Vinicius, List, I apologize for my delay in responding to your message, Vini, but you’ve given us a great deal to think about.
As I understand it, your paper asks a question of Peircean interest in relation to AI training: *By what method are a machine’s habits fixed?* A neural network is currently trained by repeatedly adjusting its responses in light of examples and errors. Examining its finished answers actually tells us little about *how those habits were formed*. Thus, your idea of 'disciplined contestation' seems especially suggestive. Instead of challenging the AI machine indiscriminately, or asking another machine to supervise it (which is being done), your method challenges a claim when it goes beyond what the available evidence justifies. In your experiments such challenges helped the machine learn the underlying rule more accurately while, conversely, contesting every apparent error drove it toward noise. To me this recalls Peirce’s insistence that genuine doubt is not adopted at will but arises from some *real* disturbance of belief. My principal question concerns what you call the 'factual ceiling', being the limit of what the evidence warrants. In your experimental worlds the *governing law* is known in advance because it has been built into the LLM's synthetic universe; so the network itself must infer that law from the evidence given it. However, in real-world inquiry *determining what the evidence warrants is itself part of the inquiry*. How can the system judge that a claim has exceeded its warrant without relying on an interpretation that may itself require correction? Admittedly, you explicitly acknowledge this problem in t he paper. Notably, your proposal would preserve an auditable history of how the machine was trained: what was claimed, where it was challenged and, subsequently, how its habits changed over time. Such a record would allow a community of inquiry to question not only the machine’s conclusions but also the evidence, the challenges involved, and the standard by which training was stopped. Yet semiostasis -- which I understand to be stability (cessation) within the limits of evidential justification -- does not at all certify that the machine has reached the truth. Rather, it justifies ending training under specified evidential conditions while leaving those conditions open to criticism. As I see it, one of your paper’s most Peircean, not to mention possibly most valuable contributions, is this: The AI machine’s learning history becomes a* public object of inquiry *rather than remaining hidden inside a putatively completed output. Best, Gary R On Wed, Sep 9, 2026 at 10:09 AM Vinicius Romanini <[email protected]> wrote: > Dear friends, > This preprint of a paper I am working on might be of some interest to this > thread. > I would be happy to have feedback. > Vinícius > > > Em dom., 12 de jul. de 2026 às 17:32, Peter Skagestad < > [email protected]> escreveu: > >> Gary, List >> >> Yes, to Doug Engelbart in his pathbreaking 1962 report the locus of >> intelligence was neither the person nor the computer, but the system that >> incorporated both. Later Doug located intelligence in systems incorporating >> multiple persons; at the time of his death he was reportedly working on a >> book titled ‘Boosting Collective IQ’. >> >> In 1993, while stressing the diagrammatic nature of reasoning, I was not >> yet sufficiently sensitive to the dialogical nature of reasoning, something >> Joe Ransdell pointed out in his article. I was still strictly off line in >> 1993; I got my first internet account in the fall of 1994, and promptly >> joined Peirce-L. >> >> While I took a strong interest in AI in the nineteen eighties and >> nineties, I have not kept up with the latest developments and am woefully >> ignorant of the current state of this technology. Were I still teaching I >> would of course have to get up to speed, but I am happily retired. >> >> Best, >> Peter >> >> Sent from my iPhone >> >> On Jul 12, 2026, at 11:32 AM, Gary Richmond <[email protected]> >> wrote: >> >> >> >> Peter, List, >> One of the things that stood out for me in re-reading "Thinking with >> Machines" is that Engelbart understood the computer as but one component >> within a much larger system including various other artifacts, human >> beings, language, methods of inquiry, training, etc.. We've been discussing >> in this and related thread whether