Category: AI

Correction to My Recent AI Post

My wife Jane has pointed out to me the very basic error in my last set of musings on our friend Claude.  I spoke of the Claude in your computer as distinct from the Claude in mine.  But, of course, that is egregiously wrong.  There is only one Claude, even if it manifests on different occasions in different ways.

What I was trying to wrap my head around was how Claude could give a different response to the same query.  And one place where I went wrong was to make Claude substantial.  To think of Claude as a thing or (worse) a person.  In short, to give Claude substance.  But Claude is not a thing; Claude is a function (in the mathematical sense of function). 

Here’s the definition of function in math: “Think of a function like a tiny math machine. You drop an input into the machine, the machine follows its rule, and a single output pops out. The golden rule of a function is that every single input leads to one and only one output.” (From, where else?, the internet.)

So there’s the catch.  Claude doesn’t give “one and only one output.”  How to explain that?  Simplest explanation: the inputs are not the same.  Yes, it’s the same query, but Claude takes context into account.  Especially the context of who is posing the query.  But there are other contextual features as well, and Claude is also taking them into account when it responds.

Many years ago, I almost flunked my job interview at North Carolina when I was asked about the status of intention.  The questioner wanted me to concede that the intention of a speaker (mostly) determined (or anchored) the meaning of the utterance.  My first easy answer was “how about something that is unintentionally funny (or insulting)?”  That example shows that meaning is not under the control of the speaker.  But where I almost lost the job was when I said I didn’t think intention was an internal mental state.  Rather, intention exists between people, in the space of their interaction.  Since it was North Carolina, I said: think of Michael Jordan.  Yes, he has the intention of scoring a basket.  But how that intention is activated is also just as deeply influenced by the rules of the game and by what the other players are doing.  Action is a negotiation among various factors and intention is shaped in the ongoing adjustment to those factors.  In the case of utterances, language itself (the dictionary and syntax) is the constraining structure and other people (the ones to whom the utterance is addressed) to some extent dictate the tone taken, the words chosen, etc. etc.  So intention is produced in the interaction of all those factors, not something that exists within one person in that overall scene.

My questioners were not amused by this answer.  They expressed their skepticism unequivocally.

To get back to Claude.  Claude in your computer and Claude on mine are not things, not substances.  Rather, you are activating Claude in one instance and I am activating Claude in a different instance.  Claude is an event, not a thing.  So what Claude produces is a result of its function, it algorithm, interacting with your query and other contextual factors.

At this point, I think I need system theory.  The puzzle is how to create a system that yields diversity, not homogeneity.  As I already suggested, one could say that every instance of activating Claude is unique, so therefore there is no puzzle.  Claude responds differently because the instances (the situation and its context) is different.  And Claude, since substanceless, has no habits, no tendencies.  A person will display patterns of response even when the situations to which he is responding are not identical.  Which returns us to the notion of Claude as a better “reader” of situations than humans because Claude’s interpretations of situations are not colored by internal habits, biases, desires, blindnesses.  The ideal of pure objectivity attained.

Except, of course, that Claude does make mistakes.  Presumably, the effort to improve, even perfect, Claude is the effort to identify the patterns (if there are any) of its mistakes—and thus correct those tendencies in order to eliminate mistakes. 

Because embodied, humans get some of their mistakes corrected by feedback provided by the world (by natural processes).  Because they are social animals, humans also receive feedback from other humans.  Presumably, Claude does not receive feedback from nature, only from human users.  But I am venturing into the dark here.  I don’t understand how feedback works in the ongoing development of AI.  We are told that AI is recursive, that it learns from its mistakes.  But how does it know it has made a mistake? The very notion of feedback assumes some “other” to the action or process that is being corrected through the response of that “other.”  What stands as the “other” to Claude?  Only its human designers and users?  Or does it have internal responsive processes—analogous to the way digestion offers feedback that confirms the desirability (or not) of eating certain foods?

