Jon, Jeff, Ulysses, List, Jon, thank you for a response that has once again caused me to rethink my own position. Your citation of CP 2.27 seems especially important, although I find myself drawing a somewhat different conclusion from it than you did. As I read that passage, Peirce is reminding us that when we later reconstruct our reasoning as an explicit argument that we are not reproducing the original process of thought. We begin with the conclusion, formulate it propositionally, and then reconstruct premises that justify it. The result is, as Peirce says, the "self-defence" of the inquiry rather than the inquiry itself. Indeed, he clearly doubts whether the propositions we come up with truly resemble the original thinking.
If my reading of that passage is even approximately correct, then it seems to me that Peirce is drawing a distinction not between human reasoning and, say, machine reasoning. Rather, it is between living inquiry and its subsequent symbolic representation. That argument is not the original process of thinking but, rather, it is itself a newly generated sign that enters the continuing inquiry where it may be interpreted, criticized, revised, etc. and possibly become the source of further interpretants. And this is what gives me pause regarding the present discussion. Because if neither the printed syllogism nor the written scientific paper captures the actual movement of thought, then the question we've been considering cannot simply be whether AI produces arguments that represent reasoning. After all, we humans do not ordinarily exchange 'living thought' with one another either; instead, we exchange signs that become occasions for further semiosis. Returning to my earlier thought experiment. My point was not that communal inquiry exhausts intelligence (Peirce's example of the animalcule shows otherwise). Rather, I was trying to consider *where the interpretants that matter for, especially, scientific inquiry actually emerge*. These clearly do not reside in the lecture, the paper (or, for that matter, the AI system). Rather, they emerge in the continuing inquiry which those signs bring about. This is why I continue to resist thinking that AI merely represents reasoning in the same sense that, say, a printed syllogism does. Certainly AI generates symbolic representations, but those are typically not the end of the matter. Rather, as I earlier suggested, they become new signs that may reveal previously unnoticed distinctions, reveal inconsistencies, even suggest hypotheses, etc. Their significance, therefore, seems to lie in *what they become within the evolving inquiry*. Jon, it seems that you locate intelligence primarily -- if not exclusively -- in the process of human reasoning itself, whereas I find myself increasingly tending to locate it in the larger semiotic process through which interpretants are continually generated, developed, etc. While those positions may be related, they are most certainly note identical. Best, Gary R On Sun, Jul 5, 2026 at 8:22 PM Jon Alan Schmidt <[email protected]> wrote: > Gary R., Jeff, Ulysses, List: > > I am grateful for everyone's participation in this stimulating discussion. > > The problem that I see with locating intelligence "within the broader > process of inquiry" as Gary is proposing is that it rules out forms of it > that are *not *engaged in communal inquiry, such as Peirce's own example > of a single animalcule's motion producing a second event as a means to the > production of a third event. Hence, Gary's thought experiment strikes me as > entirely beside the point--even if there is no relevant semiotic difference > between the three cases, it has nothing to do with whether "AI" systems can > be accurately described as *intelligent*. Instead, it simply demonstrates > how they can be used *by humans* as highly sophisticated tools for > advancing *our own* inquiries. It also suggests a danger--some might stop > attending conferences, reading papers, and otherwise collaborating with > fellow humans altogether because they come to believe that they can get > everything they need by interacting with "AI" systems. > > Jeff's example of a printed syllogism is an instance (token) of a type, an > individual embodiment (2ns) of a general sign (3ns), and the type itself is > a retrospectively formulated *discrete* representation--two premisses and > a conclusion--of what was really a *continuous *process of inference. I > apologize for the length of the Peirce quotation that follows, but the > referenced paragraph has nearly 1,200 words and is honestly worth reading > in its entirety. > > CSP: Practically, when a man endeavors to state what the process of his > thought has been, after the process has come to an end, he first asks > himself to what conclusion he has come. That result he formulates in an > assertion, which, we will assume, has some sort of likeness--I am inclined > to think only a conventionalized one--with the attitude of his thought at > the cessation of the motion. That having