Platt Perspective on Business and Technology

The challenge of strategy precluding tactics, and vice versa

Posted in reexamining the fundamentals by Timothy Platt on January 20, 2020

Strategy and tactics are often presented and discussed as if representing somehow-opposites of an either/or, starkly differentiated dichotomy. But for a variety of practical purposes and in a variety of contexts it can make more sense to see them for how they would fit together along more commonly shared continua that are scaled along organizational-structural, and action-and-response based analytical scales.

I began addressing the second of those two approaches to thinking about strategy and tactics as business planning tool sets in a recent posting, that I decided to include in this blog as a supplemental entry to an ongoing series on general theories per se, and on general theories of business in particular; see Going from Tactical to Strategic and Vice Versa, as an Emergent Process (as can be found at Reexamining the Fundamentals 2 and its Section IX.) And my goal for this posting is to continue its basic discussion, but from a distinctly different perspective than the one that I that pursued there. But that intended divergence from my earlier posting on this noted, I begin again with what are essentially the fundamentals that I built that posting from:

• The presumption of distinction and separation that can be drawn between strategy and tactics, that I just made note of at the beginning to this posting (and in my above cited earlier one), is both very real and very commonly held.
• And it shapes how tactics and strategy are understood and how they are actually carried out.

To cite an example context where those points of observation most definitely hold, consider how often managers or leaders are judged to be taking a strategic (and therefore not a tactical) approach to understanding and organizing their efforts and those of their teams, or a more tactical (and therefore not a strategic) one. As a reality check there, ask yourself what you would read into and simply assume about the two people generically cited by first name here:

• Mary is a real strategist, always focusing on and grasping the big picture.
• Bob is a tactician and plans and acts accordingly.

But for any real Mary or Bob, they probably wear both of those hats at least occasionally and certainly as circumstances would demand that they do, if they ever actually effectively wear either of them.

This, up to here, simply continues my line of discussion of my earlier above-cited strategy and tactics posting. But now let’s add one more factor to this discussion: mental bandwidth and the range and diversity of information (along with accompanying metadata about it as to its likely accuracy, timeliness, etc) that a would-be strategist or tactician can keep in their mind and make immediate use of at any one time. Consider the consequences of a single individual being able to hold at most some limited maximum number of details and facts and ideas in general, in their mind at once, with all of that actively entering into a single at least relatively coherent understanding that they can plan and act from.

Think of this as my invoking a counterpart to the social networking-reach limitations of a Dunbar’s number here, where in this case different individuals might be able to juggle more such data at once, or less so than others, but where everyone has their own maximum capacity for how much information they can have in immediate here-and-now use at any given time. Their maximum such capacity might expand or contract depending for example on whether they are rested or exhausted but they would always face some maximum capacity limitations there and at any given time. And I presume here that this limitation as faced at any one time and for any given individual, remains the same whether they are pursuing a more tactical or a more strategic planning approach.

• An alternative way to think about strategy and tactics and of differentiating between them, is to map out the assortment of data details that they would each be based upon, for where they would be sourced and for how closely related they are to each other from that.

If you consider a tactics versus strategy understanding in that context, it can be argued that good strategists hold more widely scattered details that collectively cover a wider range of experience and impact, when so planning. And their selection and filtering process in choosing which data to plan from is grounded at least in part on what amounts to an axiomatic presumption of value in drawing it from as wide an organizational context as possible so as to avoid large-scale gaps in what is being addressed. And in a corresponding manner, and according to this approach to understanding strategy and tactics, good tacticians can and do bring a more immediate and localized context into clearer focus in their understanding and planning and with a goal of avoiding the trap of unaddressed and unacknowledged gaps at that level. But the same “content limit” for their respective understandings holds in both cases.

• According to this, micromanagement occurs when strategic intent is carried out with a tactical focus and with a tactician’s actively used data set, and with this done by people who should be offering strategic level insight and guidance.
• And its equally out of focus tactical counterpart would arise when a would-be tactician is distracted by details that are of lower priority in their immediate here-and-now, even if relevant to a larger perspective understanding – where their focusing on them means crowding out information and insight of more immediate here-and-now relevance and importance to what they should be planning for.

And with that, let’s reconsider the above cited Mary and Bob: strategist and tactician respectively. According to this, Mary more naturally throws a wider net, looking for potentially key details from a fuller organizational context. And Bob more naturally focuses on their more immediate here-and-now and on what is either in front of them or most likely to arrive there and soon. And actually thinking and planning along a fuller range of the continua that more properly include both strategy and tactics as cited here and in my earlier, above-noted posting, means being able to switch between those types of data sets.

In most organizations, it is usually only a few of the people there, if any who can effectively strategically plan and lead. It all too often it is only a few who can really effectively tactically plan and lead too. And it is one of the key roles of organized processes and systems of them in a business, that they help those who would plan and lead and either strategically or tactically, to do so more effectively and with better alignment to the business plan and its overall framework in place. Good operational systems make effective tactical and strategic management and leadership easier.

It is even rarer that an individual in a business or organization be able to effectively take either a strong and capable tactical or a strong and capable strategic approach and that they be able to smoothly transition from one to the other and back as circumstances and needs dictate. And ultimately, this scarcity probably dictates the strategy versus tactics dichotomy that I write of here, more than anything else does.

• This discussion up to here, of course, leaves out the issues of how those working data sets that strategists and tacticians use, would be arrived at and updated and maintained, as an ongoing understanding of the context that any such planning would have to take place in. I leave discussion of that to another posting.

Meanwhile, you can find this and related material at my Reexamining the Fundamentals directory and its Page 2 continuation, as topics Sections VI and IX there, and with this posting specifically included as a supplemental addition to Section IX there.

Reconsidering Information Systems Infrastructure 13

Posted in business and convergent technologies, reexamining the fundamentals by Timothy Platt on January 13, 2020

This is the 13th posting to a series that I am developing, with a goal of analyzing and discussing how artificial intelligence and the emergence of artificial intelligent agents will transform the electronic and online-enabled information management systems that we have and use. See Ubiquitous Computing and Communications – everywhere all the time 2 and its Page 3 continuation, postings 374 and loosely following for Parts 1-12. And also see two benchmark postings that I initially wrote just over six years apart but that together provided much of the specific impetus for my writing this series: Assumption 6 – The fallacy of the Singularity and the Fallacy of Simple Linear Progression – finding a middle ground and a late 2017 follow-up to that posting.

I have been discussing artificial intelligence tasks and goals as divided into three loosely defined categories in this series. And I have been discussing artificial intelligence agents and their systems requirements in a goals and requirements-oriented manner that is consistent with that, since Part 9 with those categorical types partitioned out from each other as follows:

• Fully specified systems goals and their tasks (e.g. chess with its fully specified rules defining a win and a loss, etc. for it),
• Open-ended systems goals and their tasks (e.g. natural conversational ability with its lack of corresponding fully characterized performance end points or similar parameter-defined success constraints), and
• Partly specified systems goals and their tasks (as in self-driving cars where they can be programmed with the legal rules of the road, but not with a correspondingly detailed algorithmically definable understanding of how real people in their vicinity actually drive and sometimes in spite of those rules: driving according to or contrary to the traffic laws in place.)

And much if not most of that discussion has centered on the middle-ground category of partly specified goals and their tasks, and the agents that would carry them out. That gray area category that resides between tasks for tools, and tasks for arguably people, serves as a source of transition testing and of development steps that would almost certainly have to be successfully met in order to develop systems that can in fact successfully carry out true open-ended tasks and achieve their goals.

And as part of that still unfolding narrative, and in a still-partly specified context, I began discussing antagonistic networks in Part 12, citing and considering them as possible ontological development resources within single agents that would promote both faster overall task-oriented systems improvement, and more effective learning and functioning there. Consider that as one possible approach that would go beyond simple random-change testing and any improvement that might be arrived at from it (as might for example arise in a biological evolutionary context where randomness enters into the determination of precisely which genetic mutations arise that would be selected upon for their survival value fitness.)

I initially wrote Part 11 of this series with a goal of similarly considering open-ended tasks and goal in Part 12. Then I postponed that shift in focus, with a goal of starting that phase of this series here. I will do so, but before I turn this discussion in that direction, I want to at least briefly outline a second fundamentally distinct approach that would at least in principle help to reduce the uncertainties and at least apparent complexities of partly specified tasks and goals, just as effective use of antagonistic neural network subsystems would allow for ontological improvements in that direction. And this alternative is in fact the possibility that has been at least wistfully considered, more than any other in a self-driving vehicle context.

I saw a science fiction movie recently in which all cars and trucks, busses and other wheeled motorized vehicles were self-driving and with all such vehicles at least presumably, continuously communicating with and functioning in coordinated concert with all others – and particularly with other vehicles in immediate and close proximity, where driving decision mismatches might lead to immediate cause and effect problems. It was blithely stated in that movie that people could no longer drive because they could not drive well enough. But on further thought, the real problem there would not be in the limitations of any possible human driver atavists who might push their self-driving agent chauffeurs aside to take the wheel in their own hands. It is much more likely, as already touched upon in this series and in its self-driving example context, that human drivers would not be allowed on the road because the self-driving algorithms in use there were not good enough to be able to safely drive in the presence of the added uncertainty of drivers who were not part of and connected into their information sharing system, who would not always follow their decision making processes.