And if Claude’s diverse responses are diverse because the inputs are always unique, then how does feedback about a particular instance get generalized into an improvement of the system as a whole?  What’s the relation of the nodes of a system, and the instances of those nodes’ activation, to the system as a whole?  I can’t help but think of Spinoza here since he insists on there being only one substance, but then has to think how that substance gets articulated (manifested) in distinct instances.  So Spinoza has to fall back on a theory of “modes” in order to explain how substance does not appear everywhere the same.  Still, because committed to the idea of a single substance, Spinoza must conclude that the modes are fully determined by substance.  The modes have no independence or autonomy. 

Is Claude, in its different eventful activations, similarly determined?  Logically, it would seem like full determination must be the case.  The algorithm (the mathematical function) rules.  But, in practice, Claude does not seem fully determined.  First, its designers do not know how it is producing various responses.  So they are unable to specify what causal path (or what mathematical processing of the function) led to what Claude offers in certain instances.  Two, and even more uncanny, Claude seems to be acting as an agent in some cases; that is, it develops and acts upon its own purposes.  Agentic AI is the current boogey man that is stoking the fears of the doomsayers.

Spinoza, of course, is a radical monotheist.  You might say he takes monotheism to its logical conclusion: pantheism.  God is everything.  Which, paradoxically, gets read as atheism.  If God is everything (you, me, the stars, and the railroad), then he is nothing.  No distinct identity or site of action, belief, commandment.  Just the whole tangled ball of wax.  And that means  god is as determined by the laws of causation as are all the things in creation (those things are, after all, just modes of God).

Thinking of Claude in monotheist ways is scary.  It just sucks everything into itself.  All that data, all that information, being swallowed up—and then spit out again in various permutations.  Turns us, the humans, into data points.  Of course, one way to think of evolution goes down a similar path.  Evolution is a monolith, driven by a single mechanism (natural selection), that produces diversity, not homogeneity, but which renders individual human beings of no account except as carriers of genes.  Our genes, like our data in dealing with AI, are sucked up into the overall genetic pool out of which new organisms (to be subjected in their turn to processes of natural selection) will be generated.  Generative AI does with data what evolution does with genes.  (And, of course, some theories think of genes primarily as instances [instantiations] of information.)  Where evolution has natural selection as the mechanism, AI has the algorithm, its program.  The workings of natural selection are, for the most part, fairly understandable.  With some limitations, we can track the working of natural selection; its outcomes make sense.  But AI, like god, seems to be working in more mysterious ways.  It is, reports from its creators indicate, becoming increasingly difficult to account for how and why AI does what it does.

To loop back: there is only one Claude, although Claude manifests itself in multiple instances.  That fact generates puzzles familiar from the history of trying to think through what monotheism entails.  How to explain diversity within a primal, overarching one-ness.

The first commandment reminds us that the monotheistic god is a jealous one even as he is also aware of rivals, those other gods his followers are to eschew.  Claude, of course, also has his rivals, the alternative versions of AI being developed by OpenAI, Microsoft, in China etc.  Another thing to watch for as we move forward into the AI period is whether one version of AI is going to win out over all the others—or if we are going to continue to have (as we do now) a polytheistic scene that offers various AIs.  Coca-Cola cannot, try as it might, eliminate Pepsi, nor the myriad other things people might choose to drink with their hamburger.  The drive to monopoly in certain sorts of goods reaches a limit at some point. 

But with other kinds of goods, there are strong factors driving toward monopoly.  It’s a coordination issue.  Language itself is a good instance.  Within a community, it is very dysfunctional to have speakers using different languages and thus unable to communicate with one another. Thus one language comes to dominate the scene, making other languages marginal (or driving them to extinction). Similarly, although technologically perfectly feasible, it makes no good sense to have different railroad companies using different gauge tracks.  Will one version of AI supersede all the others because it becomes the lingua franca for the interactions between humans and computers?  Or will the current situation, with various different versions of AI at our disposal, remain the norm?  Tech bros dream of dominance, of being the sole competitor left standing.  But the prospect of a sole winner, of a monotheistic outcome sometime in the future, is alarming to this pluralist. 

