Monday, June 16, 2014

The Bullshit at the Heart of Humanity

I explained, in a recent blog post, Why Humans Are So Screwy.   But I didn't quite finish the story there.   Here I'll explain an additional aspect of Screwy Human Nature -- the nature of the self-delusion that lies at the heart of our selves.

My previous post identified two major culprits where human screwiness is concerned:

  • The conflict between the results of individual and group (evolutionary) selection, encoded in our genome (as described excellently by E.O. Wilson)
  • The emergence of civilization, to which we are not adapted, which disrupted the delicate balance via which tribal human mind/society quasi-resolved the above-mentioned conflict (as described excellently by Sigmund Freud)

What I want to point out here is the next chapter in the story -- the way these conflicts  impact our “selves” – our “autobiographical self-models”, which play such a large role in our inner lives.   They play a major role in causing our selves to comprise self-damagingly inaccurate models of the thought, feeling and behavior patterns of which we are actually constituted. 

Put relatively simply: in order to avoid the pain that we are conditioned to feel from violating individual or group needs, or violating civilized or tribal standards of individual/group balance, we habitually create false self-models embodying the delusion that such violations are occurring in our minds and actions much less often than they really are.  Emotional attachment to these sorts of inaccurate self-models is perhaps the most directly important cause of human mental and social suffering.

Our Problematic Selves

The primary culprit of human suffering has often been identified as the “self” – meaning the autobiographical, psychosocial self; the self-model that each of us uses to symbolize, define and model our own behavior.   One of the most commonly cited differences between normal human psychology and the psychology of “enlightened” spiritual gurus is that the latter are said to be unattached to their autobiographical selves – indeed they are sometimes said to have “no self at all.”

I think there is some deep truth to this perspective; but one needs to frame the issues with care.   Any mind concerned with controlling a body that persists through time, has got to maintain some sort of model of that body and the behavior patterns that are associated with it.   Without such a model (which may be represented explicitly or implicitly), the mind could not control the body very intelligently.  Any such model can fairly be called a “self-model.”   In this sense any persistently embodied intelligence is going to have a self.

The problem with the human self, however, is that it tends to be a bad model – not a morally bad model, but an inaccurate one.  The self-models we carry around in our minds, are generally not very accurate models of the actual behavior-patterns that our bodies display, nor of the thought-patterns that our minds contain.   And the inaccuracies involved are not just random errors; they are biased in very particular ways.

Our self-models are symbols for clusters of behavior-patterns that are observed to occur among our bodily behaviors and our internal cognitive behaviors.   This is not in itself bad – symbolic reasoning is critical for general intelligence.   However, we are very easily drawn to make incorrect conclusions regarding our symbolic self-models – and to become emotionally attached to these incorrect conclusions.

And this brings us straight back to the two conflicts that I highlighted in my earlier blog post: Self versus Group (Wilson), and Evolved Self/Group Balance versus Civilized Self/Group Balance (Freud).   These layered contradictions yank our self-models around willy-nilly.   Each modern human feels great pressure to be both self-focused and group-focused; and to balance self and group in a tribal way, and in a civilized way.

What’s the simplest way for a person to fulfill all these contradictory requirements?  -- or rather, to feel like they have at least done a halfway-decent job of fulfilling them?

That’s easy: To bullshit themselves! 

Human self-models are typically packed with lies -- lies to the effect that the person is fulfilling all these contradictory requirements much better than is actually the case.  Because when a person clearly sees just how badly they have been fulfilling these contradictory requirements, they will generaly experience a lot of bad emotion – unless that person has somehow managed to let go of the expectations that evolution and society have packed into their brains and minds.

