What is the role of biology in AI?

Question

Biology is used in AI terminology. What are the reasons? What does biology have to do with AI? For instance, why is the genetic algorithm used in AI? Does it fully belong to biology?

Answer 1

Biological organisms (such as animals or plants) are the main examples of intelligent systems that we are aware of (excluding artificially intelligent systems, so as not to discuss whether current AI systems are really intelligent or not). Consequently, biological life is often an inspiration for AI researchers to develop AI systems.

There are numerous examples of AI systems that have been introduced (at least, partially) based on or just inspired by the biology. Here are a few examples.

• Reinforcement learning is based on a similar way that animals (such as dogs or pigeons) can learn. For more details, see Sutton & Barto's book (especially chapters 14 and 15).

• Artificial neural networks are very approximative models of human neural networks.

• Genetic algorithms are roughly based on Darwin's theory of evolution.

• Ant colony optimization algorithms (and, in general, swarm intelligence) are based on the way real ants (and, respectively, biological swarms) behave. (There is even a rap song dedicated to ants).

There are probably other examples that don't come to my mind right now. See also this and this questions.

There are cases where AI discoveries have also helped the development of biology or related fields (such as neuroscience and psychology). For instance, Sutton & Barto (on page 4) write

Of all the forms of machine learning, reinforcement learning is the closest to the kind of learning that humans and other animals do, and many of the core algorithms of reinforcement learning were originally inspired by biological learning systems. Reinforcement learning has also given back, both through a psychological model of animal learning that better matches some of the empirical data, and through an influential model of parts of the brain's reward system.

Answer 2

Evolutionary game theory and evolutionary algorithms

I see the connection arising mostly thought Evolutionary Game Theory and Evolutionary Algorithms. Evolutionary algorithms are an analog of natural selection, where successive generations of a given decision making agent are more optimized than previous generations. Like organisms in nature, this process uses "reproduction, mutation, recombination and selection".

There are a couple of recent articles from Quanta Magazine. One, "The Math That Tells Cells What They Are" discusses mathematical optimization as the core function of fundamental biological systems.

"Through evolution, these cells have figured out how to implement Bayes' trick using regulatory DNA."

"Natural selection [seems to be] pushing the system hard enough so that it ... reaches a point where the cells are performing at the limit of what physics allows."

This second quote is exactly the goal of Artificial Intelligence, where utility is limited by physics (computing resources). One way for an algorithm to increase utility is to increase computing power, but the other method is to refine the algorithm to make strong decisions more efficiently. (MCTS vs. Brute Force where a model is intractable, as an example.)

A second article "Mathematical Simplicity May Drive Evolution’s Speed" talks about Genetic Algorithms

"Creationists love to insist that evolution had to assemble upward of 300 amino acids in the right order to create just one medium-size human protein. With 20 possible amino acids to occupy each of those positions, there would seemingly have been more than 20300 possibilities to sift through, a quantity that renders the number of atoms in the observable universe inconsequential."

The game of Go on a 19x19 board has a similar quality--the number of potential gamestates is vastly exceeds the number of atoms in the universe, and, even if the entire universe were converted to computronium, the game would still be intractable.

"The fatal flaw in their argument is that evolution didn’t just test sequences randomly: The process of natural selection winnowed the field. Moreover, it seems likely that nature somehow also found other shortcuts, ways to narrow down the vast space of possibilities to smaller, explorable subsets more likely to yield useful solutions."

This would also be an accurate description the process of pruning a search space. The article concludes that, although there is still much research to be conducted:

“The idea of thinking about life as evolving software is fertile.”

The process of optimization in nature and in computer science is similar in spirit, if not in fact.

Automata as a form of artificial life

The second factor may arise out of the mythology of AI, via speculative fiction. In science fiction, the idea of automata as a form of artificial life is persistent. Shows & films like Westworld, BladeRunner, and the Alien franchise, with David the Android as a prime example of a superior, artificial species that may supplant humanity, are extremely popular. These are all based on Phillip K Dick's ideas explicated in Do Androids Dream of Electric Sheep, the plot of which turns on evolutionary game theory, written about 5 years before the field was formalized! (Dick's influence can even be seen in Google's "Nexus" naming convention for their Android phone;) Underneath all of this is also the idea that Artificial Intelligence itself is a function of nature, with humans as merely the vehicle for the next form of dominant life.