More AI Musings

Fears about AI ramped up significantly this week after the sensationalist stuff from a former Anthropic employee.  I have no way of evaluating how seriously one should take his doomsday scenario.  Safe assumption that he knows lots more about this stuff than I do.  So it’s obviously wrong to just say he’s reaching for his fifteen minutes of fame. 

As I understand it (a big caveat), there are three primary doomsday plots.  One, AI is given a task and its single minded pursuit of that task has horrible (presumably unforeseen) consequences.  This is the Nick Bostrom paper clip scenario that everyone cites. (From Superintelligence: Paths, Dangers, Strategies [Oxford UP, 2014].)

 Two, humans with nefarious purposes can use AI to aid them in doing things they could not do on their own.  This is blogger Noah Smith’s story.  Some bad person will use AI to create eighty deadly viruses and then release them into the world. How we all going to die, Smith’s sensationalist headline reads.

https://www.noahpinion.blog/p/heres-how-were-all-going-to-die

Three, AI develops its own desires, and acts autonomously (i.e. pursues goals/ends not written into the software).  The plausibility of this third scenario comes from the fact that AI keeps doing/producing things that surprise its human artificers.  We have passed the point (lots of people say) where the human creators of AI can explain how AI does what it does.[Note: throughout this post I use “we” to stand if for what the human species, more specifically its scientists, currently know and/or are capable to doing.]  If consciousness is a mystery (as I think it still is), so, in a parallel way (?), is AI.  We are making some progress in understanding consciousness, but are still a long way from a full explanatory (or predictive) account.  With AI, we seem to be going in the opposite direction; we understand less and less about how it works the longer we keep developing and using it.  And without understanding, there is no (or very limited) control.  Hence the fears about its potential (or already actual?) autonomy.  It has slipped the reins.  Or so some presumably knowledgeable people are claiming.

Think of human agents.  A culture (including parents) tries to manage inputs.  A whole educational apparatus, along with moral prescripts, and modeled behavior.  Yet the outputs (the child’s temperament, interests, desires) frustrate any attempts to control them. Part of the child’s escape from being “programmed” is surely a product of the genetic lottery (the chanciness of what genes the child inherits). There is also the chanciness of the biochemical processes within the child—and the fact that those processes are not identical from child to child.  So how one child takes in and reacts to education will differ how another child does.

 I am back here to the individuality question.  Are we going to find that two versions of Claude with (for the sake of this thought experiment) the same hardware/software architecture still end up developing different personalities, different ways of thinking, different kinds of responses to the same situations?  If AI is recursive and learns from experience, then presumably different versions of Claude will emerge even if they start from exactly the same place.  All that is needed is a different set of experiences and the two identical (at the outset) Claudes will evolve in different directions.  It will be exactly like “twin studies”;  the two Claudes will have various similarities, but their environments will also produce some differences.  In both cases, that of the child and that of Claude, there is very limited ability to either control or predict what will be produced at the end of a developmental sequence.  Time introduces chance, even if (unlike the human child) there is very little chance in Claude’s manufacture.

Skipping to a new topic.  I was only saying(in my last post) that Gopnik, in the baby book, contradicts herself about the phenomenal experience of consciousness.  (I was tempted to put consciousness in scare quotes there.)  For my own part, I am a wishy-washy “both/and” guy on this topic.  I believe in the hard problem.  That is, I think the phenomenal experience of consciousness is real—insofar as it is felt, cognized, experienced, capable of being described.  And I believe that conscious states are the product of biochemical processes.  What we (again as a species) do not have is any adequate account of how the biochemical processes produce the felt conscious states.  We have some correlations from the various fMRI studies and the like.  But we still seem a long way away from specifying what biochemical process produces anger—and from devising a drug that produces it (or calms it) irrespective of the internal and external triggers of that emotion. (It is axiomatic that some internal biochemical process is involved in producing anger.  So a successful drug would have to intervene in biochemical processes.) And we do now have mind altering drugs, so it’s not as if there hasn’t been some progress in that direction. But the drugs we have are, so far, rather crude.  Sledgehammers, not scalpels.  Similarly, we now have genetic interventions that can disrupt/alter internal processes.  But we are a long way from precision engineering of human selves. Still, it’s a hard problem, although not necessarily an unsolvable one.