The above analysis of the conflicts in human nature lets us specifically identify four kinds of lies that are typically packed into human selves.  There are two kinds of Wilsonian lies:
  • Lies about how a person has acted against their own goals and desires
  • Lies about how a person has disappointed the others around them

And there are two kinds of Freudian lies:

  • Lies about how a person has repressed their true desires, in order to adhere to general social expectations
  • Lies about how a person has violated general social expectations, in effort to act out their true desires

What if a person could avoid these four kinds of lies, and openly, transparently acknowledge all these kinds of violations to themselves, on an ongoing basis during life?  This would allow the person in question to form an accurate self-model -- not the usual self-delusional self-model biased by the Wilsonian and Freudian contradictions.   But this sort of internal self-honesty is far from the contemporary human norm. 

The problem is that evolution has wired us to become unhappy when we know we have acted against our own goals and desires; OR when we know we have disappointed someone else.   And civilized society has taught us to become unhappy when we violate social expectations; but evolution has taught us to become unhappy when we don’t balance self and group in the way that is “natural” in a tribal context.    So, inside the privacy of our minds, we are constantly tripping over various evolved or learned triggers for unhappiness.  The easiest way to avoid setting off these triggers is to fool ourselves that we haven’t really committed the “sins” required to activate them – i.e. to create a systematically partially-false self-model. 

The harder way to avoid setting off these triggers is to effectively rewire our mind-brains to NOT be reflexively caused unhappiness when we act against our goals/desires, disappoint others, violate social expectations, or balance self and group in tribally inappropriate ways.  Having done this, the need for an inaccurate, self-deluding self-model disappears.  But performing this kind of rewiring is very difficult for human beings, given the current state of technology.   The only reasonably reliable methods for achieving this kind of rewiring today involve years or decades of concentrated effort via meditation or other similar techniques.

And what would a human mind be like without a dishonesty-infused, systematically inaccurate self-model?  Some hints in this direction may be found in the mind-states of spiritually advanced individuals who have in some sense gone beyond the negative reinforcement triggers mentioned above, and also beyond the traditional feeling of self.   My friend Jeffery Martin's recent study of the psychology of the spiritually advanced  (soon to be published) suggests that, without a self in the traditional sense, a person’s mind feels more like an (ever-shifting) set of clusters of personality/behavior patterns.   One of the lies the self tells itself, it seems, is about its own coherence.  Actually human beings are not nearly as coherent and systematic and unified as their typical self-models claim.

Goals Beyond the Legacy Self

So much for the complexly conflicted present.  Let's think a bit about the possibly better future.

In the current state of human nature, self and goals are intimately wrapped up together.   

Substantially, we pursue our goals because we want our self-model to be a certain way – and we do this in a manner that is inextricably tangled up with the various lies the self-model embodies.

But consider, on the other hand, the case of a post-Singularity human or human-like mind that understands itself far better than contemporary humans, thus arriving at far more accurate – and likely less unified and coherent – self-model than a typical pre-Singularity human mind.   What will the goals of such a mind be?  What will a mind without a coherent self --without a self built around lies and confusions regarding self vs. group and repression and status -- actually want to do with itself?

Considering our primary current examples of minds that have discarded their traditional autobiographical selves -- spiritual gurus and the like – provides confusing guidance.   One notes that (with nontrivial exceptions) the majority of such people are mainly absorbed with enjoying the wonder of being, and sometimes with spreading this wonder to others, rather than with attempting to achieve ambitious real-world goals.   The prototypical spiritually advanced human is not generally concerned with pursuing pragmatic goals, because they are in a sense beyond the typical human motives that cause people to become attached to pursuit of such goals.   This makes one wonder if the legacy self – with all its associated self-deception -- is somehow required in order for humans to work hard toward the achievement of wildly ambitious goals, in the manner for instance of the scientists and entrepreneurs who are currently bringing the Singularity within reach.

But it’s not clear that the contemporary or historical spiritual guru is a good model for a post-Singularity, post-legacy-self human mind.   I suspect that in a community of post-delusory-self minds, avid pragmatic goal-pursuit may well emerge for different reasons, mostly unrelated to legacy human motives. 

Why would a community of post-delusory-self minds pursue goals, if not for the usual human reasons of status and ego?   Here we come to grips with deep philosophical issues.   I would argue that, once the conflicts that wrack human nature are mostly removed, other deep human motives will rise to the fore – for instance, the drive to discover new things, and create new things.   That is: the drives for pattern, creation and information.   

One can view the whole long story of the emergence of life and intelligence on Earth as the manifestation of these “drives”, as embedded in the laws of physics and the nature of complex systems dynamics.   From the point of view of the Cosmos rather than humanity in particular, the drives for pattern, creation and information are even deeper than the conflicts that wrack human nature. 

If spiritually advanced humans, having cast aside self and ego and status, tend not to pursue complex goals of discovery and creation, this may be because, given the constraints of the human brain architecture, merely maintaining a peaceful mindstate without self/ego/status-obsession requires a huge amount of the brain’s energy.   The simple, blissful conscious state of these individuals may be bought at the cost of a great deal of ongoing unconscious neural effort. 

On the other hand, once the legacy human brain architecture becomes flexibly mutable, most of the old constraints no longer apply.   It may become possible to maintain a peaceful, blissful conscious state – relatively free of Freudian repression and individual/group conflicts – while still avidly pursuing the deeper goals of gaining more and more information, and creating more and more structures and patterns in the universe.   Here we are far beyond the domain of the currently scientifically testable – but this is indeed my strong suspicion.

Current human nature got where it is largely via the advent of certain technologies – the technologies of agriculture and construction that enabled civilization, for example.   The folks who invented the plow and the brick weren’t thinking about the consequences their creations would have for the emergence and dynamics of the superego  -- but these consequences were real enough anyway.
Similarly, the next steps in human nature may well emerge as a consequence of technological advancements like brain-computer interfacing and mind uploading – even though the scientists and engineers building these technologies will mostly have other goals in mind, rather than explicitly focusing their work toward reducing conflict in the human psyche and bringing about an era where self is less critical and discovery and creation are the main motivations.

Growth, joy and creation beyond the constrictions of the self-delusory self -- I'm more than ready!

Conceptor Networks

I read today about a new variant of recurrent neural nets called Conceptor Networks, which look pretty. interesting,

In fact this looks kinda like a better-realized variant of the idea of "glocal neural nets" that my colleagues and I experimented with a few years ago.

The basic idea, philosophically (abstracting away loads of important details) is to

  • create a recurrent NN
  • use PCA to classify the states of the NN
  • create explicit nodes or neurons corresponding to these state-categories, and then to imprint these states directly on the dynamics

So there is a loop of "recognizing patterns in the NN and then incorporating these patterns explicitly in the NN dynamics", which is a special case of the process of "a mind identifying patterns in
itself and then embodying those patterns explicitly in itself", which I long ago conjectured to be critical to cognition in general (and which underlies the OpenCog design on a philosophical level...)

There is some hacky Matlab code here implementing the idea; but as code, it's pretty specialized to the exact experiments described in the above technical report...

My intuition is that, for creating a powerful approach to machine perception, a Conceptor Network would fit very well inside a DeSTIN node, for a couple reasons
  1. It has demonstrated ability to infer complex dynamical patterns in time series
  2. It explicitly creates "concept nodes" representing the patterns recognized, which could then be cleanly exported into a symbolic system like OpenCog

Of course, Conceptor Networks are still at the research stage, so getting them to really work inside DeSTIN nodes would require a significant amount of fiddling...

But anyhow it's cool stuff ;)

Monday, June 09, 2014

Review of "More Than Nature Needs" by Derek Bickerton

I've been a fan of Derek Bickerton's writing and thinking on linguistics since happening upon Language and Species in a Philadelphia bookstore, disturbingly many decades ago.   More Than Nature Needs, the latest addition to Bickerton's canon, is an intriguing and worthy one, and IMO is considerably deeper than its predecessor Adam's Tongue.

