Search
MapFull

Play as a Cognitive Primitive

18 Jan 2021 06:37 - 07 Jun 2021 01:04

    • Epistemic Status

      • This was a presentation for Refactor Camp in 2015, somewhat polished up. The highlight is definitely this video of my dog playing with her best friend on the beach:
    • Play as a Cognitive Primitive

      • Background: back in my grad school days, I gave a lot of thought to questions about the nature of mental representation, influenced in large part by the work of Phil Agre and David Chapman, who had done some stellar work in picking apart some previously unexamined concepts and revealing their histories and problems. For my part, I wrote a paper and started a collective project aimed (loosely speaking) at rethinking representation – in the first case, asking what representation would be like if we assumed that emotion was built into the way they are used at a very basic level; and in the second, trying to take narrative or stories as the proper cognitive primitives of minds, rather than statements of logic. Neither of these ideas were terribly original, but they seemed neglected, and still do. My past efforts to explore them today strike me as kind of amateurish and embarrasing. But a blog is the ideal place for amateurish speculation – and another roughly similar approach suggested itself to me the other day.]
      • You wouldn՚t think that dogs would have a very developed sense of irony. They do, however, clearly have the concept of play, which strikes me as a very similar sort of thing, quite possibly involving the same cognitive mechanism. Both involve modulating ordinary actions and attitudes, somehow framing them in a way that serves to limit their energies. Your hipster ironist chooses a style of dress and behavior that is interpretable on several levels, some of which contradict or comment on another. Dogs at play act like they are fighting, yet manage to not commit serious injury; somehow, they communicate to each other their lack of serious intent to harm even as they attack. In both cases, ordinary actions are done, as it were, in quotation marks. The ironist may disturb our sense that things should act straightforwardly, but if even dogs can hedge their bets, maybe some facility like that is more basic to cognition than we generally think.
      • Play has always been something of a mystery to biologists, because it serves no obvious survival function. Oh, sure, people make up just-so stories that suggest it has utility in terms of learning or socialization, but play resists this sort of facile explanation. One of the essential (definitional) features of play is being decoupled from serious outcomes, and the only things natural selection can explain are those that have life and death outcomes.
      • Perhaps the mystery of play and the mystery of representation are the same thing.
      • Let՚s imagine a robot, perhaps one smart enough to do tasks around the house. For the sake of illustration, let՚s imagine it works as a GOFAI-person 40 years ago would have conceived it – contained within its computer brain is a model of the world to some level of detail, and it has the ability to use this model to generate representations of future states of the world, including those that are the results of its own actions. In order to figure out what to do next, it uses this machinery to generate a variety of possible futures based on particular courses of action it can take, and it picks the one with the highest utility.
      • But unlike Dr. GOFAI, we don՚t imagine our robot has a sophisticated symbol-processing system inside, which lets it spin whatever elaborate representations it wants. Instead think of an insect, or something evolved just a bit further than an insect. Insects have machinery for reacting to their environment, but not so much for representing it. So instead of assuming representation as in GOFAI, the exercise is to imagine ourselves in the position of natural selection, and ask how the first basic things like representaiton might evolve? What is the minimal sort of mechanism you could add to a reactive creature to give it the ability to envision the future?
      • Obviously we don՚t know, but my best guess is that the simplest way to get representation, is as a kind of  mimesis . Our mythical creature doesn't so much represent future world states as experience them. It doesn't think “about” them; it has no mechanism for that. It doesn't manipulate a set of tokenized representations, it re-activates past experiential states in such a way as to feel what they would be like if they were realized. From there, behaviors that are aimed at achieving pleasant future states or avoiding unpleasant ones can be triggered accordingly.
      • Is a form of quotation a cognitive primitive? In a sense it must be, because any kind of mental representation has to recall a past state of affairs but not entirely. (Note: this is almost exactly Marvin Minsky's K-lines).
      • Somehow you get the ability to recreate mental states (eg, of having just ate) without having to actually be in the corresponding physical state. My conjecture is that whatever primitive mechanism enables this is also at the root of play and that explains why play seems so basic and yet hard to describe – it is at the root of description itself.
      • In this view, play and the ability to create simple representations of the world are very closely linked. They both involve a decoupling, the ability to put a kind of quotation mark around basic sensation. This quotation ability is not some fancy high-level metacognition, it's more basic than that, a fundamental mechanism of thought.
      • Of course in humans the ability to experience imaginary states is orders of magnitude more powerful and sophisticated. We do have the ability to create symbolic representation, and we can pretty much completely decouple them from lived experience (or think that we can). We are hallucinating imaginary states of being all the time, it's a fundamental part of our nature.
    • Sketch of Refactor Camp talk

      • I՚m going to talk about play, but not about any of its ordinary manifestations, in games or politics or culture or whatnot. There is a very interesting literature on that, from Homo Ludens to Richard Schecter՚s approach to theater and ritual. (yes that sounds pretentious – but I did read it!). Homo Ludens is quite remarkable in how, despite being a heavey and erudite tome of classical scholarship, it manages to embody some of the spirit of play it is trying to describe.
      • But I՚m a geek not a humanist, so I like to think about what play might have to say about the architecture of cognitive machinery. In particular, the fact that play is manifestly present in non-human animals is intriguing, because it suggests that if it does play a role in cognition, it is quite low in the hardware stack – prior to language, obviously, possibly prior to the evolution of mammals.
      • Play as a cultural primitive

        • Huizinga, Homo Ludens
          • Play is an essential constituent of culture
        • “civilization arises and unfolds in and as play. ”
        • “Now in myth and ritual the great instinctive forces of civilized life have their origin: law and order, commerce and profit, craft and art, poetry, wisdom, and science. All are rooted in the primaeval soil of play.”
      • Bateson՚s theory of animal play

        • Play involves a kind of framing or quoting of normal behavior
        • Signals on two levels:
          • Level 1: “I՚m fighting!”
          • Metacommunication: “But not really!”
        • “it is evident that a very important stage in this evolution [of play signalling] occurs when the organism gradually ceases to respond quite automatically … and becomes able to recognize the sign as a signal: that is, to recognize that the other individual's and its own signals are only signals, which can be trusted, distrusted, falsified, denied, amplified, corrected, and so forth.”
        • Play involves a kind of quoting of mental states that suggests that it could be at the root of more sophisticated forms of mental representation.
        • Extend Bateson view slightly to self-signalling. And what is thinking but re-activating mental structures, but carefully framed?
      • The machinery of play

        • the ability to put “not really” scare quotes around a mental state
        • seems equally useful for basic intelligent action!
        • lookahead rather than planning.
        • That is, to choose a course of action,
        • in a manner that isn՚t merely reacting to current stimuli
        • You need to envision the future consequences
        • In both play and proto-planning, your mental machinery has to somehow recreate mental states corresponding to situations that are not actually present, and similarly feign mock versions of the actions that would normally respond to them.
      • Failure in dogs
      • Does play explain representation or the inverse? cf Homo Ludens that play is not “in” culture, it is culture, or something like that.
      • Maybe go through my enactive model? Too much for one talk…but it is so simple and obvious.
      • The relationship between state and action – this is the basis of OOP, trying to remodularize or refactor these things. OOP is out of fashion among the computationally trendy – for a variety of reasons, but mostly because the powerful vision of Alan Kay was boiled down to a thin bureaucratic and weak set of practices. Java is an ideal language for work, but it sucks for play – and for learning.
    • Random

      • “God alone is worthy of supreme seriousness, but man is made God’s plaything, and that is the best part of him. Therefore every man and woman should live life accordingly, and play the noblest games and be of another mind from what they are at present. . . .”
        • Excerpt From: J. Huizinga. “Homo Ludens.” iBooks. (quoting Plato, Laws)
      • Richard Schechner, Playing, The Future of Ritual.
        • In the West, play is a rotten category tainted by unreality, inauthenticity, duplicity, make-believe, looseness, fooling around, and inconsequentiality…The reason … is because the multiple realities of playing are situated inside a pyramidical hierarchy of increasing reality leading from unreal make-believe to “just the facts, Ma՚am”.
      • These views provide a reading of ritual as a subjunctive, as the construction of an ‘‘as if’’ world…these texts on ritual assume a world that is fragmented and broken. The subjunctive world of ritual resides in inherent tension with such a broken world, and such a subjunctive world is at least implicitly understood to be limited and temporary. Ritual, then, involves the endless work of building, refining, and rebuilding webs of relationships in an otherwise fragmented world. The work of ritual ceaselessly builds a world that, for brief moments, creates pockets of order, pockets of joy, pockets of inspiration…..Once ritual is viewed in this way…. we come to realize that ritual is something that is happening to some extent all the time, in the most seemingly common, mundane aspects of our lives. Indeed, the ability, really the necessity, of humans to reside simultaneously in multiple worlds, ceaselessly playing upon the boundaries between them, may be universal. It is certainly pervasive in our everyday lives. We therefore explored how ritual in fact shares much with play.
        • – Ritual and Its Consequences, An Essay on the Limits of Sincerity, Adam B. Seligman et al
      • Cued and detached representations in animal cognition
        • Peter Gärdenfors
        • Relevant to Play-as-cog-primitive
      • Labeling something as a cognitive primitive suggests it might also be a computational primitive, that is, the basis of a programming language. But what would an interpreter for a play-based language look like?
        • No idea...except the essence of >play< is:
          • spinning representations of imaginary futures
          • having these resonate emotionally with action, that is, they become objects of desire and motivation
          • Well that just looks like rational action selection, what's so playful about it? I guess it's a refusal to rank these futures by some external metric, it's more they compel action, they are the means by which a mind compels itself into action.
          • Damn cannot quite get at the concept, still. None of this is remotely suitable for publication, I don't even get what I'm getting at.