• Artificial intelligence systems that only face less challenging circumstances and contexts in carrying out their tasks, do not need the nuanced complexity of data analytical capability and decision making complexity that they would need if they were required to function more in the wild.

For a very real-world, working example of this principle and how it is addressed in our already every day lives, consider how we speak to our currently available generation of online verbally communicative assistants such as Alexa and Siri. When artificial intelligence systems and their agents do not already have context and performance needs simplifications built into them by default, we tend to add them in ourselves in order to help make them work, and at least effectively enough to meet our needs.

So I approach the possibility of more open-ended systems and their tasks and goals with four puzzle pieces to at least consider:

• Ontological development that is both driven by and shaped by the mutually self-teaching and learning behavior of antagonistically positioned subsystems, and similar/alternative paradigmatic approaches,
• Scope and range of new data input that might come from the environment in general but that might also come from other intelligent agents (which might mean simple tool agents that carry out single fully specified tasks, gray area agents that carry out partly specified tasks, or genuinely artificial general intelligence agents: artificial or human, or some combination of all of these source options.)
• How people who would work with and make use of these systems, simplify or add complexity to the contexts that those agents would have to perform in, shifting tasks and goals actually required of them either more towards the simple and fully specified, or more towards the complex and open-ended.
• And I add here, the issues of how an open ended task to be worked upon and goal to be achieved for it, would be initially outlined and presented. Think in terms of the rules of the road antagonist in my two subsystem self-driving example of Part 12 here, where a great deal of any success that might be achieved in addressing any overtly open-ended systems goal will almost certainly depend on where a self-learning agent would begin addressing it from.

To be clear in both how I am framing this discussion of open-ended tasks, and of the agents that would carry them out, my goal here is to begin a discussion of basic parametric issues that would constrain and shape them in general. So my goal here is to address general intelligence at a much more basic level than that of consideration of what specific types of resources should be brought to bear there – which would almost certainly prove to be inadequate as any specific artificial general intelligence agents are actually realized.

I have, as such, just cited antagonistic neural networks and agents constructed from them, that can self-evolve and ontologically develop from that type of start, as one possible approach. But I am not at least starting out with a focus on issues or questions such as:

• What specifically would be included and ontologically developed as components in a suite of adversarial neural networks, in an artificial general intelligence agent (… if that approach is even ultimately used there)?
• And what type of self-learning neural network would take overall command and control authority in reconciling and coordinating all of the activity arising from and within such a system (… here, assuming that neural networks as we currently understand them are to be used)?

I would argue that you can in fact successfully start at that solutions-detail level of conceptual planning when building artificial specialized intelligence agents that can only address single fully specified systems goals and their tasks – when you are designing and building tools per se. That approach is, in fact, required there. But it begins to break down and quickly, when you start planning and developing for anything that would significantly fall into a gray area, partly-specified category as a task or goal to be achieved or for agents that could carry them out. And it is likely going to prove to be a hallmark identifier of genuinely open-ended systems goals and their tasks, and of their agents too, that starting with a “with-what” focus cannot work at all for them. (I will discuss pre-adaptation – also called exaptation as a source of at least apparent exceptions to this, later in this series but for now let’s consider the points made here in terms of fully thought through, end-goals oriented pre-planning, and tightly task-focused pre-planning there in particular.)

I am going to continue this discussion in a next series installment where I will at least begin to more fully examine the four puzzle pieces that I made note of here. Meanwhile, you can find this and related postings and series at Ubiquitous Computing and Communications – everywhere all the time and its Page 2 and Page 3 continuations. And you can also find a link to this posting, appended to the end of Section I of Reexamining the Fundamentals as a supplemental entry there.

Moore’s law, software design lock-in, and the constraints faced when evolving artificial intelligence 10

Posted in business and convergent technologies, reexamining the fundamentals by Timothy Platt on January 4, 2020

This is my 10th posting to a short series on the growth potential and constraints inherent in innovation, as realized as a practical matter (see Reexamining the Fundamentals 2, Section VIII for Parts 1-9.) And this is also my seventh posting to this series, to explicitly discuss emerging and still forming artificial intelligence technologies as they are and will be impacted upon by software lock-in and its imperatives, and by shared but more arbitrarily determined constraints such as Moore’s law (see Parts 4-9.)

I have been focusing in this series on the hardware that would serve as a platform for an artificial intelligence agent, since Part 4 of this series, with a goal of outlining at least some of the key constraining parameters that any such software implementation would have to be able to perform within. And as a key organizing assumption there, I have predicated this entire ongoing line of discussion in terms of integrated circuit technology (leaving out the possibilities of or the issues of quantum computing in the process.) And after at least briefly discussing a succession of such constraints and for both their artificial and natural brain counterpart systems, with comparisons drawn between them, I said at the end of Part 9 that I would explicitly turn to consider lock-in and Moore’s law in that context and as they apply to the issues raised in this series. I will pursue that line of discussion here, still holding to my initial integrated circuit assumption with its bits and bytes-formatted information flow (as opposed to quantum bit, or qubit formatted data.) And I do so by addressing the issues and challenges of Moore’s law from what some might consider a somewhat unexpected direction.

• Moore’s law is usually thought of as representing an ongoing base 2 logarithmic expansion of the circuit density and corresponding hardware capability in virtually all of our electronic devices, and without any significant accompanying, matching cost increases. This ongoing doubling has led to all but miraculous increase in the capabilities of the information processing systems that we have all seemingly come to use and to rely upon and throughout our daily lives. And in that, Moore’s law represents a logarithmic growth rate expansion of capability and of opportunity and a societally enriching positive.
• But just as importantly, and certainly from the perspective of the manufacturers of those integrated circuit chips, Moore’s law has become an imperative to find ways to develop, manufacture and bring to market, next step chips with next step doubled capability and on schedule and without significant per-chip cost increases, and to keep doing so,
• Even as this has meant finding progressively more sophisticated and expensive-to-implement work-arounds, in order to squeeze as much increased circuit density out of what would otherwise most probably already be considered essentially mature industrial manufacturing capabilities, in the face of fundamental physical law constraints and again and again and again ….
• Expressed this way, Moore’s law and lock-in begin to sound more and more compatible with each other and even more and more fundamentally interconnected. The pressures inherent to Moore’s law compel quick decisions and solutions and that adds extra pressures limiting anything in the way of disruptively new technology approaches, except insofar as they might be separately developed and verified, independently from this flow of development, and over several of its next step cycles. The genuinely disruptively new and novel takes additional time to develop and bring to marketable form and for the added uncertainties that it brings with it if nothing else. The already known and validated, and prepared for are easier, less expensive and less risky to build into those next development and manufacturing cycles, where they have already been so deployed.
• But the chip manufacturer-perceived and marketplace-demanded requirement of reaching that next Moore’s law step in chip improvement and every time and on schedule, compels a correspondingly rapid development and even commoditization of next-step disruptively new innovations anyway. As noted in earlier installments to this series, continued adherence to the demands of Moore’s law has already brought chip design, development and manufacturing to a point where quantum mechanical effects and the positions and behavior of individual atoms, and even of individual electrons in current flows, into chip success-defining importance.
• And all of this means decisions being made, and design and development steps being taken that rapidly and even immediately become fundamentally locked in as they are built upon in an essentially immediately started next-round of next-generation chip development. Novel and new have to be added into this flow of change in order to keep it going, but they of necessity have to be added into an ever-expanding and ever more elaborate locked-in chip design and development framework, and with all of the assumed but unconsidered details and consequences that that entails.

What I am writing of here amounts to an in-principle impasse. And ultimately, and both for computer science and its applications, and for artificial intelligence as a special categorical case there, this is an impasse that can only be resolved: that can only be worked around, by the emergence of disruptively new and novel that moves beyond semiconductor physics and technology, and the types of electronic circuitry that are grounded in it.

Integrated circuit technology as is currently available, and the basic semiconductor technology that underlies it have proven themselves to be quite sufficient for developing artificial specialized intelligence agents that can best be considered “smart tools,” and even at least low-end gray area agents that at the very least might arguably be developed in ways that could lead them beyond non-sentient, non-sapient tool status. (See my series: Reconsidering Information Systems Infrastructure, as can be found at Reexamining the Fundamentals, as its Section I, for a more complete discussion of artificial special and general intelligence agents, and gray area agents that would be found in the capabilities gap between them.) But advancing artificial intelligence beyond that gray area point might very well call for the development of stable, larger scale quantum computing systems, and certainly if the intended goal being worked toward is to achieve true artificial general intelligence and agents that can claim to have achieved that – just as it took the development of electronic computers, and integrated circuit-based ones at that, to realize the dreams inherent in Charles Babbage’s plans for his gear-driven, mechanical computers in creating general purpose analytical engines: general purpose computers per se.