Adam's Tongue argues that the elements of human symbolic language likely emerged via scavenging behavior, as this was an early case in which early humans would have needed to systematically refer to situations not within the common physical enviroment of the speaker and hearer.  This is an interesting speculation, showcasing Bickerton's inventiveness as a lateral thinker.   MTNN continues in this vein, exploring the ways in which language may have emerged from simplistic proto-language.  However, MTNN draws more extensively on Bickerton's expertise as a linguist, and hence ends up being more profoundly thought-provoking and incisive.

As I see it, the core point of MTNN -- rephrased into my own terminology somewhat -- is that the developmental trajectory from proto-language to fully grammatical, proper language should be viewed as a combination of natural-selection and cultural/psychological self-organization.   To simplify a bit: Natural selection gave humans the core of language, the abstract "universal grammar" (UG) which underlies all human languages and is in some way wired into the brain; whereas cultural/psychological self-organization took us the rest of the way from universal grammar to actual specific languages.

The early stages of the book spend a bunch of time arguing against a purely learning-oriented view of language organization, stressing the case that some sort of innate, evolved universal grammar capability does exist.   But the UG Bickerton favors is a long way from classic-Chomskian Principles and Parameters -- it is more of an abstract set of word-organization patterns, which requires lots of individual and cultural creativity to get turned into a language.

I suspect the view he presents is basically correct.   I am not sure it's quite as novel as the author proposes; a review in Biolinguistics cites some literature where others present similar perspectives.  In a broader sense, the mix of selection-based and self-organization-based ideas reminded me of the good old cognitive science book Rethinking Innateness (and lots of other stuff written in that same vein since).   However, Bickerton presents his ideas far more accessibly and entertainingly than the typical academic paper, and provides interesting stories and specifics going along with the abstractions.

He also bolsters his perspective via relating it to the study of creoles and pidgins, an area in which he has done extensive linguistics research over many decades.  He presents an intriguing argument that children can create a creole (a true language) in a single generation, building on the pidgins used by their parents and the other adults around them.   I can't assess this aspect of his argument carefully, as I'm not much of a creologist (creologian??), but it's fascinating to read.  There is ingenuity in the general approach of investigating creole language formation as a set of examples of recent-past language creation.

The specific linguistics examples in the book are given in a variant of Chomskian linguistics (i.e. generative grammar), in which a deep and surface structure are distinguished, and it's assumed that grammar involves "moving" of words from their positions in the deep structure to their new positions in the surface structure.  Here I tend to differ from Bickerton.  Ray Jackendoff and others have made heroic efforts to modernize generative grammar and connect it with cognitive science and neuroscience, but in the end, I'm still not convinced it's a great paradigm for linguistic analysis.  I much more favor Dick Hudson's Word Grammar approach to grammatical formalization (which will not be surprising to anyone familiar with my work, as Word Grammar's theory of cognitive linguistics is similar to aspects of the OpenCog AGI architecture that I am now helping develop; and Word Grammar is fairly similar to the link grammar that is currently used within OpenCog).

Word Grammar also has a deep vs. surface structure dichotomy - but the deep structure is a sort of semantic graph.  In a Word Grammar version of the core hypothesis of MTNN, the evolved UG would be a semantic graph framework for organizing words and concepts, plus a few basic constraints for linearizing graphs into series of words (e.g. landmark transitivity, for the 3 Word Grammar geeks reading this).   But the lexicon, along with various other particular linearization constraints dealing with odd cases, would emerge culturally and be learned by individuals.

(If I were rich and had more free time, I'd organize some sort of linguistics pow-wow on one of my private islands, and invite Bickerton and Hudson to brainstorm together with me for a few weeks; as I really think Word Grammar would suit Bickerton's psycholinguistic perspective much better than the quasi-Chomskian approach he now favors.)

But anyhow, stepping back from deep-dive scientific quibbles: I think MTNN is very well worth reading for anyone interested in language and its evolution.   Some of the technical bits will be slow going for readers unfamiliar with technical linguistics -- but this is only a small percentage of the book, and most of it reads very smoothly and entertainingly in the classic Derek Bickerton style.   Soo ... highly recommended!