Sunday, April 12, 2015

Computational Theology

This post revisits a line of thought I first discussed in this article.

The Turing Test for nonhuman intelligence is overly anthropocentric. It privileges the ability to act human as the sole criterion for intelligence. A spacefaring species of aliens would not pass the test, for example. Indeed humans have historically had difficulty recognizing intelligence even in other humans, as the history of contacts between "advanced" and "savage" humans makes clear. (Orson Scott Card made our inability to recognize the "humanity" of intelligent aliens the central focus of the first two books of the Ender series.)

The Turing Test also privileges the human timescale. If an intelligent being were to "think" on a cosmological timescale we would have a hard time holding a conversation with it, even if it was very smart, because we might have to wait a million years to get an answer to a question.

Is it possible to formulate a non-anthropocentric definition of intelligence which captures our intuitions but is not restricted by our prejudices about physical embodiment or spatial or temporal scale? What are some of the characteristics that an intelligence must posess regardless of its embodiment?

One idea is problem solving, which was the focus of a lot of early AI research. Problem solving in turn implies goals, since without goals there are no problems; problems are thwarted goals, requiring action by the agent possessing the goal to unthwart them.

An additional feature of intelligence is learning: it should get better at solving similar problems as it encounters them repeatedly. Learning, in turn, implies memory, i.e., the storage of information in a physical medium, and the accessing of that stored information so as to improve performance on future instances of problems similar to those solved in the past.

Improvements in the ability to retain and deploy information about solutions to previously encountered problems have been critical in the evolution of human intelligence. Such improvements include the evolution, development or invention of learning ability, language, writing, the printing press, and the computer.

Animals clearly have the ability to store information about solutions to problems in their brains, but to a lesser degree. Notably, they have minimal ability to accumulate cultural knowledge across generations.

Are there any other systems that demonstrate any of these abilities? Life includes an information storage medium (DNA) that is accessed by living things to solve problems. It has the "goal" of reproducing itself, and it improves its problem solving ability over time. It thus satisfies the definition of intelligence described above.

The notion of "intelligent life" is reversed from its normal meaning -- not that some specific form of life has crossed the threshold of intelligence, but that life itself satisfies a non-anthropocentric definition of intelligence.

This idea raises what might be called the segmentation problem. If life is intelligent, is it one intelligence, or many? Could we divide life on earth into subsets, each having their own goals, memories, problem solving and learning abilities? Of course: we ourselves are an example; we regard ourselves as distinct intelligences, not as parts of some generalized intelligence of life on earth. We also divide ourselves into superindividuals -- tribes, nations, sects, corporations -- each of which exhibits a degree of intelligence.


No comments:

Post a Comment