one can say that 'thought', in this >> view, is not simply or exactly something that occurs 'inside the head' but, >> rather, is spread across an 'ecology of signs'. >> >> "Thought . . . is more without us than within. It is we that are in >> it, rather than it in any of us." >> (CP 8.256) >> >> The Peircean argument that seemed to support this view is that thought >> proceeds through signs, that reasoning is diagrammatic and, further, that >> inquiry depends upon the observation and criticism of publicly available >> signs. The *continuity of inquiry *therefore extends beyond any >> individual thinker into the larger semeiotic processes through which signs >> continually generate further interpretants. Still, the tentative conclusion >> we may have arrived at in a related thread is, as Jon expressed it, that >> LLMs might at best be considered to be quasi-intelligent, not intelligent >> per se. >> >> As you argue in your paper, the philosophical significance of >> computational technology lies in the possibility of extending the >> conditions under which semiosis itself develops and that Engelbart's >> intelligence augmentation seems to be quite compatible with Peirce's >> conception of *inquiry as the communal growth of meaning. *Of course, >> when your paper appeared in 1993, the kinds of responsive (or, as some say, >> 'conversational') AI systems now available simply did not exist. As I've >> noted, today we have computational systems capable of summarizing >> arguments, comparing interpretations, proposing hypotheses, exposing >> inconsistencies, suggesting analogies, etc. An intriguing question is *how >> these new symbolic technologies alter the ecology of inquiry itself*. That >> is to say that in my view today's AI systems invite us* to reflect >> on Engelbert's idea of intelligence augmentation under fundamentally new >> historical conditions*. If that is so then the question becomes: How may >> advanced computational sign systems participate in the continuing semeiotic >> processes through which human communities generate, criticize, revise, and >> extend inquiries? >> >> Your paper now strikes me as most relevant to the present discussion. >> Rather than asking whether AI is intelligent, it asks us to consider how >> these new technologies might transform the ways in which intelligence >> itself develops within inquiry. >> >> Best, >> >> Gary R >> >> On Thu, Jul 9, 2026 at 3:13 PM Peter Skagestad <[email protected]> >> wrote: >> >>> Thanks Gary, >>> >>> I would also recommend Joe Ransdell’s article’The Relevance of Peircean >>> Semiotics to Computational Intelligence Augmentation’, which discusses and >>> expands upon my article. >>> >>> Best, >>> Peter >>> >>> Sent from my iPhone >>> >>> On Jul 9, 2026, at 2:59 PM, Gary Richmond <[email protected]> >>> wrote: >>> >>> >>> Peter, Jon, List, >>> >>> Peter, thank you for pointing to your paper, "Thinking with Machines: >>> Intelligence Augmentation, Evolutionary Epistemology, and Semiotic," which >>> I remember reading many years ago when Joe Ransdell recommended that I read >>> it and another of your papers, "Peirce, Virtuality, and Semiotic." I found >>> both of them most relevant and valuable at that time. While many of the >>> links on Arisbe are not working, I was able to find "Thinking with >>> Machines" on the Wayback Machine. >>> https://web.archive.org/web/20070522213854/http://www.cspeirce.com/menu/library/aboutcsp/skagesta/thinking.htm >>> >>> >>> Of your three papers on Arisbe, only the link to "Peirce, Virtuality, >>> and Semiotic" is still live. >>> https://www.bu.edu/wcp/Papers/Cogn/CognSkag.htm >>> The third paper, "The Mind's Machines," is unfortunately a dead link on >>> Arisbe (I've asked Ben Udell to write a short note to the List on the >>> problem assessing many papers on Arisbe). >>> >>> I met Doug at the 2001 ICCS conference, the first of many I was to >>> attend, once at lunch meeting which included Aldo de Moor, John Sowa, and >>> Mary Keeler, and another time just preceding John's presenting his session >>> paper when I was able to speak with Doug briefly one-on-one. I must admit >>> that I was a bit intimidated at the time having just entered the world of >>> "conceptual structures" and having heard from Aldo of Engelbart's storied >>> career in developing the computer mouse, interactive computing, >>> hypertext, collaborative knowledge systems, networked