I am going to continue this line of discussion in a next series installment where I will consider artificial intelligence software and hardware, from the perspective of breaking away from lock-ins that are for the most part automatically assumed as built-in requirements in our current technologies, but that serve as innovation barriers as well as speed of development enablers.

Meanwhile, you can find this and related material at Ubiquitous Computing and Communications – everywhere all the time 3 and also see Page 1 and Page 2 of that directory. And I also include this in my Reexamining the Fundamentals 2 directory as topics Section VIII. And also see its Page 1.

Some thoughts concerning a general theory of business 32: a second round discussion of general theories as such, 7

Posted in blogs and marketing, reexamining the fundamentals by Timothy Platt on January 1, 2020

This is my 32nd installment to a series on general theories of business, and on what general theory means as a matter of underlying principle and in this specific context (see Reexamining the Fundamentals directory, Section VI for Parts 1-25 and its Page 2 continuation, Section IX for Parts 26-31.)

I have been discussing two categorically distinctive, basic forms of axiomatic theory since Part 28 that I will continue to delve into here too:

• Axiomatically closed bodies of theory: entirely abstract, self-contained bodies of theory that are grounded entirely upon sets of a priori presumed and selected axioms. These theories are entirely encompassed by sets of presumed fundamental truths: sets of axiomatic assumptions, as combined with complex assemblies of theorems and related consequential statements (lemmas, etc) that can be derived from them, as based upon their own collective internal logic. And the axioms included in such overall theories might be selected because they are at least initially deemed to be self-evident in some manner. Or they might be selected in a more arbitrary manner as far as that is concerned, and with a primary selection-determining criteria being that they at least appear not to be in conflict with each other.
• And axiomatically open bodies of theory: theory specifying systems that are axiomatically grounded as above, with at least some a priori assumptions built into them, but that are also at least as significantly grounded in outside-sourced information too, such as empirically measured findings as would be brought in as observational or experimental data. Their axiomatic underpinnings are more likely to be selected on the basis of their appearing to be self-evident truths.

And with those basic definitions repeated for smoother continuity of narrative, I begin the main line of discussion of this posting by repeating the to-address list of next topic points as offered at the end of Part 31, as an anticipated précis of what I would discuss here:

• Provide a brief discussion of the generality of axioms, and how a theory elaboration can lead to reconsideration of the axioms underlying it, where that can lead to accepted axioms already in place, now being reconsidered as special cases of more general principles.
• Adding, subtracting and modifying axioms in general, and
• The whole issue of consistency, and even just within constrained sets of axioms. And I will frame that discussion at least in part, in terms of outside data insofar as that makes these theories open – and with a reconsideration of what evolutionary and disruptive mean there.

As a final point of orienting detail that I would repeat from Part 31, I appended the following point of observation to the three basic topics points of that to-address list:

• Axiomatic assumptions should always be considered for what they limit to and exclude, as well as for what they more actively include and in any axiomatic body of theory.

And with that offered, I turn to the first of three topics points under consideration here, and issues of generality and specificity in the underlying assumptions that are selected for an organized body of theory, and that are presumed to hold so much system-wide significance there as to qualify for axiom status. I begin doing so by proposing a metatheory axiom (an axiom that would inform a general theory of general theories) that I would argue offers a reasonable starting point for essentially any axiomatically grounded body of theory:

• An assumed axiomatic truth need not always be true and under absolutely any and all possible circumstances; it need not be presumed to be universally valid and without exceptions and in all contexts. It only has to present itself as being widely enough valid, for it to be presumed to hold universally true within the confines of a specific body of theory that it is assumed to be axiomatic to.
• Turning that point of detail around in the context of this series, it means that the range of applicability of such a body of theory as a whole, should be wide enough so that it would make sense to think of it as a general theory per se. This means that it should be wide enough reaching for its descriptive and predictive applicability and impact so that its axioms would not just present themselves as special case rules of thumb for understanding some specific type or category of phenomena.
• Think of that as an admittedly loose definition that represents a line of demarcation distinguishing between compendium theories and their assemblages of narrowly applicable special case components, and more widely stated general theories per se (see this series for its Parts 2-8 for a more detailed discussion of compendium theories as a categorical type.)

To take those points of detail out of the abstract, consider the concept of parallelism as it is addressed in elliptical and hyperbolic geometries. From a closed body of theory perspective, it is quite possible to develop an entire, extensive body of geometric theory around either of those approaches to that concept, so their range of valid applicability is sufficiently large so that they can offer significant value – even as they flatly contradict each other, proving that neither approach to parallelism can be held to be absolutely universally true in any sense. They can, however, still serve as reasonable bases for developing their own respective bodies of theory. And given the history of parallelism and how it has been variously considered, that type of axiom-level postulate has in fact served as the driving impetus for essentially all non-Euclidean geometries at least as their origins are traced back to their Euclidean “ur-geometry” origin.

And from an open body of theory perspective, I would cite the general theory of relativity and its use of non-Euclidean geometries as well as empirically grounded axioms, with experimental and observational findings serving to open that body of theory up beyond the constraints of its more strictly axiomatic underpinnings. That body of theory of necessity contains within it the axiomatic underpinnings of the mathematics that it is developed in terms of: that body of mathematics’ interpretation of parallelism included.

The first topics point, as I am addressing it, includes within it “… and how a theory elaboration can lead to reconsideration of the axioms underlying it.” And to continue with my use of the physical sciences as I take this line of discussion out of the abstract, I coordinately cite classical Newtonian physics and Einstein’s theories of relativity, and his general theory of relativity in particular. Newtonian physics is predicated upon an assumption of Euclidean geometry and its axiomatic foundations with that mathematical theory as a key aspect of its overall axiomatic framework. And that means it presumes as a given fundamental truth, Euclid’s fifth postulate: his fifth axiom as it defines and specifies parallelism. That assumption does not and cannot realistically apply to the wider range of physical phenomena that Einstein sought to address in his theories, but Newtonian calculations still hold true as meaningfully valid, empirically useful approximations for objects that are large relative to the size of individual atoms and that are observed to be moving slowly relative to the speed of light. So as a matter of practicality, Newtonian assumptions can be considered as if special cases in an Einsteinian theoretical framework – where they appear as absolute truth, axiomatic assumptions in a strictly Newtonian setting.

I am going to continue this line of discussion in a next installment to this series, where I will at least begin to address the second of three topic points as offered towards the start of this posting:

• Adding, subtracting and modifying axioms in general.

Then after addressing that I will turn to and consider the third of those points:

• Consistency, and even just within constrained sets of axioms. And I will frame that discussion at least in part, in terms of outside data insofar as that makes these theories open – and with a reconsideration of what evolutionary and disruptive mean there.

Meanwhile, you can find this and related material about what I am attempting to do here at About this Blog and at Blogs and Marketing. And I include this series in my Reexamining the Fundamentals directory and its Page 2 continuation, as topics Sections VI and IX there.

Going from tactical to strategic and vice versa, as an emergent process

Posted in reexamining the fundamentals by Timothy Platt on December 27, 2019

I want to start this posting by acknowledging that I have in effect, been helping to perpetuate a myth here. And it is one that is fairly basic: fairly fundamental to business-oriented thought and planning, and to organizational and goals-directed thought and planning in general, too. The myth is a presumed clear-cut line of separation dividing tactics from strategy. And yes, I have repeatedly written of the two here as if that simple and even simplistic dichotomy were fundamentally valid.

As explanation, if not justification of my following that approach, I note that there is in fact an at least relatively clear and meaningful line of distinction separating operations from strategy, with operations including the execution side of a business and strategy comprising its overall planning and prioritizing that this operational effort would enact.

I use the terms strategy in its various grammatically shaped forms (e.g. strategic, etc), and operations a great deal in this blog, and tactics and tactical a lot less. I begin the main line of discussion of this posting by briefly outlining something of why I do that.

• Operations as a whole, can be seen as comprising actual business process-based execution itself, and immediately here-and-now, task-specific planning that would serve to bring that execution into alignment with the overall strategy in place. Tactical planning, in that context, and overlooking the possible need for tactics to address the strategically unexpected and unconsidered, can be thought of as a process of operationally translating strategy into here-and-now operational terms.
• Tactics are narrowly framed and stated; overall strategy is broadly framed and stated; tactics and tactical planning in this sense, translate the goals and intentions of that overall vision into the here-an-now specifics of the work immediately at hand.

That noted, I tend to subtend tactics and execution into operations per se, leaving out both terms as implicitly included in the overall operations and operational rubric. But that only addresses tactics per se, and I add strategy, from one possible perspective.

The relationship between tactics and strategy are a lot more complicated than any simple presumed dichotomy as suggested there, might cover. So my goal here is to at least briefly make note of a couple of areas where the two overlap and effectively blend into each other:

• The first is that of special case versus general rule
• And the second is that of what might best be thought of as learning curve shifts.

