Thursday, September 21, 2017

Trusting Experience or Data: Brains to IT to Souls



Certain principles of mathematics really are universal, and are understood better if one considers multiple examples together. One of the really fundamental issues in all intelligent systems – from brains to souls to AI – is the issue of how much trust to place in the reported experience data from someone else.

The issue of trust is even more complex in political systems, but, since those are not really intelligent systems, I will get to those examples later if at all today.


To see this issue with mathematical clarity, it is important to begin by understanding the basics of trust factor issues in three types of brain – roughly, mouse/rat, monkey and human. The mathematics and neuroscience behind this is summarized/reviewed at www.werbos.com/Mind.htm, which has links to the most recent published papers, such as the open access paper Regular Cycles of Forward and Backward Signal Propagation in Prefrontal Cortex and in Consciousness, by Paul Werbos and Yeshua (J.J.) Davis, Front. Syst. Neurosci., 28 November 2016.


The mouse brain basically learns from a compressed database of all its past experience, its OWN past experience. Mice pay a lot of attention to other mice and other creatures, and yes they do learn in real time as they update their memories, but their learning is based on their own direct experience. The great advance in the monkey brain over the mouse is “mirror neurons,” which allow the monkey to assimilate the experience of OTHER monkeys or humans (or… ?) and learn from it, when they actually see  the other creature. (Think of the old apt phrase “monkey see, monkey do,” and of imitation learning.) Humans take this one step further by communicating experience from one person to another without the receiver having to actually see the original experience on-site.
From a mathematical viewpoint, this is simply a logical progression, the development of new levels of learning system capable of exploiting more and more of the total information available in the environment. But more powerful learning systems using more of the data require some tricky advances in fundamental brain design, relevant to IT design as well. (Please be careful however in implementing things which can be scary powerful and dangerous! I suppose that rats would agree that humans can be very dangerous to them.)

The key challenge in building or understanding a vicarious learning system, whether monkey-level or human level, is WHAT TO TRUST.

A mouse can get away with a simple compressed database of memories – with records roughly made up of (linked sequences) of VERBS (what it was doing) and OBJECTS (the part of the environment it was acting upon). In all cases, the SUBJECT – the actor – is itself, and it does not need to think about its separate individual identity. (Some studies of the psychology of early childhood report a similar lack of personal learned identity.)
But for a vicarious learning system, identifying WHICH SUBJECT is having the experience (playing the role of actor) is extremely important. For a human level system, it is essential to have a kind of trust factor addressing the question of whether the experience as communicated really did happen, with what level of confidence. It is amazing that human brains do all this with such little VISIBLE difference in the basic wiring diagrams of the brain; however, those of us who have worked with complex algorithms know that a few simple innocuous looking feedback terms can yield radical changes in system properties at times, and we already know that humans only have a tiny percent difference in DNA from other primates. Evolution naturally seeks maximum benefit from minimal cost changes.

At this moment, people are preparing to restructure the economies which rule human society by building new transaction systems (and markets) based on ideas like blockchain. The key idea is to have trust factors which are “1” or “0” . You trust it or you don’t. IT people have for many years tried to develop more powerful belief factor or trust factor systems, where trust is not “0” or “1” but a continuous number. Because storage and attention are limited assets in brains, even when they use a continuous trust factor, they still make decisions whether to assimilate a personal or alien experience record AT ALL, with or without caveats; such limits limit our performance, but they are unavoidable facts of life for all finite systems like us.

Trust factors, or their equivalent and their cousins, will be a crucial challenge in developing new types of internet of things (IOT) capable of respecting and serving humanity, and getting us off of the quick fix of turning the whole earth over to one big integrated AI system. A major problem is that no one on earth has a good enough mathematical handle on how to do them right. Luda, who did the official reviews for the Russian AI society, reported lots and lots of work years ago on belief factor systems, none of which worked. They remind me of the neural networks before I discovered backpropagation, none of which really worked, but could similar new mathematical insight fix the situation? Maybe, but it would have to be grounded in deep understanding, just as my work on backpropagation was based on understanding not only of the credit assignment problem but of Freud’s psychodynamics.

One key insight: trust factors cannot be just scalars, as in many IT designs today. For example, in a database to support decision-making, some here would say there should be at least a Republican trust factor and a Democratic trust factor in the system, factors which would not be equal to each other. Trust is a vector, not a scalar. But it is NOT SO SIMPLE, NOT EVEN CLOSE!

Of course, running NSF panels (a crucial part of my own personal database of experience, very rich and very valuable), one can learn how different people can be trusted in different areas, at different levels.

Recent discussions with Vedanta Society people reminded me (and helped make me more conscious) of another important aspect of trust factors in human life:


How could we automate THAT? Actually, maybe that’s a very dangerous question. How could we UNDERSTAND that?

Well, maybe just one comment. In my view, the “gating” phenomena connected people’s minds within the noosphere are really just an extension of the same kind of symmetry principle we see in mirror neurons, which in turn is an extension of gating/symmetry principles present even in the visual system of the fish. (I cite Olshausen’s paper in Arbib’s first brain handbook as a nice cartoon-like explanation of how this works, though I try to be more general and formal in my own papers discussing that topic.)

