Page 188 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

178 CHAPTER 8 The New AI: Basic Concepts, and Urgent Risks

There is reason to expect that orders of magnitude more capability will be
possible here, beyond the GPU generation.
Even beyond that is the kind of capability which a new kind of electronic/
photonic hardware might offer, if we combine the capabilities of neural networks
and of quantum computing in the most powerful way which is technically possible
[34]. But do we really want to do that?
Back at the time of the first rebirth of neural networks, Jasper Lupo, who
presided over the $100 million DARPA research in this area, said at the large meet-
ings: “This is bigger than the atomic bomb.” At the time, this seemed like just
another example of the hype which was confusing people with a kind of misdirected
optimism. But as time goes on, there are more and more hints that he might have
been right after all. As with most dramatic new technologies, it could be used for
good or ill.
Years ago, someone asked me: “Are humans really mature enough to use this
technology for the good? Shouldn’t we wait until they know themselves better?”
But in this case, the serious mathematical principles involved are a key part of
what we need to know in order to know ourselves better. We are walking a tightrope
between risks, on the one hand (as Section 5 will discuss), and the need for humans
as a whole to understand themselves better and achieve the level of integrity (mental
integration) necessary even to survive as a species. There are times when that seems
like an impossible goal, but the experience of the last many years has shown me
clearly that “impossible” problems can at times be solved in the end, if one has
the discipline to really remember the larger goal.

4. NEED FOR NEW DIRECTIONS IN UNDERSTANDING BRAIN
AND MIND
The deep learning revolution in AI and computer science was basically a cultural
revolution. In the understanding of brain intelligence and of consciousness in the
mind, the obstacles to greater progress are also mainly cultural, but more complex,
because many cultures are involved, and because many cultures have a great
variety of misleading conventional wisdoms. As with AI technology, there are
levels and levels of understanding and progress possible. Painful as it is, I will
narrow my scope here (except at the end of this section), and say just a little on
a few simple questions: “Could it be that there is backpropagation in the brain,
and that the brain actually uses the kind of clocks shown in Fig. 8.9 as it learns
to predict its environment better and better over time? Could it even be that the
higher intelligence of the entire brain really is evolved to ‘try’ to maximize its
expected future utility, just as the rational expectations people in economics say,
and falls short only because it takes time to learn and because the optimization
problem is a difficult one requiring levels and levels of approximation? Is life a
game of probabilities in the end?”