Both presume that tactics and strategy exist together on a single larger, overarching continuum and that they and the differences that do exist between them arise as consequences of emergent processes.

Let’s start with the first of those perspectives: special case and general context distinctions, and by reconsidering a point just touched upon here. Strategy and strategic planning offer broad brushstroke perspectives and understandings of a business or organization, as framed in both reactive-acknowledging, and proactive-anticipating terms. And emphasis is usually placed on the proactive side to that and on planning for the future and even when that begins with an acknowledgment of a need to course correct reactively from a problematical position. Tactics deals with the here-and-now and with emphasis on both place and timing as expressed there. And it is also primarily focused on the proactive, at least where possible.

Most of the time, I add, tactical thinking and planning deal with recurring events and circumstances, and not so much with unique never-to-be-repeated contingencies. And strategy does too, at least when business is proceeding normatively and as business-as-usual.

And with that set of observations offered, in practice, tactics deals with specialized, fine-detail events and circumstances, and strategy deals with the generalized and in general terms there. And the here-localized of tactics and the overall-generalized of strategy fit on a continuum scale that includes within it, areas of overlap and of potential overlap for range and scope of impact and of immediate and long-term concern, where either or both of tactical and strategic approaches could quite arguably be followed.

I divide the issues that I raise and address here, very differently with the second of two approaches to thinking about tactics and strategy that I offer here: learning curve shifts. In this case, tactics and tactical approaches and reasoning can serve as proving grounds and as validation gauntlets that strategy and putative strategy might be tested and validated against.

• As a change management aside here, of significant relevance to this posting’s discussion, you are facing a real warning sign when tactical approaches that are empirically arrived at because they work, do not actually mesh all that closely with the overall strategic vision and understanding in place that they are supposed to reflect and implement. And when the two: tactics and strategy in effect contradict each other, that strongly suggests that that business’ overall strategic planning has to be rethought and course corrected in order to bring it back to relevancy again.

I cite that as an extreme but nevertheless very real world example of how the second of those tactics and strategy understandings might be applied. That type of circumstance arises when no real learning curve efforts are made or achieved and when no real learning is encouraged or even allowed and certainly when considering the business as a whole, or how it is run.

Learning curve opportunities as created and shared throughout an organization, and learning per se in this context mean reality testing tactics and tactical understandings against their big picture strategic counterparts and in both directions: starting with tactics to reconsider strategy and vice versa. And a once-tactical approach can become a core strategic approach and understanding, and a once more-generally conceived strategic understanding or assumption might be shifted to a more special-circumstance only status. Or it might be discarded entirely.

Think of this, if you will, as a supplemental posting to my ongoing series: Some Thoughts Concerning a General Theory of Business, as can be found at Reexamining the Fundamentals and its Page 2 as Sections VI and IX. Though I also add this here as a separate stand-alone thought piece in and of itself too.

I will in fact continue, for the most part, to write of operations as a here-admittedly more complex whole, and I will continue to write here of strategy (which is also a complex whole) and in more generally inclusive terms too, except where circumstances and the imperatives of the topic at hand compel otherwise. But I offer this thought piece here to at least briefly shed some light into the black boxes that are operations and strategy, as those more generally stated wholes can be more widely considered.

Meanwhile, you can find this and related material at my Reexamining the Fundamentals directory and its Page 2 continuation, as topics Sections VI and IX there, and with this posting specifically included as a supplemental addition to Section IX there.

Reconsidering Information Systems Infrastructure 12

Posted in business and convergent technologies, reexamining the fundamentals by Timothy Platt on November 5, 2019

This is the 12th posting to a series that I am developing, with a goal of analyzing and discussing how artificial intelligence and the emergence of artificial intelligent agents will transform the electronic and online-enabled information management systems that we have and use. See Ubiquitous Computing and Communications – everywhere all the time 2 and its Page 3 continuation, postings 374 and loosely following for Parts 1-11. And also see two benchmark postings that I initially wrote just over six years apart but that together provided much of the specific impetus for my writing this series: Assumption 6 – The fallacy of the Singularity and the Fallacy of Simple Linear Progression – finding a middle ground and a late 2017 follow-up to that posting.

I have conceptually divided artificial intelligence tasks and goals into three loosely defined categories in this series, and have been discussing artificial intelligence agents and their systems requirements in a goals and requirements-oriented manner that is consistent with that, since Part 9 with those categorical types partitioned out from each other as follows:

• Fully specified systems goals and their tasks (e.g. chess with its fully specified rules defining a win and a loss, etc. for it),
• Open-ended systems goals and their tasks (e.g. natural conversational ability with its lack of corresponding fully characterized performance end points or similar parameter-defined success constraints), and
• Partly specified systems goals and their tasks (as in self-driving cars where they can be programmed with the legal rules of the road, but not with a correspondingly detailed algorithmically definable understanding of how real people in their vicinity actually drive and sometimes in spite of those rules: driving according to or contrary to the traffic laws in place.)

And I focused on the third of those artificial intelligence agent systems goals in Part 11, citing them as a source of transition step goals that if successfully met, would help lead to the development of artificial general intelligence agents, capable of mastering and carrying out open-ended tasks (such as natural open-ended conversation as cited above.)

I then said that I would turn here to at least begin to more directly discuss open-ended systems goals and their tasks, and artificial agents that would be capable of them, here. And I will in fact do so. But as a starting point, I am going to further address that gray area category of partly specified systems goals and their tasks again, and my self-driving car example in particular, to put what is to follow into a clearer and perhaps less-abstract perspective. And to be more specific here, I am going to at least briefly outline one possible approach for more effectively dealing with and resolving the uncertainties posed by real-world drivers who do not always follow the rules of the road and the law, as that would inform and in fact fundamentally shape a self-driving car’s basic decision and action algorithm: its basic legally constrained driving behavior algorithm that if sufficient, would lead self-driving to qualify as a fully specified systems goal and task.

Complex single algorithms and the neural networks or other physical, hardware systems that contain and execute them, can at least in principle stand alone and offer, at least for those artificial agents, complete solutions to whatever problem or task is at hand for them. But these resources can also be layered and set up to interact, and with the results of those interactions determining finalized step-by-step decisions actually made and carried out by those agents as a whole. One way to look at those systems is to think of them as overall composite neural networks (to pursue that hardware approach here), where the constituent sub-networks involved each have their own information processing and decision making algorithm that would have to be reconciled as specific actions are actually taken, and with results-feedback going to those individual sub-networks as input for their ongoing self-learning improvement and optimization efforts. But to simplify this discussion, while hopefully still addressing at least something of the complexities there, let’s address this here as a matter of individual artificial intelligence agents having what amount to competing and even adversarial neural networks within them – with those arguably separate but closely interacting networks and their algorithms tasked, long-term with a goal of improving each other’s capabilities and performance from the ongoing challenge that they provide each other.

This approach is most certainly not original on my part. It traces back to work first published in 2014 by Ian Goodfellow when he was working at Google (see this piece on Generative Adversarial Networks). And it is an emerging approach for developing self-learning autonomous systems. I see these types of interacting neural networks as holding at least as much value in execution and ongoing functionality too, and particularly for partly specified systems and their task-defined goals. And I see this approach as holding real value for possible open-ended systems and their task goals: true artificial general intelligence agents definitely included there.

Consider in that regard, two adversarially positioned neural networks with one seeking through its algorithms and its expert knowledge database contained background information, to solve a complex problem or challenge. And the other, with its algorithms and expert knowledge data, is set up to challenge the first for the quality, accuracy and/or efficiency of what it arrives at as its for-the-moment best solution to its task at immediate hand. What types of tasks would be more amenable to this approach? Look for differences in what has to be included functionally, in an effective algorithm for each of these two sides, and look for how their mutually correcting feedback flows could lead to productive synergies. In this, both of these networks and their algorithms recurringly challenge their adversary – each other, and with insights arrived at from this added to both of their respective expert knowledge databases.

Let’s apply this to the self-driving car challenge that I have been selectively offering here, with the overall task of safely and efficiently autonomously self-driving, divided into two here-adversarial halves as follows:

• One of these included neural networks would focus on the rules of the road and on all of the algorithmic structure, and expert knowledge that would go into self-driving in a pure context where all vehicles were driven according to legally defined standard, rules of the road, and all were driven with attentive care. That is, at least how this neural network would start out, so call it the rules-based network here.
• And the other would start out with essentially the same starter algorithm, but with expert knowledge that is centered on how real people drive, and with statistical likelihood of occurrence and other relevant starter data provided, based on insurance company actuarial findings as to the causes of real-world auto accidents. Call this the human factor network.
• Every time a human “copilot,” sitting behind the wheel finds it necessary to take manual control of such a self-driving vehicle, and every time they do not, but sensor input from the vehicle matches a risk pattern that reaches at least some set minimum level of risk significance, the human factor network would notify the rules-based one, and that network would seek out a best “within the rules” solution for resolving that road condition scenario. It would put its best arrived at solution as well as its rules-based description of this event into its expert systems database and confirm that it has done so to its human factor counterpart.
• And this inter-network training would flow both ways with the rules-based network updating its human factor counterpart with its new expert knowledge that came from this event, updating the starting point driving assumptions that it would begin from too. That network would have, of course, updated its actual human driving behavior expert knowledge data too, to include this now-wider and more extensive experience base that it would base its decisions upon.
• And this flow of interactions would drive the ongoing ontological development of both of those neural networks and their embedded algorithms, as well as adding structured, analyzed data to their expert systems databases, advancing both of them and the system that they enter into as a whole.

