I joined the computer revolution in the 1960’s, installing software, designing databases, and writing programs. This was giving knowledge to machines and teaching them how to do what we wanted them to do. The natural learning and developing intelligence of the two infants at home was even more impressive, and so raised a question: could research into how we build intelligence in machines help up understand human learning? Discussion of related questions with a college classmate, by then a professor, convinced me I needed to study with Marvin Minsky, co-Founder of the MIT Artificial Intelligence Laboratory. That decision, made in 1974, began the the research I have done since.

What then is the main objective of the Infant Peggy Study (IPS)? to gather information about one child and her life over sufficient time to detail and to trace her development and learning through the early years.

Knowing that adequate analysis could not be done “in real time,” Peggy’s Mom and I borrowed video equipment (a Black and White Reel to Reel Portapack). We recorded “weekly” video sessions focused on varied themes and activities, typically 6-8 per half hour tape. The physical corpus includes roughly 300 videotapes, supplemented by 780 parental text notes and vignettes. Digitizing and building the on-line corpus was not easy, nor is bringing it into a usable form: transcription of the video clips (an ongoing component of corpus construction) now requires a workday of the Analyst for each five minute video clip. With augmented transcriptions available, deeper analysis can begin.

Can Intelligent Systems do the analysis for us? I’m not ready to accept that answer for questions that involve the bases of the human mind. My discussion with Marvin Minsky to recruit him as the AI subject matter expert for my doctoral thesis committee raised the essential issues. Here is a synopsis of that discussion.
Context: Seymour Papert, the other co-Founder of the MIT AI lab, had agreed to be my Ph.D. thesis supervisor. Planning was the primary conceptual model of thought in the AI Lab at that time. I had argued successfully with Papert that human thought was different, that what I knew from personal reflection, from observation of children, and from anthropology, was that we needed to explore different way of thinking about thinking. Papert agreed; he had for five years been Jean Piaget’s mathematician at Geneva’s Institute of Genetic Epistemology and saw my proposing the case study method for exploring learning, both in and beyond the lab, as extending Piaget’s best research.
Discussion: When I explained my proposal to Minsky, and argued that however promising, the machine intelligence of that day (mid 1970’s) could not compare with the intelligence seen everyday in the behavior and thinking of people, most especially of children when the focus was on the issue of learning, Marvin said I would not be able to get a degree in Artificial Intelligence. I explained that was not what I wanted. He did agree to my proposal for a negotiated program I designated as “Variform Intelligence,” focused on both human and artificial intelligence; such is my doctoral degree. During that conversation, the primary shared programmatic ground was our common interest in development of a general theory of intelligence involving its function as embodied knowledge and structure, as impacted by the knowledge acquisition (regardless of whether the body learning was a machine or a person).

The current AI success is a quite remarkable triumph of engineering based on research in neurophysiology, computation and fabrication. Even given that efforts have been well funded and networks have opened to machines the Global Knowledge Inheritance, this last fifty year years have produced human level intelligence systems; that seems little short of a miracle, especially if one notices that live, biological systems, required not millions of years, but billions of years to do the same.

The development of mind in individual humans is “little short of a miracle” as well, if we look at the life process: from a single cell, the fertilized egg, to the minds we encounter everywhere, all the time. Human nervous system complexity is more than astronomical, as Maturana summarizes:
“Some 10 exp11 (one hundred billion) interneurons interconnect some 10 exp6 (one million) motorneurons that activate a few thousand muscles, with some 10 exp7 (ten million) sensory cells distributed as receptor surfaces throughout the body. Between motor and sensory neurons lies the brain, like a gigantic mass of interneurons that interconnects them… into an ever changing dynamics,”
Maturana and Varela, p.159, The Tree of Knowledge

How did and do minds grown in this human embodiment ever become able to think, to learn and imagine all the knowledge that creators of AI have made available to machine intelligences? And how is it that humans, with a brain refreshment cycle speed of 3 per second, have created the knowledge that underlies the power of mechanical minds operating so much faster (at speeds measured in “megaflops,” i.e. millions of floating point operations per second). I believe it is a consequence of people thinking using representations of things, processes and relations they have encountered.