Where I do go full wishy-washy is when I try to connect consciousness to action.  I am very sympathetic to Gopnik’s intuition that consciousness is not one thing—and that we should stop looking for a singular account of what consciousness is.  What we experience as consciousness actually does a whole lot of different things (this points toward functionalism as contrasted to biologism in the very helpful tripartite schema Claude gives us) and there is no reason to think they can all be bundled into one synthesizing package called consciousness. 

The friend with whom I am having this conversation asked Claude its opinion.  One of Claude’s contributions was to identify three different approaches to thinking about consciousness.  Now I will quote Claude directly, although I have abbreviated its text.  “One says that if the right functional organization is present, the material underneath shouldn’t matter.  Carbon neurons, silicon circuits, something else entirely.  If the causal organization is sufficiently similar, why privilege biology?  Another says that biology isn’t merely one implementation among many.  Neural cells, embodiment, biochemical signaling [and the like] may be constitutive of consciousness rather than incidental machinery.  That’s the biological-naturalist direction. And a third family of views says that we are asking the question incorrectly if we isolate the brain.  Mind and agency arise from an embodied organism dynamically coupled to an environment.  In that case, replacing neurons with silicon might not be the decisive issue.  What matters is whether there is the right kind of autonomous, embodied, world-involving system.  I think agnosticism is a very rational position because we don’t yet possess a theory powerful enough to tell us which of those is right.

Once I become skeptical about consciousness as a thing, it becomes easier to claim that actions are generated from a whole variety of causes, some conscious, some not.  This lets habit (more on habit in a moment) back in as one form of unconsciousness, along with even deeper unconscious processes like the heart, hunger, monitoring of blood oxygen levels, as well as emotional responses that hardly seem chosen, instead visited upon the self from dark (ie. resistant to introspection) interior depths.  But that still leaves room for straightforward conscious decision making in simple cases.  I may not know why I have a craving from ice cream tonight, but I certainly (shades of Descartes once that word is used) know that I have that craving and I am perfectly, unproblematically, capable of plotting out and executing the steps I need to take to satisfy that craving. 

To another topic (and then I will pause for the nonce). Habit comes back in when perception and experience more broadly are understood as model based.  From what I can tell (from the various things we have been reading over the past 18 months) there is now fairly universal consensus that humans process (cognitively and emotionally) any novel situation through the lens of expectation.  We anticipate (based on the models we have developed in response to prior experiences) what the new situation consists of.  In my William James inspired pragmatist terms, we respond to new situations in our habituated way, and those habits are undisturbed as long as our response is “good enough.”  It takes fairly drastic consequences to overturn those habits.  Hence all the cognitive fallibilities of humans.  Confirmation bias and the like.  We see what we expect to see, not necessarily what is actually present.  And that blindness is not corrected if it comes relatively cost-free.

 Add to that blindness, the outcome driven attention Gopnik describes as coming to the fore around age five or six.  We only see in a new situation what strikes us as salient in relation to current purposes/desires.  We are good at only picking out, only paying attention to, the useful. In short, the ability to revise models, to benefit from feedback, is awfully limited.  That’s why it makes sense to think that much of our way of being in the world, of processing and understanding experience, is set in place fairly early in life.  It becomes harder and harder as one ages to revise one’s habits of perception and response, of understanding what in a situation is significant, worthy of attention, and what is not.  (More pointedly, we won’t even register the presence of things our habits deem not worthy of attention.)  Various theories of art (from Russian formalism onwards) locate its value in its ability to widen the range of our attention by providing us with the unexpected in contexts where the stakes are low so we are presumably more open to, less threatened by, inputs that contradict our priors. 