collaboratories, and >>> human-computer interfaces that later became foundational technologies for >>> modern AI development and human-AI collaboration. That, I should note, is a >>> partial list. >>> >>> I'd like to quote a section of your conclusion of "Thinking with >>> Machines" to suggest the themes of your paper: >>> >>> >>> >>> >>> >>> *In the above we have examined and documented a coherent set of ideas >>> common to Engelbart's augmentationism, Popper's evolutionary epistemology, >>> and Peirce's pragmatistic semiotic. Let us summarize these ideas in eight >>> theses, not all of them explicitly shared by all three thinkers: 1. Human >>> evolution proceeds exosomatically, through the evolution of external >>> artifacts.2. Human knowledge does not reside in the head, but in a subclass >>> of our evolved artifacts: Engelbart's "external symbols", Popper's "World >>> 3", Peirce's "signs".3. The Whorfian hypothesis: Our fundamental world view >>> is shaped by one of our evolved exosomatic organs, namely our language.4. >>> The Neo-Whorfian hypothesis: Our thinking is also shaped in a variety of >>> ways by other exosomatic organs, specifically the tools by which we >>> manipulate external symbols.(84)5. Abstract, critical thought is made >>> possible only by the existence of written language. This thesis is also >>> embraced by Buehler and Havelock.6. The possible degrees of abstraction (as >>> well as the possible kinds of intellectual projects) available to us depend >>> on the logical or effective features of the symbols by which we represent >>> abstractions -- Arabic numerals, parentheses, and the like -- i.e. not what >>> the symbols look like, but what they can and cannot do.7. The possible >>> degrees of abstraction available to us depend on the physical features of >>> the tools with which we produce and manipulate symbols -- pencils, >>> computers, and the like.8. The digital computer is most fruitfully regarded >>> as a means for augmenting human intelligent activity by automating and >>> accelerating the production and manipulation of symbols, rather than as an >>> independent source of intelligent behavior.* >>> >>> I would encourage all those interested in the topic to read "Thinking >>> with Machines." Thanks again, Peter, for directing us to it. >>> >>> Best, >>> >>> Gary >>> >>> >>> On Thu, Jul 9, 2026 at 8:03 AM Peter Skagestad <[email protected]> >>> wrote: >>> >>>> Gary, list, >>>> >>>> It may be of interest that I argued for the relevance of Peirce’s >>>> semiotic to Engelbart’s intelligence augmentation in my 1993 paper >>>> ‘Thinking with Machines’, available on the Peirce Telecommunity website. >>>> Doug read the paper, we had a couple of conversations about it, and I think >>>> I can claim the credit for Doug’s taking an interest in Peirce in his later >>>> years. >>>> >>>> Best, >>>> Peter >>>> >>>> Sent from my iPhone >>>> >>>> On Jul 9, 2026, at 12:54 AM, Gary Richmond <[email protected]> >>>> wrote: >>>> >>>> >>>> Jon, Ulysses, Jeff, List, >>>> >>>> I agree that "AQI" (or even "AQ-I") is quite unlikely to catch on >>>> any day soon. >>>> >>>> Recently I've been reflecting on Doug Engelbart's project,* Intelligence >>>> Augmentation* (IA), which my friend and colleague, Aldo de Moor, has >>>> done much 'hands-on' work through his work on *CommunitySense *( >>>> https://www.communitysense.nl). Aldo describes his project as >>>> "Building on and generalizing from Engelbart’s views," and summarizes >>>> Engelbart's vision in "Making Doug’s Dream Come True: Collaboratories >>>> in Context" (https://communitysense.nl/papers/iccs02b.pdf) >>>> >>>> In [Engelbart's] view, the only feasible approach [to developing >>>> 'collaboratories' functioning as mediators of distributed groups working >>>> with systems of information tools] is a long term, pragmatically >>>> guided whole-system evolution.