Representations
An internal representation is a description used by a mind to selectively refer to discriminated elements of experienced things, relations and processes acting upon or acted on by the human. In short, it’s what we think is happening or what we are doing in the world. Sharable representations are translations of individual’s internal representations for communicating of information among human minds about elements of the world that activate other individiuals’ more or less corresponding internal representations. The long course of earthly life has evolved humans predisposed to use motor and sensory stimuli to form such internal representations and to extend their phylogenetic predispositions with further articulations of experienced aspects of descriptions during the course of new born, of infant, and of childhood development. This experience based, individually embodied ontogenetic development of representations is what we mean by learning.

If we observers try to understand learning by others, it is important to distinguish between the representations we apply in making sense of situations and behavior (which generally settle with maturity into culturally common place representations) and the different representations plausibly inherent in the newborn, those rapidly developing infant and the child, and those sharable representations gradually shaped by the individual’s expanding circle of communication with others as individuals grow and mature. This is the current objective of the Infant Peggy Study.

As Analyst of the infant Peggy’s learning, the component tasks I see now before me are these:

Ascribing representation structures challenges the mature analyst (who has spent a lifetime making distinctions and associations) to imagining different ways of interpreting and coping with the world as it appears to the infant and child. Any analyst would have to consider how a subject, at her state of development, perceives the world, what actions comprise her repertoire of behaviors and what in the situation makes evident her motives. It is not certain that this will be possible. The Infant Peggy Study is an exploration, one which I believe will be illuminating though not a proof validating a theory.

What aspects of IPS, beyond the corpus itself, give hope for progress through the effort? How does one imagine constructing representations which can build out from the infant’s experience and yet are fit to the external world?

My approach expects the representations of the infant to differ significantly from those observers with mature minds. Consider the picture of Peggy’s Standard Objects [linked by “Describing Objects” under Ongoing Work at the sidebar bottom]. Adults can be expected to describe these objects to one another by naming their types, noting differences of shape, noting as well their ability to be stacked on and included in one another. Contrast with that perspective, the descriptions in the table [linked by “Objects and Uses,” which follows “Describing Objects” in Ongoing Work in the sidebar]. If the infant’s actions are primary in her approach to the world, the responses of the objects might well be what she attends to if “making things move” is her intention (as observation suggests). If she intends to grasp or mouth an object, or involves other sensory modalities, the affordances would be different again. Representation development could be followed by tracing Peggy’s attempted uses of objects singly and jointly. I believe it will be fruitful to undertake tracing such developments in the different impacts of Peggy’s experiments, in her witness of use by others (accidental or instructive) and even in the effectiveness over time of language mediated information.

In the IPS world, our favorite creature Peggy is not alone. But the other creatures of her family are as external to her as are all the other physical things of her world. They can act and Peggy can receive sensory stimuli from what they do, but they do not change her representations of the world; that is entirely her prerogative.

Changing her representations of the things of her world begins with her performing whatever actions she chooses from her repertoire upon whatever external element of her world she chooses, then observing what happens next. The action-result sequence has status comparable to the stimulus-response arc. In a creature’s being, it has the advantage of being initiated by and thus controlled by the agent herself.

The representations which Peggy develops and through which she becomes able to think about the objects and events, and even behavior in her world, I suspect are constructed to a significant extent through the everyday communication experiences of living. Consider this a conjecture. The language-engaged aspects of IPS focus early on communication production in situations where the activity is recorded in sufficient context to be reliably interpreted even while pre-verbal (and where the speech production of others is also recorded) with a sufficiency of incidents that permit a time-based line of development to be uncovered.

From the beginning at 18 weeks, the transcriptions are formed as dialogues integrating Peggy’s activities with the speech and action of speakers. The Analyst has made an effort to note Peggy’s vocal expression through all the clips transcribed. Our referencing them is simple in the beginning: her first utterances are coded “NVV” (non verbal vocalization), with parenthetical interpretation when possible. As her utterances become more language like, the are coded as PVV (potential or probably verbal vocalization). These vocalizations developed into sound-sequences that could be characterized as polysyllabic – if they were in recognizable words.These have been coded by quoted aurally recognizable spellings. For anyone interested in checking an utterance interpretation, the recorded sounds can be located through the transcript’s episode time codings.

As Analyst, I see four general categories of work on my near agenda.

That should keep me busy for a while.

Analyst,
Bob
RWLawler@NLCSA.net

Dated: 8/07/2026