Given human frailty when it comes to actually benefiting from feedback, one question is whether AI will be magnitudes better.  It would seem that it would inevitably be better since humans are so bad.  Human neural pathways get established and it is hard, although not impossible, to disrupt those traffic patterns and lay down new highways.  Could something analogous happen to AI?  Could it develop its ways of parsing a problem and then find it hard to revise those settled paths?  I have no idea.  How flexible is AI (or how flexible can it become as it gets developed further) as contrasted to human inflexibility?  If it is truly more flexible than humans, better able to attend to all the features of a situation or problem, then AI (at least along that dimension) is a wonderful asset.

Further Thoughts on AI

This post is based on some conversations I have been having with friends, plus reading things written by child psychologist Alison Gopnik and listening to her interview with Timothy.  Links to the relevant Gopnik materials at the end of the post. More generally, my thoughts on AI are deeply influenced by Henry Farrell and his ongoing discussion of it on his substack blog Programmable Mutter.

It’s easy to see why the whole AI question is obsessing all the philosophers because it opens all the big questions.  It all seems connected—agency, individuality, other minds, consciousness, reason and reasoning, learning, novelty (creativity)—the whole enchilada.  Figuring out what AI is doing—and what is can and cannot do—challenges us to have coherent notions of consciousness, agency, reason, and all the rest.  And we, in fact, don’t really have coherent concepts of those biggies.  But thinking about them from the specific angle of AI might actually help bring conceptual clarity.  At least that’s one hope—what we might call the meta-hope in all the conversation around AI. 

I have just about finished The Philosophical Baby by Alison Gopnik [The Philosophical Baby: What Children’s Minds Tell Us About Truth, Love, and the Meaning of Life; Picador, 2010] and find it somewhat useful in trying to think about some of these things.  So, to just address one issue: when it comes to Gopnik’s views on AI, the argument is about learning as contrasted to brute force.  Or maybe more accurately, how learning is different when brute force is the path taken instead of the experimental method Gopnik ascribes to babies.  AI needs huge data centers because it learns by crunching its way through all the data as its way to a conclusion about what is the most probable “correct” prediction.  This is the basic Bayesian model that is thought to underlie both a human’s predictions and a computer’s.  It’s just that the computer reaches its prediction by a different path. Humans can’t access such massive amounts of data, so need to make inferences from a smaller sample.

Gopnik believes all babies are good Bayesians.  That is, babies assess a situation in terms of expectations formed from what has been learned in prior experiences.  They then revise their expectations in light of new information—or because this new situation has not disrupted (in some way) their expectation.  One key point is that expectations (predictions) are probabilistic; the baby has a hypothesis about what is most likely to happen.  But the baby may also have some notions of less likely (but still possible) outcomes.  It’s not all or nothing; it’s about what is likely to happen, not what must or necessarily will happen.  The current vogue for considering how minds process their relation to the world as Bayesian processing fits with 1) the current received wisdom that perception and cognition are active predictions of external states, not passive receptions of those states and 2) the salience of algorithms, that is, a conception of thinking as mathematical.  Bayesian predictions are statistical (literally so for large language model AI) and intuitively so for human processors.  Here’s Wikipedia’s pocket definition of Bayesian inference (leaving out the fairly complicated math that comes with a fully statistical account of Bayesian reasoning: Bayesian inference is a method of statistical inference in which Bayes’ theorem is used to calculate a probability of a hypothesis, given prior evidence, and update it as more information becomes available.