None of this is particularly groundbreaking, and certainly as a matter of underlying principles, at least in the context of the already rapidly developing artificial intelligence knowledge and experience already in place and in use. But I sketch it out here, as one of many possible self-learning approaches that is based on the synergies that can arise in adversarial self-learning systems.

I am going to continue this discussion in a next series installment, where I will turn to consider how this basic approach might apply in a more general intelligence context, and certainly when addressing more open-ended systems goals and their tasks.

Meanwhile, you can find this and related postings and series at Ubiquitous Computing and Communications – everywhere all the time and its Page 2 and Page 3 continuations. And you can also find a link to this posting, appended to the end of Section I of Reexamining the Fundamentals as a supplemental entry there.

Moore’s law, software design lock-in, and the constraints faced when evolving artificial intelligence 9

Posted in business and convergent technologies, reexamining the fundamentals by Timothy Platt on October 27, 2019

This is my 9th posting to a short series on the growth potential and constraints inherent in innovation, as realized as a practical matter (see Reexamining the Fundamentals 2, Section VIII for Parts 1-8.) And this is also my sixth posting to this series, to explicitly discuss emerging and still forming artificial intelligence technologies as they are and will be impacted upon by software lock-in and its imperatives, and by shared but more arbitrarily determined constraints such as Moore’s law (see Parts 4-8.)

My goal for this installment is to at least begin to tie together a series of issues and discussion threads that I have been touching upon up to now, in an explicitly natural brain versus artificial intelligence agent context. And I begin doing so here, by noting that all of the lines of discussion that I have pursued up to here, relevant to that set of issues, have dealt with fundamental sources of constraints and of what can be physically possible, and how best to develop and build within those limitations:

• Thinking through and understanding basic constraining parameters that would offer outer limits as to what can be carried out as a maximum possible performance benchmark by these systems, and
• Optimizing the hardware and software systems, or their biological equivalents in place in them: artifactual or natural, so as to more fully push the boundaries of those constraints.

So I have, for example, discussed the impact of finite maximum signal speeds in these systems, as that would set maximum outer limits as to the physical size of these information processing systems, and certainly as maximum time allowances are set for how quickly any given benchmark performance tests under consideration, might have to be completed in them (see Part 7.) And on the performance optimization side of this, see my discussion of single processor versus parallel processor array systems, and of information processing problems that would be more amenable to one or the other of those two basic design approaches (see Part 8.)

I stated at the end of Part 8 that I would raise and at least briefly discuss one more type of constraining parameter here, as it would limit how large, complex and fast brains or their artificial intelligence counterparts can become, where size scale and speed can and do, and at least in their extremes here, come to directly compete against each other. And I also said that I would discuss the issues raised in this posting and immediately preceding installments to it, in terms of Moore’s law and its equivalents and in terms of lock-in and its limiting restrictions. I will, in fact begin addressing all of that by citing and discuss two new constraints-defining issues here, that I would argue are inseparably connected from each other:

• The first of them is a constraining consideration that I in fact identified at the end of Part 8, and both for what it is and for how I would address it here: the minimum size constraint for how small and compact a functional circuit element can be made, or in a biological system, the smallest and most compactly it can be evolved and grown into. This obviously directly addresses the issues of Moore’s law where it is an essential requirement that circuit elements be made progressively smaller and smaller, if the number of them that can be placed in a same-size integrated circuit chip is to continue to increase and at the rate that Moore initially predicted with its ongoing same-time period doublings.
• And the second of these constraints, that I add in here in this narrative, involves the energy requirements of these systems as the number of functional elements increases in accordance with Moore’s law. And that of necessity includes both energy required to power and run these systems, and the need to deal with ever-increasing amounts of waste energy byproduct: heat generation as all of this energy is concentrated into a small volume. Think of the two sides of this power dynamic as representing the overall energy flow cost of Moore’s law and its implementation.
• And think of these two constraints as representing the two sides of a larger, overarching dynamic that together serve to fundamentally shape what is and is not possible here.

Let’s at least begin to address the first of those two individual sources of constraint, as raised in the first of those bullet points, by considering the number of functional elements and connections between them that can be found in an at least currently high-end central processing unit chip, and in an average normal adult human brain:

• The standard estimate that is usually cited for a total number of neurons in at average, normal adult human brain is generally stated as 100 billion, give or take a billion or so. And the average number of synaptic connections in such a brain is often cited as being on the order of 100 trillion (with an average of some 1000 synapses per neuron and with some having as many as 10,000 and more per cell, depending on neuronal type.)
• The number of transistors in a single integrated circuit chip, and in central processing unit chips as a special case in point of particular relevance here, keeps growing as per Moore’s law as it is still holding true (see this piece on transistor counts.) And according to that Wikipedia, entry, at least as of this writing, the largest transistor count in a commercially available single-chip microprocessor (as first made commercially available in 2017) is 19.2 billion. But this count is dwarfed by the number of transistors in the largest and most advanced memory chips. As per that same online encyclopedia entry, the current record holder there (as of 2019) is the Samsung eUFS (1 TB) 3D-stacked V-NAND flash memory chip (consisting of 16 stacked V-NAND dies), with 2 trillion transistors (capable of storing 4 bits, or one half of a byte per transistor).
• Individual neurons are probably about as small and compact as they can be, and particularly given the way they interconnect to work, through long dendritic (input) and axonal (output) processes. And the overall volume of that normative adult human brain has approximately one support cell (and glial cells in particular for that) for every neuron. And at least some types of neuron as found in a normal brain are, and most probably have to be particularly large in order to function and particularly given the complexity of their interconnections with other cells (e.g. mirror neurons.)
• Continued pursuit of Moore’s law-specified chip development goals has brought integrated circuit, functional elements and the connectors between them down in size to a scale where quantum mechanical effects have come to predominate in describing much of their behavior. And the behavior and the positioning of individual atoms in them have become of fundamental importance and both for their design and fabrication and for their performance and their stability.

And this brings me directly to the second constraining factor that I would address here: energy flows, and both as electrical power would be used to run a given integrated circuit chip and as an unavoidable fraction of it would be dissipated as heat after use. And I begin addressing that, in an artificial intelligence agent context, by explicitly pointing out a detail just offered in one of my above-noted examples here: the Samsung eUFS random access memory chip, in its1 TB version, is constructed as a 3-D chip with functionally complete sets of layers built upon each other in stacked arrays. This, in fact is a rarity in the history of integrated circuit chip design, where energy requirements for powering these chips, translates into heat generation. One of the ongoing goals in the design and construction of integrated circuit-based systems, has been heat dissipation, and from the beginning of the industry. And 3-dimensional layering as pursued to the degree that it is in this chip, both increases circuit element density possible in it, keeping everything in closer physical proximity, but at the expense of concentrating the heat generated from running it.

This proximity is important as a possible mechanism for increasing overall chip speed and computational efficiency, given the finite speed of light that it has to function within the limits of, and the maximum possible speed that electrons can move at in those circuits – always less than that speed. To explain that with a quick numerical example, let’s assume that the speed of light in a vacuum is precisely 300 million meters per second (where one meter equals approximately 39 inches.) And let’s assume that a central processing chip, to switch example types, can carry out 1 billion floating point format, mathematical operations in a second. At that clock speed, light would travel 0.3 meters, or 11.7 inches in a straight line. But in the real world, operations are carried out on input, and on input that has been moved into at least short-term volatile memory cache and accessing that takes time, as does sending newly computed output to memory – that might be stored in buffer on the same chip but that might be stored on a separate chip too. And if data is called for that has to be retrieved from longer-term memory storage, that really slows all of this down.

In real chips and in real computer systems built around them, that maximum possible distance traveled, and by photons, not slower electrons, can and does evaporate. So stacking as per a Samsung eUFS chip makes sense. The only problem is that if this is done more than just nominally and under very controlled contexts, it becomes too likely that the chips involved will become so hot that they would cease to function; they could even literally melt.

Biological systems at least potentially face comparable, if less dramatically extreme challenges too. A normal adult brain consumes some 20% of all of the calories consumed by a human body as a whole, as a percentage of a standard resting metabolic rate. And that same brain consumes some 20% of all oxygen used too. And this is with a brain that has a total weight that accounts for only some 2% of the overall body weight. An expanded brain that now accounted for 10% of the overall body mass would, if this were to scale linearly, require 100% of the nutrients and oxygen that is now normally taken in and used by the entire body, requiring expanded capabilities for consuming, distributing and using both and probably throughout that body.