* It is crucial not just to create good >>>> socio-technical systems, consisting of co-evolving Human and Tool Systems. >>>> Also, one should continuously improve this design process, in other words >>>> “improve the improvement capability of the organization”*. >>>> >>>> >>>> In rereading Aldo's paper I was reminded that as early as 2000 I had a >>>> keen interest in this kind of research: See a paper Aldo, Mary Keeler, and >>>> I collaborated on, "Towards a Pragmatic Web" ( >>>> https://repository.tilburguniversity.edu/server/api/core/bitstreams/953ea141-d6d4-4cf9-961e-40542bd4d2f3/content >>>> ). >>>> >>>> Recently I've been working on the possibility of building on and >>>> developing Engelbart's and de Moor's work on collaboratories in >>>> consideration of AI, an idea that I've been referring to as IAAI >>>> (Intelligence Augmentation meets Artificial Intelligence). That acronym may >>>> not take hold either, but I think it captures the gist of the idea of a >>>> contemporary approach to how we humans might approach grappling with such >>>> extraordinary LLM developments. >>>> >>>> Best, >>>> >>>> Gary R >>>> >>>> >>>> >>>> xx >>>> >>>> On Wed, Jul 8, 2026 at 12:39 PM Jon Alan Schmidt <[email protected]> >>>> wrote: >>>> >>>>> Gary R., Ulysses, List: >>>>> >>>>> I find your latest formulations considerably less objectionable than >>>>> ascribing genuine intelligence to "AI" systems themselves. They are indeed >>>>> diagrammatic tools that augment *human *intelligence, facilitating >>>>> inquiry when they are intelligently used *by humans* for that >>>>> purpose. Since they at least arguably qualify as quasi-minds, we *might >>>>> *be able to refer to them accurately as *quasi*-intelligent; but even >>>>> if so, I doubt that "AQI" would ever catch on as a replacement for "AI." >>>>> :-) >>>>> >>>>> Thanks, >>>>> >>>>> Jon >>>>> >>>>> On Wed, Jul 8, 2026 at 1:26 AM Gary Richmond <[email protected]> >>>>> wrote: >>>>> >>>>>> Jon, Ulysses, List, >>>>>> >>>>>> Jon, I found your mention of the conference presentation on HAI >>>>>> especially relevant. It immediately brought to mind Douglas Engelbart's >>>>>> research program which he called *intelligence augmentation* (IA) >>>>>> [See his paper (1962), "Augmenting Human Intellect: A Conceptual >>>>>> Framework," https://www.dougengelbart.org/content/view/138] >>>>>> Engelbart's central idea, as I understand it, is that the really >>>>>> important >>>>>> question is not whether machines themselves become intelligent, but how >>>>>> human intelligence may be augmented through interaction with >>>>>> computational >>>>>> systems. >>>>>> >>>>>> In any event, I find myself thinking less about AI considered in >>>>>> isolation than about the larger semiotic processes within which both >>>>>> humans >>>>>> and machines participate. In other words, the most interesting unit of >>>>>> analysis is most definitely *not* the machine itself, but the >>>>>> continuing inquiry in which machine-generated signs aid human thinking >>>>>> about important issues, perhaps eventually even some of our most >>>>>> criticall, world-unsettling 'wicked problems'. >>>>>> >>>>>> Ulysses' description of LLMs as diagrams also seems quite relevant >>>>>> here. A diagram, after all, is not itself an argument or an inquiry (let >>>>>> alone an inquirer). Yet diagrams have long played an indispensable role >>>>>> in >>>>>> mathematical and scientific discovery because, as Peirce >>>>>> definitively showed, especially the 'manipulation' of them permit >>>>>> relations to become visible that might otherwise remain unnoticed. If AI >>>>>> systems are understood somewhat analogously, then perhaps their >>>>>> significance lies *not* in their possessing intelligence than in >>>>>> augmenting the semeiotic processes through which inquiry develops. >>>>>> >>>>>> I wonder whether Engelbart's idea of IA points towards a formulation >>>>>> which might reconcile our differing emphases. The machine need not >>>>>> be regarded as an autonomous intelligent agent, nor merely as a passive >>>>>> artifact. Rather, its significance may lie in the way that it can augment >>>>>> the continuing semiosis through which communities of inquiry generate >>>>>> fruitful concepts, programs, discoveries, etc. >>>>>> >>>>>> Best, >>>>>> >>>>>> Gary R >>>>>> >>>>>> On Tue, Jul 7, 2026 at 11:48 PM Ulysses <[email protected]> wrote: >>>>>> >>>>>>> I am at least commenting on the semiotic and categorical >>>>>>> capabilities of the AI model itself in order to understand what it is >>>>>>> capable of contributing to broader inquiry and what human users need to >>>>>>> bring to the AI model if they are to use them for inquiry. >>>>>>> >>>>>>> My position is large language models are, from a semiotic >>>>>>> perspective, a diagram. Which is to say, potentially very useful for >>>>>>> inquiry but not an artificial “mind” capable of symbolic thought, which >>>>>>> is >>>>>>> necessary, in my