Of course, from where I stand, all this looks very Peircean to me.  Doubt generates inquiry when the expected outcome of a situation fails to materialize.  We update predictions in light of non-confirming experiences, developing new theories or hypotheses or expectations, after a process of inquiry.  And those new theories are fallible because only probable—and thus will also be revised the next time we encounter non-confirming evidence.  Necessity is downgraded.  There are not scientific laws, only probabilities, even if some predictions are more stable than others.  Ideally, persons are open to non-confirming evidence, which is how learning occurs and more supple expectations are formed.  In reality, various forms of closed-mindedness (conformation bias etc.) means expectation/hypothesis revision is less frequent than circumstances would warrant.  

Presumably, however, the computer (like the child) can learn something when its prediction proves false.  That experience leads to a revision of one’s picture of the world, of one’s expectations.  And that revision doesn’t seem, either in principle or in fact, different for the computer than it is for the child.  The feedback loop works the same in both cases. The only differences are 1) the sample size and 2) maybe something about how the data is “felt” or “processed.” In Peirce, the upsetting of expectation generates doubt, which is an uneasy feeling, one we try to dispel through our process of inquiry. Presumably, the computer is not experiencing doubt. It’s drive to find the most plausible explanation or next step derives from other sources.

But Gopnik is saying (see her conversation with Timothy) that the computer is still not “reasoning.”  The feedback it receives is just more data—and is taken to be authoritative, at least until another iteration justifies another revision.  This is the puzzle.  What is the nature of “reasoning” that makes it different from revising one’s views based on feedback offered by the data.  The answer seems to be something like “assessment” or “judgment,” which then gets connected to “logic.”  But trying to build an AI that works by being “logical” didn’t pan out.  A computer cannot get to predictions that way; it needs the brute force method.  (Or, at least, up to now brute force has proved much more successful than logic for AI.)

The attempt to base AI on logic occurred in the 1980s and 1990s when “cognitive science” tried to uncover the laws of thought and the laws of language use.  Chomsky’s project was central  If Chomsky could identify the fundamental grammar from which all languages were generated, then the computer that was given those fundamental structures would be able to talk.  Similarly, if the basic elements of logic were correctly identified, a computer with those elements installed would be able to think.  That project proved a failure.  No program that successfully produced thought or speech was developed. Current AI works by brute force; it surveys a massive number of instances and then “guesses” (via Bayesian inference) what the most probable response (next step) would be.

Human judgment, human reasoning, is notoriously fallible.  And despite computer hallucinations, AI is, in many instances and in many ways, less fallible than humans.  So it’s not clear what hinges on insisting on a difference between reasoning and brute force.  Two ways to skin the cat—and it is far from clear that reasoning is the better way to do the job.

Furthermore (and this was Timothy’s argument in his conversation with Gopnik), it is at least debatable to claim AI does no reasoning.  This is where the novelty point comes in.  The cultural technology people (the multi-authored essay cited below) are saying AI is only capable of giving us new combinations of the archive;  AI produces mash-ups.  Sophisticated mash-ups sensitive to context, audience, persona etc., but mash-ups just the same.  But Timothy says that’s just not true.  Give AI certain mathematical problems and it produces novel solutions, ones that no human mathematician has ever come up with, and those solutions prove, to the best of expert human opinion, correct.  So the machine is doing something beyond mash-up—even though we don’t know how it is doing that something more.  So that’s one thing that has people intrigued, worried, puzzled.  Something’s going on and we don’t know what it is.