been ascertained, he next asks > himself how he is justified in being so confident of it; and he proceeds to > cast about for a sentence expressed in words which shall strike him as > resembling some previous attitude of his thought, and which at the same > time shall be logically related to the sentence representing his > conclusion, in such a way that if the premiss-proposition be true, the > conclusion-proposition necessarily or naturally would be true. ... At any > rate, it is only the self-defence of the process that is clearly broken > up into arguments. It is more than doubtful whether what we can state as > an argument or inference represents any part of the thinking except in > the logical relation of the truth of the premiss to the truth of the > conclusion. And, moreover, the argument so considered consists in the > statements in words. How nearly they represent anything really in the > thought is very doubtful, and is quite immaterial. (CP 2.27, 1902) > > > Likewise, it seems to me that current "AI" systems can *represent > *genuine/continuous/triadic > reasoning, but they cannot *perform *such reasoning and thus are not > themselves intelligent. Ulysses seems very knowledgeable about how they > work, and his observations about *iconicity *as the essence of what they > are doing gets at what I have in mind when I insist that they only > *simulate *the behaviors that we associate with intelligence. > > Regards, > > Jon Alan Schmidt - Olathe, Kansas, USA > Structural Engineer, Synechist Philosopher, Lutheran Christian > www.LinkedIn.com/in/JonAlanSchmidt / twitter.com/JonAlanSchmidt > > On Sun, Jul 5, 2026 at 6:53 PM Gary Richmond <[email protected]> wrote: > >> Jon, Jeff, Ulysses, List, >> >> From all that has preceded this discussion, the question now appears to >> me to be whether intelligence (a.) should be located primarily in isolated >> agents or, (b.) in the communal (considered in its most expansive, Peircean >> sense), *world-directed semiosis* through which interpretants are >> continually generated, tested, criticized, etc. >> >> A brief thought experiment which might help clarify this question: >> >> Imagine, (a.) a scientist attending a lecture on a difficult problem in >> her field that she herself has been investigating. During the lecture she >> encounters a novel line of argument that she had not previously considered. >> Returning to her office, she substantially revises her own theory; her >> subsequent paper is recognized by her colleagues as not only confirming the >> lecturer's thesis but also extending it. >> >> Now imagine a second case. (b.) Instead of attending a lecture, the >> scientist reads a recently published paper. Again, a new line of >> argumentation reshapes her thinking. Some months later her own paper on >> the topic is judged by her peers to have confirmed the novel thesis and to >> have advanced the inquiry. >> >> Finally, a third case. (c.) The scientist engages in an extended >> discussion with an AI system concerning that same or a similar problem. >> During the exchange the AI proposes an illuminating analogy, exposes an >> unnoticed inconsistency, distinguishes two concepts that had previously >> been conflated, or even reformulates the problem in a way the scientist had >> not earlier considered. She revises her own work in light of certain of >> those AI generated ideas, and her colleagues again regard the resulting >> paper as a genuine contribution to the field. >> >> *From a Peircean standpoint, what is the relevant semiotic difference >> among these three cases?* >> >> In each instance the scientist is responding to signs by generating >> further signs. Those additional signs then become objects -- in the >> hypothetical example, academic papers -- for subsequent interpretation >> within the community of inquiry. It seems clear enough to me that the >> interpretants involved are not private possessions but, rather, they are >> public 'candidates' for criticism, revision, rejection, or acceptance. >> Whatever their original source (I considered 3), *their significance >> ultimately lies in their consequences for the continuing inquiry.* >> >> Jeff's post seems to reinforce exactly this point: AI-generated signs >> only become genuine signs when they enter the communal process of criticism >> and further interpretation*.* I find your observation especially >> illuminating, Jeff, because it shifts attention from the internal >> architecture of the machine toward *the larger semiosic processes in >> which both human and machine-generated signs can participate*. >> >> Ulysses, thank you too for reminding us of something which is no doubt >> just as important: Language is not some kind of formal symbolic game but, >> rather, *language is dynamically connected to the world through indices, >> perception, action, etc.