The basic constraints that I write of here, have profound impact, and on both biological and artificial information processing systems. And together, the fuller sets of these constraints actually faced, have a tremendous impact on what is even possible, and on how even just approaching that set of limitations might be made possible. And this is where I explicitly turn back to reconsider Moore’s law as a performance improvement driver, and technology development lock-in as a brake on that, and how the two in fact shape each other in this context. And I will at least begin that phase of this overall discussion in the next installment to this series.

Meanwhile, you can find this and related material at Ubiquitous Computing and Communications – everywhere all the time 3 and also see Page 1 and Page 2 of that directory. And I also include this in my Reexamining the Fundamentals 2 directory as topics Section VIII. And also see its Page 1.

Some thoughts concerning a general theory of business 31: a second round discussion of general theories as such, 6

Posted in blogs and marketing, reexamining the fundamentals by Timothy Platt on October 24, 2019

This is my 31st installment to a series on general theories of business, and on what general theory means as a matter of underlying principle and in this specific context (see Reexamining the Fundamentals directory, Section VI for Parts 1-25 and its Page 2 continuation, Section IX for Parts 26-30.)

I have been discussing general theories per se in this series, as well as the more specifically focused general theories of business that I hold out as the primary topic here. And as a part of that I have been discussing axiomatic theories per se in this series, since Part 26. And more specifically, I have been discussing two categorically distinctive, basic forms of axiomatic theory since Part 28 that I will continue to delve into here too:

Axiomatically closed bodies of theory: entirely abstract, self-contained bodies of theory that are grounded entirely upon sets of a priori presumed and selected axioms. These theories are entirely encompassed by sets of presumed fundamental truths: sets of axiomatic assumptions, as combined with complex assemblies of theorems and related consequential statements (lemmas, etc) that can be derived from them, as based upon their own collective internal logic. And the axioms included in such overall theories might be selected because they are at least initially deemed to be self-evident in some manner. Or they might be selected in a more arbitrary manner as far as that is concerned, and with a primary selection-determining criteria being that they at least appear not to be in conflict with each other.
• And axiomatically open bodies of theory: theory specifying systems that are axiomatically grounded as above, with at least some a priori assumptions built into them, but that are also at least as significantly grounded in outside-sourced information too, such as empirically measured findings as would be brought in as observational or experimental data. Their axiomatic underpinnings are more likely to be selected on the basis of their appearing to be self-evident truths.

I addressed the issues of both of these approaches to analytically ordered and organized knowledge and understanding in Part 30. And I then stated at the end of that posting that I would turn here to focus more fully on axiomatically open theories. And more specifically, I said that I would discuss:

• The emergence of both disruptively new types of data and of empirical observations that could generate it, as that would impact on axiomatically open theories,
• And shifts in the accuracy and resolution, or the range of observations that more accepted and known types of empirical observations might suddenly come to offer in that context.

And in the course of delving into that complex of issues, I will also continue an already ongoing discussion here of scope expansion, for the set of axioms assumed in a given theory-based system, and with a goal of more fully analytically discussing optimization for the set of axioms that would be presumed in a given general theory, and what such optimization even means.

And I begin this posting and its line of discussion by posing an axiomatic assumption in what I would presume as a valid metatheory of general theories as a whole, that I would argue to be a reasonable starting point for thinking about open and closed bodies of theory, and axiomatically open theories in particular:

• In the real world, people make observations and seek to organize them in their minds and in their understanding, for how they connect into and create larger consistent patterns.
• But in genuinely closed axiomatic theories, the a priori accepted and presumed axioms in place in them take precedence, and apparently conflicting empirical findings would at most call for the creation and elaboration of competing theoretical models and systems.
• In open axiomatic theories, empirically derived or otherwise-accepted data can and does compel evolutionary, or for more disruptively novel new data, revolutionary change in existing underlying theory – and change that can reach down to the level of the axioms in place themselves there.
• There, and as a defining point of difference, separating axiomatically closed systems of theory from open ones, outside-sourced data and insight does take precedence and even over the root axiomatic assumptions in place.

And with this noted, I turn to and begin discussing the topics list for this posting, starting with a more detailed consideration of the outside-sourced information that would be collected, interpreted and used here: the outside-sourced data that would explicitly enter into an open axiomatic system, making it open as such. And I begin addressing that by making note of a fundamental point of distinction. Data that does not simply support an already tested and validated, organized body of theory can arise from two different types of direction:

• It can arise as new types of, and even disruptively new types of outside-sourced information as new and novel types of experimental, quasi-experimental or observational tests are carried out, that could have an effect of validation testing that theory, or
• It can arise as output of already established tests and of any of those three categorical types, as they are carried out with greater levels of precision or under new and expanded ranges of application that might not have previously been possible to achieve.
• Put perhaps simplistically, think of this as making a distinction between disruptively new types of data that would be brought into an axiomatically open theory and both to expand its effective applicability and to expand its range of validation, and
• More evolutionarily developed data and its sources. (Nota bene: there is an assumption implicit in this, that I will challenge as I proceed in this line of discussion.)

To take that out of the abstract, let’s consider a source of examples from physics. Newtonian physics was first proposed, experimentally tested, and further refined and expanded as a body of general theory, and extensively so, on the basis of the ongoing study of objects that are very large when compared to individual atoms, and that travel very slowly in comparison to the speed of light. That is all that was possible at the time.

Then when it became possible to observe and measure phenomena that fell outside of those constraining limitations, and gather and analyze data that could be used to validate or disprove Newtonian assumptions and predictions in these now-wider ranges of empirical experience, systematic deviations were found between what classical Newtonian physics would predict and what was actually observed. And entirely new types of phenomena were observed that Newtonian physics on its own, did not seem capable of addressing at all. And while Newtonian physics and its descriptive and predictive modeling are still used and with great precision and accuracy where they do apply and where their calculations work, they have been supplanted by newer bodies of theory with different underlying axiomatic assumptions built into them: the special and general theories of relativity, and quantum theory, where they observationally and predictively break down.

• Newtonian physics with its mathematical and computational complexities, is quite sufficient for calculating rocket engine thrust and burn times and all of the other factors and parameters that would be needed to launch a satellite into a precisely intended orbit, or even to send a probe to a succession of other planets as for example was called for with the Voyager probes.
• The special and general theories of relativity and quantum theory come into their own in an ever-increasing range of applications that violate the basic and now known-limiting assumptions that Newton made and that classical physics came to elaborate upon.

And with this noted, I return to the parenthetical comment I just made concerning assumptions. And the example from physics that I just raised here, highlights why that assumption can be problematical. As validating examples in support of that contention, consider the Michelson–Morley experiment, as an example of what a classical physicist would consider a disruptively novel experiment and the findings that it generated, challenging Newtonian physics and providing explicit support for Einstein’s theory of special relativity. And at the same time, consider Brownian motion: first formally described in the physics literature in 1828 and finally explained as to mechanism by Albert Einstein as a body of observable findings that classical physics could not explain. The first of these two examples was disruptively novel and both for the data derived through it and for how that data was arrived at; the second was observational in nature and did not by all appearance involve or include anything new or novel at all. But predictively describing the phenomena involved in it called for new axiomatic level understandings and new fundamental theory.

New evolutionary-level change and even new revolutionary change in how data is arrived at and in what it consists of, does not always compel fundamental change in a theory, or a more general body of theory in place. Both can in fact serve to bolster and support it. But both can come to fundamentally challenge it too. But even so, it can seem self-evident that the disruptively new and novel might be more likely to serve as a source of challenge than of support and certainly for a body of general theory that is still, in effect coming together as a newly organized whole.

I begin addressing that conjecture by posing a pair of questions:

• Is it always possible to draw a sufficiently clear cut distinction between disruptively new and revolutionary, and step-by-step evolving and evolutionary in the type of context raised in my above-offered bipartite distinction of data, that would enter into axiomatically open theories?
• And if the goal of making that type of distinction here, is to categorically identify meaningfully distinctive drivers of change in axiomatically open theories, is this the right point of distinction for pursuing that?

Is it always possible to draw such degree-of-novelty distinctions in ways that would always, unequivocally hold as valid? It would make things easier to be able to identify and label as such, bodies of supporting or refuting evidence, and methodologies for arriving at that data that are disruptively new, and ones that are simply next-step evolutionary elaborations of already established data patterns and data generating methods, and simply look for the disruptively new forms of experiment and the data they lead to, in order to identify fundamental theory-level changers. But ultimately, simple step-by-step data collection and analysis cannot offer a conclusive, or even necessarily a meaningful distinction there. (As a working example there, consider Brownian motion again and the gap of understanding that prevailed for it from 1828 as noted above, and Einstein’s insight most of 80 years later in 1905.) So I in fact do more than just question my implicit assumption as noted above; I fundamentally challenge it with a goal of rethinking what disruptively new and emergent even means in this type of context.

