From personal stuff to some points about how the brain works
in a very technical way….
This past week was
exciting in many ways. Last Saturday
(December 13) we flew to Chile. We got
back yesterday (Sunday December 21). But
at one level – my life was also about constantly adjusting and using what I had
in my backpack. Despite a whole lot of complexity, sleeping in three different
places and adapting to climates from hot
and dry to freezing cold and wet, and from formal to rough wilderness, I was
able to juggle it perfectly – until the final stage, when I packed my housekey
in a pouch in a suitcase (not in the backpack), and we had an incredible mess
yesterday when the suitcases did not arrive. We are still waiting for them,
from the lost luggage department of American Airlines. An incredible mess.
It certainly was a mistake not to put the pouch in the
backpack, so why didn’t I? Aside from excess trust in American Airlines (and
the baggage handlers who ripped off the colorful cloth we used to identify my
garment bag) – I was overreacting to the mess of having too much in the
backpack many times this week, making it very hard to find things fast enough
for it to be useful to have them. Some things just fell to the bottom, so that
I really didn’t have them when I needed them anyway. In the final packing
stage, I was happy to put so many things into my two small suitcases, and to
make my backpack lighter and more manageable.
But that little pouch with the housekey in it would not have interfered;
it would probably sink to the bottom, and be hard to find – but in an emergency
(like what happened!) we could take the time to find it anyway.
So – life is like a backpack, and I think it’s useful to
make some time to think about that. Some people and some religions would want
to go to extremes – pack everything possible into it, or keep it totally empty,
but the extremes simply are not the right way. (Well, OK, the backpack is empty
right now, sitting next to a small vacuum cleaner so that I can remember to
clean it out. But that’s only for part of the time.) Life, for us, is an
exercise of our mind and our intelligence – and that always means coping with
complicated choices. When we push too
hard towards either extreme – packing in
to much or not packing in key things we need – it doesn’t work out as well, for
WHATEVER values we are pursuing. At one time, I erred by packing too much, and
at the last time, too little – but it wasn’t so bad, even at that one final
time…
Always and forever a balancing act..
But I promised some technical brain science here too.
On Monday in Chile, I gave another talk on how to build and
understand four levels of brain or consciousness – all the way from a low
level, which I call “vector intelligence,” up to level four, the mouse, which
science might build in about a hundred years if it continues on the path it is
on. In 1990, I thought that my new
mathematics of “vector intelligence” could explain how human brains work; I
gave lots of details in my chapters in the book Handbook of Intelligent
Control. I was in fact the first to develop a class of true intelligent system,
able to learn the best possible strategy of action and prediction of life “in
any environment.” But in the next few years, I realized that more powerful
intelligent systems could be built, by exploiting additional mathematical
principles. The next step up was a level I call “spatial intelligence,” brains
which can handle much more complicated streams of in input data. Vector
intelligence can EVENTUALLY get to the right predictions and behavior , but
when life gets complicated, it gets much much slower. I developed new neural
network designs which possess spatial intelligence in a very general way, much
more general than what people have reinvented this year under the name of “deep
learning.” But then I also understood
better ways of organizing experience through time, as we look ahead seconds or
years into the future; I developed and published new mathematics to handle that
case, which I think of now as “reptile level” intelligence. Only in 2009, when
I published a paper in the journal Neural networks, did I understand how to go
to the next stage, the mouse level, by adding a certain kind of creativity
system. (I knew it was needed, but only then did I see the basics of how to do
it.)
So where does the backpack come into it?
The systems with spatial intelligence only are always
“living in the present.” In a way, they are like an empty backpack. Their
feelings of what they like and what scares them ARE based on their experience
of what might come next… but they do not really visualize what might happen
more than a split-second ahead. But at the reptile level, they look ahead. Looking
ahead requires some awkward decisions about HOW FAR to look ahead.
According to my model, we creatures organize time into
things I call “decision blocks,” for which I have published the new
mathematics. But we only get to have a limited list of possible decision
blocks. I think of decision blocks as “verbs,” which may be passive or active.
The active ones are like what people in robotics call “skills” – but they do
not have the full mathematics.
So how many verbs do we learn in a lifetime? Not so many.
Maybe thousands in our whole lifetime, even though our brains have something
like 100 billion neurons. It’s really very awkward, having the ability to learn
only a few thousand verbs. It is very much like a backpack. There is a book on motor control by Vernon Brooks
(not the MIT Brooks) on motor control, which gives a good feeling for what’s
it’s like in the transition period when people learn a new verb.
Folks like Sutton have also talked about systems which learn
“options,” which can be seen as a kind of verb – but not real verbs, since they
don’t have nouns and adverbs to go with them; to add the nouns and adverbs the
right way requires more complete mathematics and approximation methods. Maybe
you could reinvent that mathematics if I say enough about it – but it was
already published, and cited in the Neural Networks paper of 2009.
Meanwhile – in 2012, my newer paper in Neural Networks goes
on to discuss the more advanced progression from mouse brains to monkey brains
to humans to sane humans. And a couple of
my blog postings here go on to discuss the two next fundamental features beyond
that – quantum intelligence and a specific type of multimodular learning
system.
=========================
Many people would think that MEMORY is the aspect of brain like a backpack. Yes,
awkward decisions must be made there too, even at the humble level of vector intelligence --
but for many purposes, the problem with memory is usually more a problem how to FIND what we want. Pribram talks about "holographic memory," where we seem to lose things only because they are buried under other memories which we find more easily. (Well, OK, that was a major issue with the backpack too.) In Neural Networks 2012, I talked about the importance of ACTIVE memory... like deciding actively to remember where we left the car in the parking lot, or like filing important emails in special folders. (Just this past week, I decided NOT to buy a chromebook, because of how important active organization is to my life, awkward or not.)
No comments:
Post a Comment