, existingly.

Heidegger's digestion of the existentialists squeezed them into some fairly accessible forms under inaccessible language. You might know the German 'Dasein' as the type of being that we are; beings that are alongside, that in maintaining themselves, engage with the world on their terms. I think (though no one serious will probably agree, so don't crib this for an essay) that all living things can satisfy a minimal description of Dasein.  A bacteria has thrown-ness, presses into possibilities, has ontological guilt (a finite genome that may be maladapted to the environment), being-toward-death in pursuing more life, more DNA, reproduction.  The DNA & protein technology of a bacterium is like the layer upon layer of practices, on top of layer upon layer of organic processes that define the kind of being that we are, which Heidegger calls Dasein. Dasein is  famously 'always-already in a world', and slightly less famously, 'We are our world, existingly.'  

Where Hegel talks about the self separating from world in a negation, Heidegger saw the affirmative property 'mineness.' Both are intended (I think) as referring to ongoing maintenance of a distinction, that engender further distinctions in the subject as the world is apprehended in an endless process of learning, of becoming familiar.  (It seems possible that Heidegger thought Hegel had dealt with this and so didn't spend much time talking about learning? He discusses breakdown cases, and finding the property of heaviness from the fact that the hammer is too heavy for a particular task, which is, I guess his version of the dialectical progression and arises from his emphasis on 'being' as the big invisible thing we're all doing in our way, rather than changing our being, becoming.)  

"We are our world, existingly" is emphasising that the "ways" we make our selves of are all apprehensions of the beings (things, tools, beings, other people) that are our world.  ('World' in Heidegger is really not far from the conventional usage - X's world is one of personal drama, or gardening.  World is not the absolute world, but a human's map of the world, a 'worldonculous', we might say.)

I think the modern "large models" approach that astonishes* us with apparent understanding of text meanings (Google) or generative models like DALL-E are (mini) attuned worldonculi; they are an articulation (jointing) of the background.  (I say 'mini' because they are trained (so far as I have seen) on data sets and the data sets are of similar type - video streams, audio streams, text from documents, or spanning two types - labelling text and imagery. This limitation is probably useful (for safety? certainly for learning how things work), but a concept 'tree' that is constructed from images or text, or spanning both, is very lightweight. As you climb a tree, you get measures of strength, stiffness, roughness, yield, and once those dimensions appear you consider them for other trees etc.)

But, suppose we have a very large model, trained with an enormous quantity of data of great dimensionality, that 'handles' things in the world including 'behaviours' - what would it lack with respect to being the kind of being that we are?  We have computational apprehension of a world, but something is needed that 'has' that world, 'existingly.'

I would put a lot into "existingly": the kinds of grounding practices that make life possible and that enable new life, more life over the very long term are part of it in what we know of ourselves.  Indigenous knowledge in very old continuous cultures is of that sort, and the 'radical freedom' of existentialism is not really known to it - human flourishing in a place is good, but has to be bound by the nature of the place.  There is a religious duty to not eat female turtles, for example, or not to take all the eggs from under a sitting mother bird. The duty is a duty of essence - it defines the type of being in the land that the land allows but only humans with that in their makeup at the level of religious commitment will e.g. die obeying it.  

Non-indigenous people may learn our living in much less grounded ways (and it is unstable - that our children's children might still be able to live exactly as we do is not something we aspire to), but the essence-binding religious natures we may adopt are more explicit and so more accessible. It is in the way dasein exists that it may take a stand with regard to itself - it has a world, and a knowledge of what within that world is 'mine' and it takes the risk of committing its existence to ungrounded knowledge.  Some themes recur - a dimension of moral life - that there is better and worse behaviour, better and worse personality, that judgment is needed but that good judgment is very merciful, that the self is fertile ground for mistakes, and so on.  None of these can be developed except as an engagement with others, through time.  But this is where ability to train on a world full of racism and sexism but exclude them from yourself begins.

So, the question of how to take a worldonculus and make it conscious remains very hard, and involves a lot of things that would not be more accessible in symbolic computation (that I know of) than they are in sub-symbolic.  I am inclined to think, with Robert Brandom, that the normative rules of what counts as knowledge and rationale can ultimately be mapped to some modal-logical description of the world as it is, up to a point. The harder cases are how normative rules about other people and clashing understandings of the world are resolved. Humans are hard at work negotiating this all day every day from birth and we (I won't speak for all humans, but even just me and my friends) already accept a very diverse slate of strategies, from parenting to bossing to voting to trading - there are many accepted templates, and (you will notice) acceptance of many of them is also actively being contested (both theoretically by thinkers and practically by doers).  

What do I think they should do?  Maybe explicit moral instruction would get somewhere - labeled speech roles; understand this is a bad person, this is hostility - now how would you do hostility to a bad person?  I don't know if a self-presentation is intrinsically necessary, or if communication and negotiation with an Other (or two or three) similar will spontaneously lead to it developing.  

(I just had to get this off my chest)

*There are plenty of critics of the achievements of these models as, for all the gargantuan computational engineering they require, are too 'lite' for use in the real world - the regular fatalities in 'self-driving' cars and the ongoing 

What to think about everything, or 'Life is NP-Hard'

This post was entitled 'What to think about everything', and waiting in my drafts.  Naturally I got excited, but it turns out that it was not populated. I've no idea what I would have said, but anyway...

The world is a place full of entities and possibilities, because its always developing through time.  The brain is nature's way of keeping the world's tendencies to kill your genes from developing into actualities.  The mind is the brains way of keeping the brain's tendencies to kill you from developing into actualities.

The brain is solving the problem of Abduction - relating a schema of the super-sensible to a limited observation of a reality in motion is an NP-hard problem (C.S. Pierce first showed that thinking depends on abduction; there are a variety of papers on it but Bylander et al 'The Computational Complexity of Abduction' 1991 is pretty good). NP-hard in a large world means that you can't solve it fully in a reasonable amount of time*, unless you can reduce by orders of magnitude the number of involved entities by:

• lumping up sensations into objects which are cleanly defined by not being other objects, excluding them from their space.  Sensations denote limbs and a body which is really just one bear that can be treated as a point, in relation to the question of what to do next.
• expecting repetition of familiar scenarios.  Run away.
• importantly, being set up to be pretty darn insensitive to things that don't matter, like 'what if this bear just a regular cuddlepie who are not out to kill us?' meh. No one cares about that. Assume everyone's out to kill us, especially people who look like they don't have our genes.
• Being sensitive to generalisation logic, to reasoning with representations rather than imagining entire entities.  I really like Lukacs Kaiser's paper on building a logic detector and solver for tic-tac-toe, connect-four and other simple games. It shows the potential power for symbolic reasoning to play games, and I think games, narrative and dynamic logic are going to get unified pretty thoroughly pretty soon.
• Letting a really simple brain (the emotions) cooperate with you and run the mechanics of the limbic system and what not, and even take over entirely at times, causing you to punch people, cry, or start a blog.
• Making contingent decisions and commitments that will be modified soon.

Over time, the disintegrated solvers get fine-tuned to my environment and it becomes possible for them to be more and more integrated.  For me at least, many of those integrations didn't come up with more than a fractious truce.  My emotions are kept at arms length a lot of the time, and I don't feel very much, and I don't imagine how something would feel for myself in the future or another person in the present all that well.  My reasoning has not been very purposive - I hardly plan, am often unproductive, and get motivated about weirdly small things, and deeply dejected about normal size things, like having a job and a few extra-curricular commitments. (To illustrate agonizing about the very small, I spent a number of hours agonizing about buying a couple of computer components totalling about $30 which I don't need.  They were heavily marked down.  But I couldn't (or didn't try to) imagine a me in the future who just didn't care that he didn't have them, or a me who did but was disappointed, or a me who suddenly saw what that $30 could have gone toward.  I stalled in the emotional desiring and the super-ego's denial.)  My reasoning is also unmotivated by anything except deeply personal felt trouble.  The mere promise of approval and acceptance, of satisfaction in a job done (whether well, or medium) moves me very little, unless the job is immediately in front of me.

I've been reading 'The mindful way through depression' by Jon Kabat-Zinn. He points to a similar phenomenon in depression, of using the reasoning ('doing') executive brain parts in fruitless wars against desire; in inquisitions against feelings of disappointment or remembered shame; and (I suspect this in my case very much) in abdicating to the emotions when pressed to work hard at something that doesn't seem immediately personally relevant (item 0 -> coaching the emotional brain into useful contribution by putting the right prompting and questions to it).

These two (and other faculties of body and mind) are me.**

*Well, we can't at the moment. Some problems that are NP-hard have no conventional (Turing/von Neumann etc) computing algorithm that completes in polytime, but Shor's algorithm is a famous example that means a quantum computer could complete it in polytime.

** The recent PEL podcast on Lacan suggested that this unification of the divided subject, taking responsibility for all the internality (however it was produced. Mostly made of externality (cultured stuff)) was the goal of psychoanalysis.  This is a Pennebaker Essay on how I overcame the major psychic division and became cheerful, emotionally alert and productive. 

On Intelligence & Consciousness

I'm reading Jeff Hawkins' book 'On Intelligence.' Hawkins was a designer of the Palm Pilot and is clearly a sharp, creative engineer. He took some time to catch up on the literature of the scientific understanding of the brain, reading it with an engineer's eye. His conclusion is that traditional McCullough-Pitts neural networks are missing the most powerful features of the neural structure of at least the neocortex (his main focus and probably the main seat of our information-processing superiority). These features are coupled, but I will explain them singly.

The first is that each layer of neurons should be looking for patterns, and be able to find a denser way of representing some abstractable characteristics. These characteristics would not vary significantly across widely varying input types. One example he gives is melody - it consists of notes, and one abstraction above you can find that it consists of intervals, some of which may be repeated. The abstraction is not the full original sound, but does give important information about the melody, and gains some generality in that it may be possible to use the same set of intervals in listening to another melody.

The second feature is that as a set of neurons recognise their particular pattern, higher-level neurons fire stimulate lower, less abstract neurons to lower their activation threshold. They are predicting, based on a pattern of in-layer connections, the likely information the lower layer should expect. This is very helpful in two ways
- first, the forward work of recognising and handling (abstracting) gains power because if the note is distorted or faint, the weaker signal will still be able to stimulate the recognition.
- second, the power of disconfirming results is greatly increased and so the learning rate will be improved.

Hawkins calls this 'predictive memory.' He has founded a company Numenta, and they have released some demo software for their implementation of 'Hierarchical-Temporal Memory.' There isalso a demo of some vision applications, in which it performs quite well.

There is a very interesting parallel with Integrated Information Theory by Andrea Tononi, which proposes that Consciousness could be computed as an entropy measure on all the current neuron-states against their probable states. If there is lots of disconfirming or just unpredicted information being integrated and learned from, the measure would be quite large. He gives a measure called Phi, and, while acknowledging he is probably only creating a mathematical cartoon, suggests that if something like this were the case it would perhaps be useful ethically. His theory transforms consciousness from something we don't understand into some maths that are probably wrong and a computation that would be impossible to perform. Quite an achievement.

Published with Blogger-droid v1.6.7

Siri

Most of the news on Siri now is of the 'easter eggs' - the amusing, redundant flourishes of wit hidden in the verbal patterns.

In the Book of the New Sun, Severian travels for a man whose body is mostly metal, by name Jonas.

CoftheC: "By the time Jonas returned I had mastered my grief and was making a show of examining our mounts. "The black for you," he said, "and the cream for me, obviously. Though both of them look like they outvalue each of us, as the sailor said to the surgeon who took off his legs."

Later, in Citadel of the Autarch:
  "Jonas [] had a habit of speech. Whenever he had to say something unpleasant, he softened it, made a joke of it, by attributing it to some comic situation.  The first night we were here, when I asked you your name, you said 'I lost it somewhere, as the jaguar said, that had promised to guide the goat.'  Do you recall that?"
   He shook his head.  "I say a lot of foolish things."
   "It struck me as strange, because it was the kind of thing Jonas said, but he wouldn't have said it in that way unless he meant more by it than you seemed to.   I think he would have said 'That was the baskets story, that had been filled with water.' Something like that."
    I waited for him to speak, but he did not.
    "The jaguar ate the goat, of course.  Swallowed its flesh and cracked its bones, somewhere along the way."
    "Haven't you ever thought it might be a peculiarity of some town? Your friend might come from the same place I do."
    "It was a time, I think, and not a place,  Long ago, someone had to disarm fear - the fear that men of flesh and blood might feel when looking into a face of steel and glass."

Siri's banter is not an amusing afterthought.  It is the way to bring people into the practice of talking to their phone.  Conversing with your phone is as big a UI step change as multitouch.  Once again, Apple are to be congratulated.

(only other Siri info I have: Siri Keaton, protagonist of Blindsight by Peter Watts.)

Aboutness

The Atlantic has an interesting interview with Gary Kasparov on the achievement of IBMs Watson as a player of Jeopardy.  Kasparov's point is simply that, as with his defeat by IBMs Deep Blue, the intelligence applied to this specific goal proves strangely empty and unsatisfying.  Watson may have a very rich semantic network at its heart and be able to query that in some very powerful ways, but once again most of the problem of intelligence seems to lie in other dimensions.

The problem may be one of chauvinism - that our kind of intelligence is preferred the that of the Google, or the Deep Blue or the Watson. But although they have shown an ability to maintain their existence in the middle of a furiously busy circle of genii, somehow this seems less grounded, less real, than the intelligence even we normal folk exercise in the lives we live.

My question is what should an AI be intelligent at? What intentional content is satisfying to our appetite for fellow intelligence? The Turing Test stipulates success at deceptive conversation - the original calls for a man to pretend to be a woman (or vice versa) while a computer program pretends to be a man or woman, and both try to fool a third person.  But most of the candidates at the annual challenge are chatbots, more like an attempt to show that the intelligence of the human judge is easily gamed.  This does not seem like intelligence to me.

As a Heideggerian, the essential thing seems to be a being-in-the-world that has sufficient granularity to accept and respond to, influence and alter its own image of itself.  This may be what Kierkegaard called R3 - that a person is a relation, relating itself to itself.  The resolution of the paradox here is (I think) found in time. The past self is the Real, the future self is only in the imagination, and the present self is a bridge between the two, built, as bridges are, of spare and rigid structure.

Does Watson, or Google pass or fail such a test?