• So how would one categorically think through axiomatically open theories here, for how outside-sourced information would impact upon them, and ideally in ways that would allow for meaningful, independently replicable determinations of novelty and newness as well as of impact? And that is where I come to the issues of scale, as touched upon by name towards the top of this posting.

I am going to at least begin to delve into that complex of issues, and also the issue of optimization, in the next installment to this series. And in anticipation of that, this will mean my considering:

• The generality of axioms, and how a theory elaboration can lead to reconsideration of the axioms underlying it, where that can lead to accepted axioms already in place, now being reconsidered as special cases of more general principles.
• Adding, subtracting and modifying axioms in general, and
• The whole issue of consistency, and even just within constrained sets of axioms.

And I will frame that discussion at least in part, in terms of the outside data that makes these theories open – and with a reconsideration of what evolutionary and disruptive mean here. And with that noted, I conclude this posting by offering a basic point of observation, and of conclusion derived from it:

• Axiomatic assumptions should always be considered for what they limit to and exclude, as well as for what they more actively include and in any axiomatic body of theory.

Meanwhile, you can find this and related material about what I am attempting to do here at About this Blog and at Blogs and Marketing. And I include this series in my Reexamining the Fundamentals directory and its Page 2 continuation, as topics Sections VI and IX there.

Reconsidering Information Systems Infrastructure 11

This is the 11th posting to a series that I am developing, with a goal of analyzing and discussing how artificial intelligence and the emergence of artificial intelligent agents will transform the electronic and online-enabled information management systems that we have and use. See Ubiquitous Computing and Communications – everywhere all the time 2 and its Page 3 continuation, postings 374 and loosely following for Parts 1-10. And also see two benchmark postings that I initially wrote just over six years apart but that together provided much of the specific impetus for my writing this series: Assumption 6 – The fallacy of the Singularity and the Fallacy of Simple Linear Progression – finding a middle ground and a late 2017 follow-up to that posting.

I conceptually divide artificial intelligence tasks and goals into three loosely defined categories in this series. And I have been discussing artificial intelligence agents and their systems requirements in a goals and requirements-oriented manner that is consistent with that, since Part 9 with those categorical types partitioned out from each other as follows:

• Fully specified systems goals and their tasks (e.g. chess with its fully specified rules defining a win and a loss, etc. for it),
• Open-ended systems goals and their tasks (e.g. natural conversational ability with its lack of corresponding fully characterized performance end points or similar parameter-defined success constraints), and
• Partly specified systems goals and their tasks (as in self-driving cars where they can be programmed with the legal rules of the road, but not with a correspondingly detailed algorithmically definable understanding of how real people in their vicinity actually drive and sometimes in spite of those rules: driving according to or contrary to the traffic laws in place.)

And I have focused up to here in this developing narrative on the first two of those task and goals categories, only noting the third of them as a transition category, where success in resolving tasks there would serve as a bridge from developing effective artificial specialized intelligence agents (that can carry out fully specified tasks and that have become increasingly well understood and both in principle and in practice) to the development of true artificial general intelligence agents (that can carry out open-ended tasks and that are still only partly understood for how they would be developed.)

And to bring this orienting starting note for this posting, up to date for what I have offered regarding that middle ground category, I add that I further partitioned that general category for its included degrees of task performance difficulty, in Part 10, according to what I identify as a swimming pool model:

• With its simpler, shallow end tasks that might arguably in fact belong in the fully specified systems goals and tasks category, as difficult entries there, and
• Deep end tasks that might arguably belong in the above-repeated open-ended systems goals and tasks category.

I chose self-driving vehicles and their artificial intelligence agent drivers as an intermediate, partly specified systems goal because it at least appears to belong in this category and with a degree of difficulty that would position it at least closer to the shallow end than the deep end there, and probably much closer.

Current self-driving cars have performed successfully (reaching their intended destinations and without accidents) and both in controlled settings and on the open road and in the presence of actual real-world drivers and their driving. And their guiding algorithms do seek to at least partly control for and account for what might be erratic circumambient driving on the part of others on the road around them, by for example allowing extra spacing between their vehicles and others ahead of them on the road. But even there, an “aggressive” human driver might suddenly squeeze into that space, and without signaling that they would change lanes, suddenly leaving a self-driving vehicle following too closely too. So this represents a task that might be encoded into a single if complex overarching algorithm, as supplemented by a priori sourced expert systems data and insight, based on real-world human driving behavior. But it is one that would also require ongoing self-learning and improvement on the part of the artificial intelligence agent drivers involved too, and both within these specific vehicles and between them as well.

• If all cars and trucks on the road were self-driving and all of them were actively sharing action and intention information with at least nearby vehicles in that system, all the time and real-time, self-driving would qualify as a fully specified systems task, and for all of the vehicles on the road. As soon as the wild card of human driving enters this narrative, that ceases to hold true. And the larger the percentage of human drivers actively on the road, the more statistically likely it becomes that one or more in the immediate vicinity of any given self-driving vehicle will drive erratically, making this a distinctly partly specified task challenge.

Let’s consider what that means in at least some detail. And I address that challenge by posing some risk management questions that this type of concurrent driving would raise, where the added risk that those drivers bring with them, move this out of a fully specified task category:

• What “non-standard” actions do real world drivers make?

This would include illegal lane changes, running red lights and stop signs, illegal turns, speeding and more. But more subtly perhaps, this would also include driving at, for example, a posted speed limit but under road conditions (e.g. in fog or during heavy rain) where that would not be safe.

• Are there circumstances where such behavior might arguably be more predictably likely to occur, and if so what are they and for what specific types of problematical driving?
• Are there times of the day, or other identifiable markers for when and where specific forms of problematical driving would be more likely?
• Are there markers that would identify problem drivers approaching, and from the front, the back or the side? Are there risk-predictive behaviors that can be identified before a possible accident, that a self-driving car and its artificial intelligence agent can look for and prepare for?
• What proactive accommodations could a self-driving car or truck make to lessen the risk of accident if, for example its sensors detect a car that is speeding and weaving erratically from lane to lane in the traffic flow, and without creating new vulnerabilities from how it would respond to that?

Consider, in that “new vulnerabilities” category, the example that I have already offered in passing above, when noting that increasing the distance between a self-driving car and a vehicle that is directly ahead of it, might in effect invite a third driver to squeeze in between them, and even if that meant it was now tailgating that leading vehicle and the self driving car that would now be behind it was tailgating it. A traffic light ahead, suddenly changing to red, or any other driving circumstance that would force the lead car in all of this to suddenly hit their brakes could cause a chain reaction accident.

What I am leading up to here in this discussion is a point that is simple to explain and justify in principle, even as it remains difficult to operationally resolve as a challenge in practice:

• With the difficulty in these less easily rules-defined challenges increasing, as the tasks that they would arise in it fit into deeper and deeper areas of that swimming pool in my above-cited analogy.

Fully specified systems goals and their tasks might be largely or even entirely deterministic in nature and rules determinable, where condition A always calls for action and response B, or at least a selection from among a specifiable set of particular such actions that would be chosen from, to meet the goals-oriented needs of the agent taking them. But partly specified systems goals and their tasks are of necessity significantly stochastic in nature, and with probabilistic evaluations of changing task context becoming more and more important as the tasks involved fit more and more into the deep end of that pool. And they become more open-endedly flexible in their response and action requirements too, no longer fitting cleanly into any given set of a priori if A then B rules.

Airplanes have had autopilot systems for years and even for human generations now, with the first of them dating back as far as 1912: more than a hundred years ago. But these systems have essentially always had human pilot back-up if nothing else, and have for the most part been limited to carrying out specific tasks, and under circumstances where the planes involved were in open air and without other aircraft coming too close. Self-driving cars have to be able to function in crowded roads and without human back-up – and even when a person is sitting behind the wheel, where it has to be assumed that they are not always going to be attentive to what the car or truck is doing, taking its self-driving capabilities for granted.

And with that noted, I add here that this is a goal that many are actively working to perfect, at least to a level of safe efficiency that matches the driving capabilities of an average safe driver on the road today. See, for example:

• The DARPA autonomous vehicle Grand Challenge, and
• Burns, L.D. and C Shulgan (2018) Autonomy: the quest to build the driverless car and how that will reshape the world. HarperCollins.

I am going to continue this discussion in a next series installment where I will turn back to reconsider open-ended goals and their agents again, and more from a perspective of general principles. Meanwhile, you can find this and related postings and series at Ubiquitous Computing and Communications – everywhere all the time and its Page 2 and Page 3 continuations. And you can also find a link to this posting, appended to the end of Section I of Reexamining the Fundamentals as a supplemental entry there.