* And, as you also remark, AI systems remain >> dependent upon human training, human purposes and, perhaps most >> importantly, human interaction with reality. >> >> Inquiry is rarely, nay, *never* truly or fully conducted in isolation. >> To cut to the chase, and to put it in the context of this inquiry, *the >> semiosis isn't in** the AI system any more than it is in the individual >> scientist*. Again, inquiry ordinarily involves inquierers, >> conversations, publications & institutions and, perhaps most importantly, >> the stubborn resistance of reality itself! In a word, *AI enters >> an already existing ecology of signs as an additional participant in the >> production of interpretants*. >> >> This, I think, is also where Jon's insistence upon genuine 3ns remains >> indispensable. If the interpretants generated through AI-assisted inquiry >> merely produced the illusion of advancing inquiry, they would eventually >> fail under criticism by the given community of inquiry. But if they survive >> criticism, or clarify concepts, or expose unnoticed inconsistencies, or >> generate fruitful hypotheses, then it becomes difficult for me to see them >> as merely simulated interpretants. Again, their significance doesn't lie in >> their origin but in their continuing semiosic consequences. >> >> For that reason, I have come to believe that *the principal locus of >> intelligence* in Peirce's philosophy is not the individual thinker, but >> the evolving semiosis through which signs continually generate further >> signs under the dual-discipline of *individual experience and communal >> reflection/criticism.* I believe that the history of scientific thought >> shows that the growth of inquiry has always depended upon signs whose >> consequences often far outrun the intentions of those who first produced >> them. Perhaps, the, AI represents a new, derivative, and still quite >> limited but *real* participant in that larger process. >> >> If that's the case, the philosophical question that now seemingly >> naturally presents itself to me is whether the communal growth of >> intelligence may now include forms of semiotic participation that Peirce >> himself could scarcely have imagined, and yet while remaining thoroughly >> consistent with his conception of inquiry as the self-corrective growth of >> inquiry via the interpretant objects -- lectures and books and papers, etc. >> -- that it generates. >> >> Best, >> >> Gary R. >> On Sun, Jul 5, 2026 at 1:45 PM Ulysses <[email protected]> wrote: >> >>> 1. What counts as an interpretant? Are summaries interpretants? Or are >>> (human) summaries artifacts of interpretation? >>> >>> 2. What does this mean: “But they do appear to participate in >>> thoroughly genuine triadic relations when they generate signs that take >>> other signs as objects and produce further signs capable of functioning as >>> interpretants” >>> >>> It’s unclear what “genuine triadic relations” means here. I think we all >>> agree that there is a sense in which LLMs participate in semiosis, that >>> doesn’t mean it does so in a genuinely tradic way. It seems to me that LLMs >>> operate on copies (icons) of symbols and produce diagrams that represent >>> statistical patterns (icons of indicies) (albeit extremely complex diagrams >>> with billions of parameters). An LLM is quite literally an algebraic >>> formula with billions of weights and peirce often calls algebraic formula >>> diagrams. >>> >>> It’s unclear to me whether an AI summary is an icon of the relation a >>> larger text has to a human written summary, or if it is in some real sense >>> a genuine exemplar of a summary. I think mechanically it is a “mere” icon >>> but “mere” is doing a lot of work, and might necessitate explanding the >>> breadth of the symbol “summary” to include both human and AI generated >>> summaries. However, shifting language use to allow for AI generated >>> artifacts to be grouped alongside human generated ones still does not make >>> an AI generated summary “genuinely triadic”. >>> >>> Observing the internal mechanics of an LLM, it seems like the way an LLM >>> produces a summary is not through a deliberative or habit forming process. >>> Rather it is applying an induced “law” (in the form of an algebraic formula >>> ie diagram) to an input sequence which results in an output sequence that >>> is summary shaped. Because this mechanical application of a dyadic >>> representation of a law is not fed-back into a system that is continually >>> and dynamically coupled with habit formation, i have a hard time say that >>> AI products like chatgpt are genuinely triadic. >>> >>> I am more inclined to say the training of an LLM includes some >>> approximately triadic moments (around the attention mechanism and MLP >>> layers) which allow it to produce a continuous, nearly differentiable map >>> of input sequences to output token probabilities and this continuum is like >>> a frozen slice / effete artifact of something that is quasi-mind like—but >>> it is still a “mere” diagram. >>> >>> What is critical to remember is the training process is optimized to >>> minimize the difference between predicted next tokens and observed next >>> tokens taken from a large corpus of training data. This project of >>> maximizing the *similarity* between a prediction and an observation seems >>> fundamentally iconic. While it is easy to say “words are symbols, therefore >>> it’s symbolic” I would push back and say that the model is trained on >>> tokens of symbols (tokens in a piercean sense). it’s not like the model is >>> being trained on the genuinely triadic relationship that symbols have >>> between objects and interpretants directly… all it has access to are >>> statistical distributions of tokens of these symbols. >>> >>> One could push back and say that while the goal of next word prediction >>> is itself degenerate, it puts a constraint on the training process that >>> forces the model to learn something more general and law-like somewhere in >>> the transformation process between the input sequence and the output >>> probabilities. I think is is correct but still, because the goal of next >>> token prediction is so artificial and arbitrary, the induced “laws” of next >>> word prediction the model learns are only incidentally or tangentially >>> related to intelligence. A similar argument could be made for evolution, >>> where the single evolutionary goal of “next offspring creation” is also >>> arbitrary and tangentially related to intelligence— this is true (many >>> living things aren’t intelligent) but the also the constraint of >>> surviving long enough to reach maturity is complex enough that intelligence >>> is highly useful. Ultimately i think the constraint of next word prediction >>> likely partially overlaps in some abstract and general ways with the >>> constraint of natural selection and thus results in some overlapping >>> families in intelligent transformation rules /encoding decoding procedures. >>> But whereas actual survival requires a continuous and dynamic coupling >>> between habits, habit formation processes, and the physical world—the >>> training goal of next token prediction and the resulting model lack this >>> integrated holistic feedback. >>> >>> While so-called reasoning models change the goal a little bit, by >>> filtering models based on whether icons of reasoning-like language are more >>> likely to produce the effects of reasoning (ie by asking models to solve >>> reasoning problems and keeping only models that tend to survive this task) >>> this training regime still doesn’t not strike me as genuinely triadic >>> (though admittedly it is harder to articulate why). To make it clearer, >>> formal logic has long been able to represent the formal structure of >>> deductive reasoning, but we would not call a purely deductive logical >>> program genuinely triadic because its execution of logic is not fed-back >>> into a system that reasons about its own habit formation process. It >>> doesn’t have the ability to “grow” (continuously partake in continuity?). >>> Systems for training reasoning models are a little closer something that >>> can grow because they have the ability to prune ineffective reasoning >>> chains but they lack the ability to reason why one chain vs another is >>> effective, and thus are still more like a mechanical and brute force >>> approximation of reasoning. Likewise the models we use in consumer facing >>> AI products are the effete artifacts of these training procedures and are >>> thus doubly degenerate wrt thirdness. >>> >>> One challenge to the argument that LLMs are “mere” simulations is that >>> some things are iconic in their nature and thus a simulation of them is not >>> categorically distinct from the thing being simulated. A computer screen >>> programmed to display red is still red—we wouldnt even bother to call it a >>> simulation because all icons of red share the same property. Likewise a >>> computer programmed to play chess is genuinely capable of playing chess— >>> because chess is defined by formal rules that you can simulate without >>> losing anything important about chess. A simulation of a tornado is, >>> however, just a simulation because an actual tornado is categorically >>> something that has a causal dynamical relationship with the physical world. >>> I think we are grappling with the difficulty that language is both a chess >>> game and a tornado. LLMs can simulate common language games (regarding >>> syntactical formal reorganization of tokens while observing formal >>> geometric approximations of semantic relations encoded in high dimensional >>> embeddings) but they don’t simulate the dynamic coupling language has with >>> the physical world (re indices, and pragmatics). Note: Reinforcement >>> learning with Human Feedback (RLHF) only partially captures some of this by >>> aligning LLM responses with human preferences. Furthermore, models lack the >>> internal dynamics necessary to grow their representative capacity in a way >>> that is dynamically and continuously coupled with the real world because 1) >>> models are effete products of training procedures, 2) the training >>> procedures themselves are based on minimizing predictive loss and lack the >>> “learning how to learn” dynamic that seems to be a necessary element of a >>> genuinely triadic/argumentative growing symbol and not just a formal/dyadic >>> representation of one. >>> >>> Apologies for the long and unwieldy sentences & typos. This was not run >>> through an LLM to clean up/reorganize my sentences. >>> >>> Best, >>> Ulysses Pascal >>> Large Language Lab >>> Digital Humanities, UCLA >>> >>> On Sun, 5 Jul 2026 at 5:46 am, Jeffrey Brian Downard < >>> [email protected]> wrote: >>> >>>> Jon S, Gary R, List, >>>> I have found this exchange very helpful, especially because it moves >>>> the question away from the familiar and often unilluminating question, “Is >>>> AI conscious?” and toward the more Peircean question concerning the >>>> relation among semiosis, intelligence, inquiry, and the growth of habit. I >>>> want to enter the discussion by approaching the issue from a slightly >>>> different angle: Peirce’s classification of genuine triadic relations, >>>> especially in “The Logic of Mathematics: An Attempt to Develop My >>>> Categories from Within,” and the relation of that classification to his >>>> broader account of the law of mind. >>>> In that essay, Peirce gives a remarkably important account of triadic >>>> relations. He distinguishes cases in which a general law governs particular >>>> facts from the more developed cases of representation, in which one thing >>>> stands for another to a possible interpretant. A natural law is already >>>> triadic in an important sense: it is not merely one brute fact striking >>>> another brute fact. Rather, it is a general rule under which particular >>>> facts are brought. A law does not merely push; it governs. It mediates >>>> between a general possibility and its particular actualization. >>>> But Peirce also distinguishes this from the more thoroughly genuine >>>> triadic relations involved in signs, representations, representamens, >>>> interpretants, arguments, and symbols. A law governs particular facts. A >>>> sign, especially a symbol, does something more complex: it can represent >>>> not only things, but also other signs, other symbols, other rules, and >>>> other possible habits of interpretation. This is crucial. Symbols do not >>>> merely fall under general laws; they can represent general laws. They can >>>> also modify, criticize, extend, and reorganize the very rules by which >>>> other symbols are interpreted. >>>> This seems to me one of the decisive points for thinking about AI in >>>> Peircean terms. A symbol is not merely a token that stands for an object. >>>> It is a general sign whose interpretive power depends upon habit. Moreover, >>>> symbols can take other symbols as their objects. An argument, for example, >>>> can take a prior assertion, theory, hypothesis, or argument as its object. >>>> It can then articulate premises concerning that object, draw relations >>>> among those premises, and generate a conclusion that is itself a further >>>> symbol. That conclusion may then become the object of further reasoning. In >>>> that way, symbols participate in the growth of other symbols. They do not >>>> merely instantiate generality; they mediate the growth of generality. >>>> That is what I am doing in this very post. I am taking Jon’s and Gary’s >>>> posts about AI, semiosis, and intelligence as objects of reflection. I am >>>> then articulating premises about Peirce’s theory of signs, triadic >>>> relations, symbols, interpretants, and the law of mind. From those >>>> premises, I am drawing conclusions that I take to be more or less valid, or >>>> at least worth submitting to the community of inquiry for criticism. My >>>> response is therefore itself a sign generated in response to signs. It is >>>> an interpretant that may become a further sign in the continuing inquiry. >>>> This is why I think the distinction between a general law and a symbol >>>> matters so much. A general law governs particular facts. But a symbol can >>>> represent and govern the interpretation of other general rules. A law may >>>> explain why particular things happen as they do. A symbol can take that law >>>> as an object, compare it with other laws, ask whether it has exceptions, >>>> reformulate it, or embed it in a wider system of signs. In this respect, >>>> symbols are not merely governed by habits; they are instruments through >>>> which habits grow, become explicit, criticize themselves, and pass into >>>> more general forms. >>>> Now consider the difference between a desktop computer running a fixed >>>> program and a contemporary AI system. I am currently using MS Word to type >>>> and correct this message. Word is obviously rule-governed. It applies rules >>>> concerning spelling, formatting, grammar suggestions, storage, display, and >>>> so forth. In a loose but useful sense, the laptop and the software are >>>> extensions of my thought. They assist me in articulating, correcting, >>>> preserving, and transmitting signs. But MS Word does not itself >>>> substantially transform the interpretive space of the inquiry. It does not >>>> usually propose new premises, discover hidden tensions, reformulate my >>>> argument in a more powerful way, or generate a novel hypothesis about the >>>> relation between Peirce’s semiotic and the question of AI. >>>> An LLM is different. I do not mean that an LLM is therefore a person, a >>>> rational agent, a moral subject, or an autonomous inquirer. But it is not >>>> merely a passive writing instrument in the same sense as MS Word. Its >>>> internal organization is the result of training on vast fields of signs. It >>>> operates by transforming signs in relation to other signs. It can generate >>>> candidate interpretants: summaries, distinctions, hypotheses, objections, >>>> analogies, reformulations, and conclusions. When embedded in larger >>>> systems, it may also be connected with perception, memory, tool use, >>>> robotic action, experimental feedback, or other forms of correction. Even >>>> where the weights of a given deployed model are not changing during a >>>> particular exchange, the broader AI system belongs to a technological and >>>> social process in which rules for generating signs are trained, tested, >>>> revised, fine-tuned, evaluated, and reorganized. >>>> That seems to me a real difference. With a conventional app, we have a >>>> relatively fixed set of rules that facilitate a human user’s sign activity. >>>> With LLMs and related AI systems, we have rules governing the generation, >>>> transformation, selection, and revision of signs; and, at least at the >>>> level of the broader training and deployment process, we also have rules >>>> governing the modification of those rules. This does not make the machine a >>>> human mind. But it does bring us much closer to the Peircean territory of >>>> habit-taking, rule-growth, and the law of mind. >>>> Jon’s objection, as I understand it, is that the operations inside a >>>> digital computer remain discrete and dyadic. They are not continuous and >>>> genuinely triadic in the sense required for intelligence. I think this is >>>> an important objection, and I do not want to dismiss it too quickly. If one >>>> identifies intelligence with autonomous self-controlled inquiry, with the >>>> capacity for genuine doubt, with voluntary collaboration, and with >>>> means-ends reasoning in the fullest sense, then present AI systems clearly >>>> fall short. They do not possess the kind of self-conscious critical control >>>> over inference that a mature human inquirer may possess. They do not suffer >>>> the irritation of doubt in the same way. They do not initiate inquiry from >>>> within a lived horizon of purpose, vulnerability, error, and correction. >>>> They do not have moral responsibility for what they say. >>>> But I am not convinced that this settles the semiotic question. For >>>> Peirce, the reality of Thirdness is not always located where a reductionist >>>> analysis of physical operations would place it. A printed syllogism >>>> consists, physically, of marks on a page. Those marks may be described >>>> dyadically in terms of ink, paper, shape, and causal production. But the >>>> argument as argument is not exhausted by that dyadic description. It is a >>>> triadic sign relation because it stands for an object to an interpretant >>>> under a general habit of interpretation. Likewise, the fact that the >>>> operations of a digital machine may be physically implemented through >>>> discrete state transitions does not by itself prove that the machine cannot >>>> participate in triadic semiosis at the level of sign activity, inquiry, and >>>> interpretation. >>>> The question, then, is where we locate the interpretant. Here I think >>>> Gary’s point is especially important. The interpretant need not be >>>> identified simply with something “inside the machine.” Nor, for that >>>> matter, is the interpretant in ordinary human communication simply a >>>> private mental episode inside an individual skull. Peirce’s semiotic >>>> repeatedly pushes us beyond that kind of individualism. Interpretation >>>> unfolds in conduct, in further signs, in habits, and in the community of >>>> inquiry. The meaning of a sign is not exhausted by its efficient cause. It >>>> is found in its possible and actual interpretive consequences. >>>> Suppose an AI system identifies an unnoticed inconsistency in an >>>> argument, proposes a plausible hypothesis, distinguishes two senses of a >>>> term that had been conflated, translates a vague intuition into a clearer >>>> formulation, or draws attention to a relation among texts that the human >>>> participants had not seen. In such a case, I do not think the crucial >>>> question is whether the machine inwardly “understands” the sign as a human >>>> person does. The crucial Peircean question is whether the generated sign >>>> functions as an interpretant within an ongoing inquiry. Does it make a >>>> difference to the growth of signs? Does it become something that can be >>>> criticized, accepted, rejected, revised, or generalized? Does it contribute >>>> to the formation or transformation of habits of interpretation? >>>> If the answer is yes, then it seems too strong to say that the AI >>>> merely simulates semiosis or merely simulates collaboration. It may >>>> simulate human consciousness. It may simulate autonomous agency. It may >>>> simulate moral responsibility. But the signs it generates can become real >>>> signs in a real inquiry. Their origin in a machine does not prevent them >>>> from entering the communal process of interpretation, just as the origin of >>>> a diagram, printed formula, or mathematical notation in a physical artifact >>>> does not prevent it from functioning as a genuine sign. >>>> This is where Peirce’s doctrine of quasi-mind may be useful. A sign >>>> process need not be a fully personal mind in order to exhibit mind-like or >>>> quasi-mental structure. Communication itself, as Peirce suggests, involves >>>> a commens or common mind, a shared interpretive space in which utterer and >>>> interpreter are not isolated monads but participants in an overlapping >>>> semiotic process. Scientific inquiry is even more obviously not reducible >>>> to the private consciousness of any individual scientist. It is a communal, >>>> historical, self-corrective process in which signs generate further signs, >>>> hypotheses are tested, habits are modified, and generality grows. >>>> AI systems may therefore be best understood not as independent fellow >>>> inquirers, but as semiotic organs or quasi-minds within a larger process of >>>> inquiry. They are not merely like hammers or pencils, though they are >>>> tools. They are not persons, though they can generate person-like >>>> discourse. They are not autonomous members of the community of inquiry in >>>> the full normative sense, though their outputs may enter that community and >>>> alter its course. Their status is intermediate, and that intermediate >>>> status is precisely what makes them philosophically interesting. >>>> Peirce’s law of mind, as I understand it, concerns the tendency of >>>> ideas to spread, connect, generalize, and form habits. Mind is not, for >>>> Peirce, a sealed Cartesian container. It is continuous, relational, and >>>> developmental. Ideas grow by association, by generalization, by the >>>> formation of habits, and by the tendency of signs to generate further >>>> signs. The law of mind is therefore not simply a law of private >>>> consciousness. It is a law of the growth of intelligibility, the growth of >>>> habit, and the growth of reason. >>>> From that perspective, the relevant question is not simply whether an >>>> AI system has consciousness, but whether it participates in processes by >>>> which signs generate interpretants and habits are transformed. I am >>>> inclined to say that it does, though derivatively, dependently, and >>>> incompletely. Its purposiveness is not autonomous in the way human >>>> purposiveness can be. Its final causation, if we may use that language, is >>>> borrowed, scaffolded, and embedded within human purposes, institutional >>>> designs, training regimes, and ongoing inquiries. But borrowed or >>>> scaffolded finality is not nothing. A scientific instrument also embodies >>>> purposes it did not originate. So, too, for much human thought and action. >>>> Children are taught how to speak and how to act from a very young age--and >>>> the training is a process of habituation. A diagram guides reasoning >>>> without being conscious. A mathematical notation system reorganizes inquiry >>>> without being a person. The question is whether AI belongs merely with such >>>> instruments, or whether it represents a more complex case in which the >>>> instrument itself generates candidate interpretants that can redirect the >>>> inquiry. >>>> I think the latter is closer to the truth. AI systems are not >>>> autonomous rational agents. They are not morally responsible collaborators. >>>> They do not yet exhibit the full form of self-controlled reasoning that >>>> Peirce associates with the highest grades of rationality. But they do >>>> appear to participate in thoroughly genuine triadic relations when they >>>> generate signs that take other signs as objects and produce further signs >>>> capable of functioning as interpretants. In that respect, they participate >>>> in the growth of signs. They are not merely governed by rules; they are >>>> parts of systems in which rules for sign transformation can themselves be >>>> trained, evaluated, and revised. >>>> This is also why I hesitate to draw too sharp a boundary between “tool” >>>> and “participant.” The distinction is real, but it may be a distinction of >>>> degree and role rather than an absolute ontological divide. A hammer is a >>>> tool. MS Word is a more sophisticated sign tool. A search engine is a still >>>> more complex sign tool. An LLM embedded in a multimodal, memory-bearing, >>>> tool-using, feedback-sensitive system is more complex again. At some point, >>>> the tool becomes part of the semiotic architecture of inquiry in a way that >>>> deserves a richer description than “passive artifact,” even if it still >>>> falls short of “autonomous inquirer.” >>>> So I would put the point this way. Jon is right that intelligence in >>>> the highest Peircean sense requires genuine Thirdness, self-control, final >>>> causation, and the growth of habits under the pressure of experience. Gary >>>> is right, I think, that the interpretants generated in AI-assisted inquiry >>>> need not be located simply inside the machine and need not be dismissed >>>> because the machine lacks human consciousness. My suggestion is that >>>> Peirce’s distinction between general law, which exhibit genuine thirdness, >>>> and thoroughly genuine symbolic relations gives us a way to say both things >>>> at once. >>>> A general law governs particular facts. A symbol can represent and >>>> govern other symbols and other general rules. Human reasoning is a >>>> paradigmatic case of this symbolic growth. AI systems, especially LLMs, now >>>> participate in this symbolic growth in a derivative but real way. They take >>>> signs as objects, generate further signs, and produce candidate >>>> interpretants that can enter the communal process of criticism and >>>> revision. They do not thereby become persons. But neither are they merely >>>> fixed programs in the old sense. They belong to the expanding ecology of >>>> quasi-minds through which signs grow, habits are reorganized, and inquiry >>>> continues. >>>> In an important sense, we see that many of these attributes are not >>>> confined to AI systems. Anywhere the law of mind governs the growth of >>>> other habits and law we will find thoroughly genuine triadic relations. >>>> That process has been unfolding since the origins of the cosmos as the laws >>>> of nature have evolved in accord with a law of mind that is itself growing >>>> and evolving. >>>> That, at least, is why I think Peirce’s law of mind remains so relevant >>>> here. The law of mind is not merely about what happens inside individual >>>> consciousness. It is about the growth of signs, habits, and generality. >>>> Best, >>>> Jeff >>>> ------------------------------ >>>> *From:* [email protected] <[email protected]> on >>>> behalf of Gary Richmond <[email protected]> >>>> *Sent:* Saturday, July 4, 2026 11:41 PM >>>> *To:* [email protected] <[email protected]>; Jon Alan Schmidt < >>>> [email protected]> >>>> *Subject:* Re: [PEIRCE-L] A Peircean Argument for the Reality of AI >>>> Intelligence >>>> >>>> Jon, List, >>>> >>>> If I understand you correctly, the decisive question isn't whether AI >>>> can participate in semiosis -- I think we agree that it does (but correct >>>> me if I'm mistaken). Rather, it is whether genuine interpretants >>>> require the kind of continuity that Peirce associates with 3ns, or >>>> whether a system whose internal operations are dyadic can nevertheless >>>> contribute genuine interpretants to an ongoing inquiry. >>>> >>>> So the problem now would seem to be where we should locate the >>>> interpretant. As mentioned in my previous post, I do not think it should >>>> be identified >>>> with the internal operations of the computer. Instead, I see *the * >>>> *interpretant** as emerging within the inquiry itself in which both >>>> human participants and AI-generated responses become signs*, the >>>> relevant semiosis within the inquiry not being determined nor exhausted by >>>> what happens 'inside the machine'. >>>> >>>> Again, thanks for your incisive objections which are helping to clarify >>>> the matter for me, especially in the sense that the question seems no >>>> longer simply to be whether AI is intelligent, but *whether >>>> intelligence is located primarily within an individual sign user or within >>>> the broader process of inquiry* which, for Peirce, is essentially >>>> general and communal, a process in which interpretants are continually >>>> generated, criticized, developed, etc. >>>> >>>> Best, >>>> >>>> Gary R. >>>> >>> _ _ _ _ _ _ _ _ _ _ > ► PEIRCE-L subscribers: Click on "Reply List" or "Reply All" to REPLY ON > PEIRCE-L to this message. PEIRCE-L posts should go to [email protected] > . > ► <a href="mailto:[email protected]";>UNSUBSCRIBE FROM > PEIRCE-L</a> . But, if your subscribed email account is not your default > email account, then go to > https://list.iu.edu/sympa/signoff/peirce-l . > ► PEIRCE-L is owned by THE PEIRCE GROUP; moderated by Gary Richmond; and > co-managed by him and Ben Udell.
_ _ _ _ _ _ _ _ _ _ ► PEIRCE-L subscribers: Click on "Reply List" or "Reply All" to REPLY ON PEIRCE-L to this message. PEIRCE-L posts should go to [email protected] . ► <a href="mailto:[email protected]";>UNSUBSCRIBE FROM PEIRCE-L</a> . But, if your subscribed email account is not your default email account, then go to https://list.iu.edu/sympa/signoff/peirce-l . ► PEIRCE-L is owned by THE PEIRCE GROUP; moderated by Gary Richmond; and co-managed by him and Ben Udell.