view, for intelligence. >>>>>>> >>>>>>> The thing I might disagree with Jon on is I am probably more likely >>>>>>> to consider the training process itself as quasi intelligent, albeit >>>>>>> with a >>>>>>> lot of caveats. >>>>>>> >>>>>>> On Tue, 7 Jul 2026 at 11:27 pm, Jon Alan Schmidt < >>>>>>> [email protected]> wrote: >>>>>>> >>>>>>>> Gary R., List: >>>>>>>> >>>>>>>> As you know, I understand Peirce's late topical conception of >>>>>>>> continuity to be top-down, not bottom-up; the whole is ontologically >>>>>>>> primordial, not an assemblage of discrete parts. From that standpoint, >>>>>>>> the >>>>>>>> notion of *locating *the continuity of semiosis does not really >>>>>>>> make sense--the *entire universe* is a vast semiosic continuum >>>>>>>> from which individual signs, individual sign-users, and larger >>>>>>>> inquiries >>>>>>>> can be prescinded as artifacts of analysis. This obviously does not >>>>>>>> entail >>>>>>>> that the entire universe is properly characterized as *intelligent*; >>>>>>>> on the contrary, it still encompasses discrete things and their brute >>>>>>>> dyadic reactions, including the sequential operations within a digital >>>>>>>> computer, as *degenerate *outcomes of continuous and triadic >>>>>>>> semiosis. >>>>>>>> >>>>>>>> When most people refer to "artificial intelligence," it seems to me >>>>>>>> that they are ascribing intelligence to an individual artificial >>>>>>>> system--exactly what I find problematic--not the larger inquiry that >>>>>>>> results when humans *use *it for that purpose. In fact, one of the >>>>>>>> most interesting presentations at a conference that I attended last >>>>>>>> month >>>>>>>> advocated greater attention to human interaction with "AI" systems, >>>>>>>> which >>>>>>>> effectively turns them into "HAI" systems. Research suggests that >>>>>>>> improvements in that interface might make a bigger positive difference >>>>>>>> going forward than further advancing the capabilities of the technology >>>>>>>> itself. >>>>>>>> >>>>>>>> Regards, >>>>>>>> >>>>>>>> Jon Alan Schmidt - Olathe, Kansas, USA >>>>>>>> Structural Engineer, Synechist Philosopher, Lutheran Christian >>>>>>>> www.LinkedIn.com/in/JonAlanSchmidt / twitter.com/JonAlanSchmidt >>>>>>>> >>>>>>>> On Tue, Jul 7, 2026 at 3:11 PM Gary Richmond < >>>>>>>> [email protected]> wrote: >>>>>>>> >>>>>>>>> Jon, List, >>>>>>>>> >>>>>>>>> I wonder if it is possible that we are now considering somewhat >>>>>>>>> different questions. It seems to me that you are commenting more >>>>>>>>> on *the semiotic character of the AI system* itself, while I >>>>>>>>> increasingly find myself reflecting on *the semeiotic character >>>>>>>>> of the larger inquiry *in which AI seems to me to be capable >>>>>>>>> of participating. Your sticking point as I see it seems to be >>>>>>>>> that the internal operations of present AI systems consist entirely of >>>>>>>>> dyadic reactions and, therefore, cannot participate in the continuous >>>>>>>>> triadic semiosis that Peirce identifies with intelligence. >>>>>>>>> >>>>>>>>> Say, for the sake of argument, that everything you say about the >>>>>>>>> machine's internal operations is correct, still does it necessarily >>>>>>>>> follow >>>>>>>>> that the larger inquiry within which those operations become signs for >>>>>>>>> further interpretation cannot, then, exhibit genuine semiosis? >>>>>>>>> >>>>>>>>> So, I'm beginning to think that our principal disagreement has >>>>>>>>> less to do with AI, but increasingly seems concerned with *where >>>>>>>>> one ought to locate the continuity of semiosis itself*. Is it >>>>>>>>> primarily within the individual sign-user, or may that continuity also >>>>>>>>> characterize the larger inquiry through which interpretants are >>>>>>>>> generated, >>>>>>>>> criticized, corrected, etc.? If the latter, then intelligence may be >>>>>>>>> seen >>>>>>>>> in the living semiosis through which inquiry continually develops. >>>>>>>>> >>>>>>>>> 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. >>>> >>>> _ _ _ _ _ _ _ _ _ _ >> ► 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. > > > > -- > Vinicius Romanini, Ph.D. > Professor of Communication Studies > School of Communications and Arts > University of Sao Paulo, Brazil > www.minutesemeiotic.org > www.semeiosis.com.br > > Skype:vinicius_romanini >
_ _ _ _ _ _ _ _ _ _ ► 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.