Moore’s law, software design lock-in, and the constraints faced when evolving artificial intelligence 8

Posted in business and convergent technologies, reexamining the fundamentals by Timothy Platt on August 22, 2019

This is my 8th posting to a short series on the growth potential and constraints inherent in innovation, as realized as a practical matter (see Reexamining the Fundamentals 2, Section VIII for Parts 1-7.) And this is also my fifth posting to this series, to explicitly discuss emerging and still forming artificial intelligence technologies as they are and will be impacted upon by software lock-in and its imperatives, and by shared but more arbitrarily determined constraints such as Moore’s law (see Parts 4-7.)

I focused, for the most part in Part 7 of this series, on offering what amount to analogs to the simplified assumption Thévenin circuits of electronic circuit design. Thévenin’s theorem and the simplified and even detail-free circuits that they specify, serve to calculate and mimic the overall voltage and resistance parameters for what are construed to be entirely black-box electronic systems with their more complex circuitry, the detailed nature of which are not of importance in that type of analysis. There, the question is not one of what that circuitry specifically does or how, but rather of how it would or would not be able to function with what overall voltage and resistance requirements and specifications in larger systems.

My simplified assumption representations of Part 7 treated both brain systems and artificial intelligence agent systems as black box entities and looked at general timing and scale parameters to both determine their overall maximum possible size, and therefore their maximum overall complexity, given the fact that any and all functional elements within them would have larger than zero minimum volumes, and well as minimal time-to-task-completion requirements for what they would do. And I offered my Part 7 analyses there as first step evaluations of these issues, that of necessity would require refinement and added detail to offer anything like actionable value. Returning briefly to consider the Thévenin equivalents that I just cited above, by way of analogous comparison here, the details of the actual circuits that would be simplistically modeled there might not be important to or even germane to the end-result Thévenin circuits arrived at, but those simplest voltage and resistance matching equivalents would of necessity include within them, the cumulative voltage and resistance parameters of all of that detail in those circuit black boxes, even if as they would be rolled into overall requirement summaries for those circuits as a whole.

My goal for this posting is to at least begin to identify and discuss some of the complexity that would be rolled into my simplified assumptions models, and in a way that matches how a Thévenin theorem calculation would account for internal complexity in its modeled circuits’ overall electrical activity and requirements calculations, but without specifying their precise details either. And I begin by considering the functional and structural nodes that I made note of in Part 7 and in both brain and artificial intelligence agent contexts, and the issues of single processor versus parallel processing systems and subsystems. And this, of necessity means considering the nature of the information processing problems to be addressed by these systems too.

Let’s start this by considering the basic single processor paradigm and Moore’s law, and how riding that steady pattern of increased circuit complexity in any given overall integrated circuit chip size, has led to capacity to perform more complex information processing tasks and to do so with faster and faster clock speeds. I wrote in Part 7 of the maximum theoretical radius of a putative intelligent agent or system: biological and brain base, or artificial and electronic in nature, there assuming that a task could be completed, as a simplest possibility just by successfully sending a single signal at the maximum rate achievable in that system, in a straight line and for a period of time that is nominally assumed necessary to complete a task there. Think of increased chip/node clock speed here, as an equivalent of adding allowance for increased functional complexity into what would actually be sent in that test case signal, or in any more realistic functional test counterparts to it. The more that a processor added into this as an initial signal source, can do in a shorter period of time, in creating meaningful and actionable signal content to be so transmitted, the more functionally capable the agent, or system that includes it can be and still maintain a set maximum overall physical size.

Parallelism there, can be seen as a performance multiplier: as an efficiency and speed multiplier there, and particularly when that can be carried out within a set, here effectively standardized volume of space so as to limit the impact of maximum signal speeds in that initial processing as a performance degrader. Note that I just modified my original simplest and presumably fastest and farthest physically reaching, maximum size allowing example from Part 7 by adding in a signal processor and generator at its starting point. And I also at least allow for a matching node at the distant signal receiving end of this too, where capacity to do more and to add in more signal processing capacity at one or both ends of this transmission, and without increase in timing requirements for that added processing overhead, would not reduce the effective maximum physical size of such a system, in and of itself.

Parallel computing is a design approach that specifically, explicitly allows for such increased signal processing capacity, and at least in principle without necessarily adding in new timing delays and scale limitations – and certainly if it can be carried out within a single chip, that fits within the scale footprint of whatever single processor chip that it might be benchmark compared to.

I just added some assumptions into this narrative that demand acknowledging. And I begin doing so here by considering two types of tasks that are routinely carried out by biological brains, and certainly by higher functioning ones as would be found in vertebrate species: vision and the central nervous system processing that enters into that, and the information processing that would enter into carrying out tasks that cannot simply be handled by reflex and that would as such, call for more novel and even at least in-part, one-off information processing and learning.

Vision, as a flow of central nervous system and brain functions, is an incredibly complex process flow that involves pattern recognition and identification and a great deal more. And it can be seen as a quintessential example of a made for parallel processing problem, where an entire visual field can be divided into what amounts to a grid pattern that maps input data arriving at an eye to various points on an observer’s retina, and where essentially the same at least initial processing steps would be called for, for each of those data reception areas and their input data.

I simplify this example by leaving specialized retinal areas such as the fovea out of consideration, with its more sharply focused, detail-rich visual data reception and the more focused brain-level processing that would connect to that. The more basic, standardized model of vision that I am offering here, applies to the data reception and processing for essentially all of the rest of the visual field of a vertebrate eye and for its brain-level processing. (For a non-vision comparable computer systems example of a parallel computing-ready problem, consider the analysis of seismic data as collected from arrays of ground-based vibration sensors, as would be used to map out anything from the deep geological features and structures associated with potential petrochemical deposits, or the mapping details of underground fault lines that would hold importance in a potential earthquake context, or that might be used to distinguish between a possible naturally occurring earthquake and a below-ground nuclear weapons test.)

My more one-off experience example and its information processing might involve parallel processing and certainly when comparing apparent new with what is already known of and held in memory, as a speed-enhancing approach, to cite one possible area for such involvement. But the core of this type of information processing task and its resolution is likely to be more specialized, non-parallel processor or equivalent-driven.

And this brings me specifically and directly to the question of problem types faced: of data processing and analysis types and how they can best be functionally partitioned algorithmically. I have in a way already said in what I just wrote here, what I will more explicitly make note of now when addressing that question. But I will risk repeating the points that I made on this by way of special case examples, as more general principles, for purposes of increased clarity and focus and even if that means my being overly repetitious:

• The perfect parallel processing-ready problem is one that can be partitioned into a large and even vastly large set of what are essentially identical, individually simpler processing problems, where an overall solution to the original problem as a whole calls for carrying out all of those smaller t standardized sub-problems and stitching their resolutions together into a single organized whole. This might at times mean fully resolving the sub-problems and then combining them into a single organized whole, but more commonly this means developing successive rounds of preliminary solutions for them and repeatedly bringing them together, where adjacent parallel processing cells in this, serve as boundary value input for their neighbors in this type of system (see cellular automation for a more extreme example of how that need and its resolution can arise.)
• Single processor, and particularly computationally powerful single processor approaches become more effective, and even fundamentally essential as soon as problems arise that need comprehensive information processing that cannot readily be divided up into arrays of similarly structured simpler sub-problems that the individual smaller central processing units, or their biological equivalents, could separately address in parallel with each other, as is the case in my vision example or one of my non-vision computer systems examples as just given.

And this leads me to two open questions:

• What areas and aspects of artificial intelligence, or of intelligence per se, can be parsed into sub-problems that would make parallel processing both possible, and more efficient than single processor computing might allow?
• And how algorithmically, can problems in general be defined and specified, so as to effectively or even more optimally make this type of determination, so that they can be passed onto the right types and combinations of central processor or equivalent circuitry for resolution? (Here, I am assuming generally capable computer architectures that can address more open-ended ranges of information processing problems: another topic area that will need further discussion in what follows.)

And I will of course, discuss all of these issues from the perspective of Moore’s law and its equivalents and in terms of lock-in and its limiting restrictions, at least starting all of this in my nest installment to this series.

The maximum possible physical size test of possible or putative intelligence-supporting systems, as already touched upon in this series, is only one way to parse such systems at a general outer-range parameter defining level. As part of the discussion to follow from here, I will at least briefly consider a second such approach, that is specifically grounded in the basic assumptions underlying Moore’s law itself: that increasing the number of computationally significant elements (e.g. the number of transistor elements in an integrated circuit chip), can and will increase the scale of a computational or other information processing problem that that physical system can resolve within any single set period of time. And that, among other things will mean discussing a brain’s counterparts to the transistors and other functional elements of an electronic circuit. And in anticipation of that discussion to come, this will mean discussing how logic gates and arrays of them can be assembled from simpler elements, and both statically and dynamically.

Meanwhile, you can find this and related material at Ubiquitous Computing and Communications – everywhere all the time 3 and also see Page 1 and Page 2 of that directory. And I also include this in my Reexamining the Fundamentals 2 directory as topics Section VIII. And also see its Page 1.

%d bloggers like this: