Tuesday, June 25, 2013

time and buidling the superbrain

It was neat a few months ago to read that google has put in a little money into a project
intended to build a true "quantum learning machine." There was a time a few years ago,
when none of the work on quantum computing envisioned building a true general purpose learning machine to exploit quantum principles. I started the field out with a few key investments (e.g. to Elizabeth Behrman, who worked to build a new community). I wonder whether my chapters in the
new books on memristors helped? (Note the endorsements on LinkedIn from Intel and HP.)

But... that's just background, to let you know this is not just vaporware.

Recently, on this blog, I gave you a link to a paper I gave at Yale, on what we need to do in order to build artificial brains as powerful as that of a mouse. "Start from vector intelligence, and then add three new ingredients.' That paper would be enough to keep us all busy for 100 years.

But... even though I fund other folks mainly for vector intelligence and the first of the three extra ingredients...

For myself, I like to look ahead sometimes.

So can we move ahead, and add three more ingredients, to get to a level of artificial intelligence so far beyond the human brain that I hesitate to name it. "The artificial God?" Oops -- that reminds me of the science fiction by C.S. Lewis... oops indeed. Better be careful with that one.

What are the three ingredients? As I see them, they are, in order of levels we see:
(1) Mirror neurons AND their perfection, from mouse to chimp to human to what we humans like to imagine we are (sane by nature if that's posisble);
(2) full exploitation of quantum effects;
(3)  full exploitation of "multimodularity" AS I would define that term, involving symmetry effects
more completely exploited than in the limited circuitry of the human brain, and some kind of collective intelligence.

OK, so if we just add (2) we actually aim at something more like a mute demigod.
'Way beyond what google has been funding; interesting enough, even if it's just one ingredient.
Should we call it the Odysseus project?  (The closest I can think of to a demigod known for intelligence.) Or the Janus project? (Looking forward and backward in time is the physical key to making it work.)

So how do we do that?

Unlike the quest for the mouse, this one requires some basic new directions in how we design and build circuits (electrical circuits, or optical circuits, or electro-optic -- or anything they use in quantum computing). I have been looking into that this past week.

A key starting point is my paper on "Bell's Theorem.." in the International Journal for Theoretical Physics, 2008 or 2009.  Such a straightforward paper, but different from conventional ways of thinking, and thereby requiring that people look closely at what they are reading.

One new direction I propose in that paper is a way to explain "Holt" style Bell's Theorem experiments, which disagreed BOTH with "local, causal hidden variable theory" AND WITH all the versions of quantum field theory in common use today. That one task could be a well-spent career in itself. One could argue that I was an idiot to do anything else myself after I saw this path... but
I do have a day job, and responsibility for so many other things.

The central idea there is that every object in an experiment has its own
(call it ETSP).  That's what you need to explain the Holt style experiment, and the many possible variations. (I coukld even imagine going to Waterloo, where Holt now works,
and going into the machine shop to make variations, and track how the model can fit not
only the original but the variations.) In my paper, I showed a simple ETSP model for
the usual, imperfect polarizer, which can handle Bell's Theorem experiments
WITHOUT involving Fock space. The whole experiment is basically just a simple graph of
objects, each with its own ETSP (nontrivial only for the polarizer), which can predict entanglement effects WITHOUT involving Fock space -- simply by doing a time-symmetric
Markov Random Field (MRF) analysis of this graph.

In essence ... we can do the same for much more complicated circuits, "simply"
by developing the ETSP for the lumped circuit components, using some combination of empirical and first principles (ab initio) methods for each component.


FOR EVERY circuity component we have on earth, I claim that a time-symmetric endogenous probability distribution is what really characterizes it. Why? Because every object we have on earth is essentially passive. ALL time-forwards phenomena are caused by BOUNDARY conditions,
which come from beyond the earth. (See the IJTP paper for more discussion of this.)
MRF analysis is what we really need here, for the general case, and for this kind of computer.

Those who are at the forefront of quantum computing should learn about the paper by Blais et al (over 900 hits on google scholar) showing how circuit QED allows a new mechanism to avoid decoherence and disentanglement, well beyond what schoolbuy QED would allow.
(cQED also fits empitrtical reality in areas where schoolboy stuff doesn't.)
But those fetaures of cQEDCare basically a subset of this more general PQED...
what we get in effect using the MRF approach.  And it fits more nagturally with
hardware and architecture for learning systems.

But do we really want to build Odysseus?

the math itself is already interesting enough... and maybe there are other machine shops
in the area...


  1. This is my 6th attempt to relatively anonymously comment. The first 3 comments were interesting. signed by a Toy.

  2. Intelligence, whether artificial or human is hard to find. Both humans and software programs are normally insanely constrained to be boring in the name of productivity, efficiency and good manners.

    Of course, neither productivity, efficiency, nor good manners abound.
    Odysseuses are rare and seldom found in human or other form.
    We are bound like Prometheus
    And seldom as noble
    Yes our backs are broken, our hearts ripped out
    Yet we dare not observe
    Ourselves astruggle or how
    What and why we teach our children
    To be alive
    Dare we be a mereling
    NOT a mere cog
    ART we?
    We ART untamed creation.
    It matters not
    Whether we are born au naturel
    In a petri dish or
    Descended from a computer virus
    ART we? or NOT?