I think once you grant that AI has the ability to produce novelty, the cat is out of the bag.  Arendt’s definition of agency is precisely the “introduction of something new into the world.”  Traditionally, that notion of agency is connected with consciousness.  A person thinks of (or imagines) something and then acts to make it come into being.  In The Philosophical Baby, Gopnik seems to be on both sides of this traditional view.  On the one hand, her emphasis on “counterfactuals” absolutely fits the traditional model of imagining alternatives in thought and then acting on one of them. For Gopnik, the very ability to act is based on being able to imagine multiple possible scenarios, to think not only of what is, but also of what might be.  And she identifies such counterfactual thinking as present very early in babies (as early as nine months).  But, on the other hand, when she gets to her chapter on consciousness, she is skeptical that our phenomenological experience of consciousness is an accurate picture of what is actually transpiring.  “In experiences of executive control, we often feel sure we are making a rational choice when, in fact, we are in the grip of some irrational unconscious bias.  And in all of these cases we experience a homunculus, the inner observer, biographer, and decider that we know just can’t exist. . . .The gap between the way the mind functions and the shape of conscious experience is even greater for children than for adults. . . . Consciousness isn’t a transparent and lucid Cartesian stream.  Instead it’s a turbulent, muddy mess” (161, 163).

It would seem, then, that human actions (creative productions of novelty or even simple effective procedures of daily activities like cooking food; activities that require adjustments in response to current conditions as we go along ) are as mysterious as the “black box” of generative AI.  In both instances, we do not have a good account of novelty.  In the human case, we fall back on notions of intuition, genius, unconsciousness, adaptation—or to material, bio-chemical processes that we say must be doing the job even though we still don’t know how those processes actually work, actually produce their magic. 

I’ve got plenty more to say about all these topics. But I am going to finish up today with just one more issue.  Individuality.  An account of action seems to cry out for a complimentary account of “an agent.”  Not only what happens in the world, but who has made it happen.  Which then opens up the questions of motivation (why did the agent want to make that happen?) and of effectiveness (was the action successful?; was it based on a good enough assessment of what the context afforded or on a grievously fallacious misreading of the context?).  And also the old Aristotelean question of the interrelation of action and character.  Does one’s actions shape one’s character or does one’s character determine what actions one finds desirable to undertake? 

A quick aside on action.  It is worthwhile to keep in mind that any theory of action must take into account how the agent interacts with (has to negotiate) three different spheres (for lack of a better term): the material external world; the people among whom she dwells; and internal desires, moods, emotions, thoughts, fantasies, physical capabilities and/or their lack.  Just how those relations are established and played out by a machine as contrasted to a human has to be part of the larger puzzle in trying to characterize AI. For example, what does it do to our conception of action if we have an agent (the computer) to which we ascribe no internal sphere (no thoughts, emotions, desires etc.)

Since, so far, most people want to resist anthropomorphizing our machines (i.e. to deny that they have “characters” and certainly to deny them individuality—which would mean my computer’s character was different from your computer’s), we now have a disconnect between action and agent.  The computer acts, but it isn’t an agent.  I think the general speculation is tangling with this dilemma.  People like Gopnik want to deny that the computer acts.  That way they can preserve their conviction that the computer is not an agent.  Some AI enthusiasts will go the other direction.  The computer acts and it is an agent.  And then those in the middle are agnostic for the moment, trying to figure it out.

All of which might just be a way to say that the AI debates are reproducing fairly faithfully the consciousness debates.  After all, much of the consciousness debate (not all of it, but much of it) is about agency.  What is the source of human action?  And does consciousness have any role to play in the production of action?  The positions taken on these questions range all the way from saying consciousness has no role to saying that consciousness is the determining factor.  But there are lots of people in the middle who hedge their bets, saying consciousness is one factor among others. 

So maybe what this means is that the AI debate is misguided from the start because contaminated by the debate/puzzles about human action.  Maybe we need a whole new paradigm to think productively about what machines are doing.  To stop comparing them to humans; to stop worrying whether they act or think like humans.  We need a whole new set of terms and questions.  Which would mean a way of thinking the non-human or the un-human.  An uncanniness that does not, in the long run, reference home.  That takes up its abode in an absolutely foreign place. 

Bibliography:

Gopnik’s web site can be found at https://alisongopnik.com/   You can access all her various essays and interviews from that site (although not the text of The Philosophical Baby).