And of course, my level of interest/acceptance of reports from other people is influenced by measures of plausibility, which in gturn are very much influence by my own past database of first person experience.



That’s already my main message here for now. It was about five minutes in bed this morning. I have joked to Luda: “Is my superior higher intelligence of the early morning due to a kind of cosmic consciousness, or is it just due to being close to you?” Some of both, I suppose.

She immediately asked “so indeed what should **I** make of that March 1967 newspaper experience you keep talking about? Can’t you tell me more of the details?” Well, that was another whole 5 minutes. (Can you imagine the suffering of a one-year old child with little prior experience of life, exposed to such periods of ten minutes? That’s serious, and tricky, but not for today, not even about today.)

Some immediate thoughts –

1.   My present >> 99% belief in paranormal/spiritual phenomena is mainly based on experience AFTER March 1967. The full first person experience was naturally far more compelling than hearing about it would be, but I was such a hard case that all it did was raise me to “50-50”, open mind. Open mind plus intense general curiosity led to many, many later experiences, of varying testability.
2.   I certainly can recall more about the period and the experience – not all, but enough that I feel like a guy on the psychiatrist’s couch who suddenly understands past experience better when reliving it. And of course, I am well aware of the need to resist filling in details (confabulation) as most people often do.

Did I really read the New York Times from cover to cover every morning back in those days? Yes, I am quite certain about that, I remember that from more than just that day, but there is a caveat: I read the whole news section, which was more than 30 pages, including the big editorial page. I did not read EVERY story, but all those which interested me. Other sections I paid much less attention to, except for the magazine section on Sundays and occasional headlines which would catch my eye (more often in business section, never in sports).

I would wake up more slowly then than I do now (when I am retired and utterly unconstrained!). My bed was aligned parallel to the window, parallel to the wall which contained the door.
It was a rectangular room, longer in the direction from door to bed and window than along the door wall. I would drag myself slowly from bed to door, open it, pick up the newspaper, close it, drag myself back to bed, and study the newspaper with a really fierce intensity.

I was INTO the goal of “saving the world” just as intensely then as now, maybe more intense, even though I had not yet understood that it really is a matter of avoiding species extinction. I was in fact a kind of moderate Republican, recently very active in Young Republicans (less so as my work loads grew heavier), intensely scanning the newspaper for understanding of people’s thinking and of opportunities either for me or for society to do better. I had fully understood and implemented the idea of (mundane) “sanity”, as described in www.werbos.com/Mind_in_Time.pdf (published much later in Russia).. and so in studying the newspaper I was “ALL THERE,” verbal mind, mathematical mind, and full nonverbal self all fully engaged and working together, probing the world. I described this to people later as “reading it half asleep, remembering every word”, but in fact I was combing the clarity of intense consciousness and focus with the residues of dream association which I did not focus on but which were part of the me doing the focusing.

And yes, that was every day, BEFORE I assigned any possibility to the kind of experience I had on that startling day in March 1967.

For many years, I have noticed how the gaps between the “ego” (one of Freud’s concepts of ego, the symbolic reasoning explicit self, actually a phenomenon of human culture built UPON evolution of the more fundamental wiring for communicated experience) and the conscious nonverbal life have many consequences. It is possible for people to recall memories of experience which do not really fit what the “ego” would have assigned to them at the time. And we can even hold on to memories we do not entirely trust which do not fit our theories of the time.

And so, this morning, as I relive what my routine was then, I suddenly realize it was NOT so different in its way from my “cosmic consciousness” post-dream routine today!!

In fact, I now believe that CONNECTION is the most important and essential spiritual practice of all. Yes, I now do new things in the morning, like sometimes asking questions about difficult mathematical issues – but then I was really fully immersed in active connection to my childhood version of “the watch.”
(But the New York Times was so much better for that than CNN! It brings home how much has been lost in some ways…)
Luda asks: “Could it be that I just came to understand Mao Tse Tung much better than those guys studying him full time at Harvard, to the point where I could predict the exact changes in thought and words he would use to express those changes, and the exact day of the change?” In truth, my abilities at remote empathy and in reconstructing the thoughts of other people probably were important to my academic success before Harvard. Harvard was not all success for me – but mainly because of my “trespasses,” my clumsiness in stomping on other people’s intellectual turf, not any difficulty in finding it.

So it seems it was like the later day, AFTER I had opened up, when I visited Karl Deutsch in his inner sanctum, heard him briefly express his contempt for all ideas paranormal, and SAW him connect his soul to folks in Europe to ask how they would view a new proposal I suggested for strengthening the European Union. Now I can look back and see MYSELF that way!

It is odd indeed that I totally disbelieved paranormal phenomena then, AND was deeply focused on trying to understand neural networks of brains, but used that morning period for something totally different! Yes, I experimented with my own brain as brutally as Newton did with his eye for the neural network part, but at later times of the day.

All for now. Best of luck…







No comments:

Post a Comment