Recreational Speculations



From artificial intelligence to the human collective


Recent research on artificial intelligence suggests a new vision of the human collective, interpreted as an intelligent learning system. Accordingly, it would be advantageous for a human society to favor diversity among its members, and the creative exploration of ideas.
Research on learning algorithms explores mathematical and computational principles that could yield to artificial intelligence and explain in part our own intelligence. Recent research on the mathematical difficulty of learning complicated concepts suggests that a single human would have a hard time to learn the kind of complex abstractions we take for granted. Left alone, individual adaptation could settle in sub-optimal solutions known as poor local minima. When stuck in a local minimum, we stick to our current beliefs even if a substantial belief revision could help us get a more predictive model of reality. This research also suggests a well-known strategy to reduce the local minima difficulty: parallel exploration of many different potential solutions, where the best parts of the best solutions can be combined to yield even better ones, as in genetic evolution.
How may these mathematical and computational considerations cast an interesting light about human societies and human intelligence? Genetic evolution, culture and individual learning are adaptive processes working at different time scales, and the source of our intelligence. The brain of a learning human reorganizes itself so as to capture many concepts and abstractions through configurations of neural connections. Each such configuration is a potential solution to the task of creating a 'better' human brain. The adaptive processes that give rise to a particular human brain conspire to explore the space of these neural connections configurations. Here, each human represents a potential solution, a configuration of neural connections. Language would be a mechanism to recombine and propagate - through the exchange of ideas - the good neural configurations discovered along the way. Culture (as the set of all concepts exchanged among humans and passed along to new humans) is a living and growing repository of implicit information. It helps to wire a human brain (not always in the best ways!) through education and immersion in a culture. It allow us to take advantage of the good ideas accumulated by the humans that preceded us.
Connectivity
From artificial intelligence to the human collective
This analysis opens the door to considerations beyond AI research: philosophical, psychological, social and political considerations regarding human societies. It suggests that one can view humanity as a learning system, and that creativity and the diversity of ideas are crucial to efficiently explore the space of ideas, and consequently the space of human brains. In this context, as a society, we should probably invest more and in a more diversified way in the exploration of ideas. Indeed, what could look like a too risky exploratory choice for a single individual would turn out to be a good strategy from the collective point of view: a single crazy but successful idea out of a thousand, moving humanity forward, is enough to justify the apparent waste of time of many marginals, researchers, and other artists. This analysis also suggests the importance of an education that encourages creativity and unorthodox explorations. It could also help to give a meaning to our life, in the context of this immense collective discovery enterprise.

See my 2012 paper entitled Culture vs Local Minima for a more complete treatment of the ideas above. Yoshua Bengio, Sept. 15, 2008