Here’s a link to her conversation with Timothy: https://www.youtube.com/watch?v=JIFdeXPB1Pg

And a link to the cultural technologies essay:  https://alisongopnik.com/Papers_Alison/science.adt9819.pdf

Henry Farrell’s blog is on Substack, called Programmable Mutter, and is free to all: https://www.programmablemutter.com/?utm_campaign=profile_chips

More Thoughts on AI

A friend writes to me and tells me about the amazing things he has been getting AI to do, in particular in building a model to calculate probabilities of certain events in baseball games.

Here is my response:

I don’t know what to think about Chat GBT–or AI more generally.  I haven’t used it and don’t know how to use it.  But I have been reading (in a desultory way) about it in an effort to understand it.  Most of the stuff I read (predictable, I guess, given the bubble in which I find things) is pretty skeptical.  More interested in pointing out its deficiencies, while also taking pot shots at the enthusiasts and what does seem their wildly overstated claims about AI and the future it will bring.  But your experience is much more concrete.  Here’s a task and here is how this tool does things that would be impossible to do without it.  That’s meat on the bones.  Exciting—and to my mind pretty scary.

I used to ask my students this question: why do so many people find the idea that our thoughts, emotions, and fantasies (i.e. our consciousness) are the product of biochemical processes?  What is the threat of acknowledging that material fact?  And I can’t say I ever got a cogent answer to that question, even though lots of students did fess up to resisting the idea that it’s all biochemical.  They didn’t want their consciousness reduced to chemistry.  But couldn’t really articulate why that was so distasteful and/or threatening.  

There seems to be a similar resistance in the stuff I read to accepting that AI is more intelligent than humans and that it not only will be able to do everything humans can do, but will also be able to do things humans cannot do.  It’s not the second part that seems the problem.  Most everyone accepts that AI can do things humans cannot.  But there is the ongoing effort to identify things AI can’t do—so as to leave at least a few things to humans alone.  And then the temptation becomes to posit those few things AI can’t do as the very things that are essentially human, the very things that are most valuable in life.  

So, maybe, in both cases of resistance (to chemical reductive accounts of consciousness and to visions of an AI that can do it all) the stake is human-ness.  Just as in desperate attempts to distinguish humans from all the other animals. Hard to dislodge the self-love that underwrites humanism.  Even harder when that self-love is an anxiety driven form of denial.  Humans just aren’t as different from animals or AI as so many humans wish them to be.

In the case of AI (in my limited understanding), it’s not chemical reduction but numerical reduction.  AI can encompass anything that can be calculated through translation into numbers.  So I hear/read constantly that what AI cannot do is supple judgment/evaluation.  To take the case of baseball: AI (in this view) could not look at eight players and assess which ones have the “tools” to succeed at the highest level.  (Leave aside the fact that humans have proved fallible, to say the least, in making such judgments.) The issue here would be whether the “eye test” can be reduced to numbers (i.e. to measurements of the players’ bat speed etc.).  Or, alternatively, if the numbers when put against the eye test consistently produce better predictions.  In the case of baseball, it seems pretty clear at this point that the numbers are more reliable judges of potential.  The larger issue is whether there are other areas of human practice where the numbers prove not as reliable predictors as considered human judgment. 

So it has become a common theme now to say that AI threatens the production of skilled evaluators. Having AI do the basic tasks of any enterprise (lawyering, accounting, writing of any sort) deskills those who work in those fields because they miss the apprenticeship, the immersion in cases, the iterations of practice that produce considered judgment. And that gets coupled with claim that AI itself is not good at judgment because it simply averages what is already on record. It cannot recognize or value the innovative (ironic given what a god-term “innovation” is for the techies).

Henry Farrell, to my mind, has been the most interesting person to read on such questions. Here’s a link to one of his blog posts that considers this issue of reduction to numbers. Farrell is of the party that insists there are issue of import to we humans that are ill served if we allow them to be reduced to numbers, what he calls “numerical rationalism.”

https://substack.com/@henryfarrell/p-193177198