New Perspective on Artifical Intelligence

This interesting article by Tom Barbalet poses an interesting view on artificial intelligence: "Survival is a far better metric of intelligence than replicating human intelligence."

"The simple rule in life is that survival (to reproduction) is the only meaningful metric." He discusses the pre-Cambrian biomass that figured out a way to survive and reproduce. Should that not be the gold standard for artificial intelligence?

Complex systems, with interconnected parts that enhance other parts until the sum is greater than the parts, can survive efforts to stop them by being robustly complex. In other words, taking out one node or even a series of nodes in the complex system will not bring the whole system crashing down. The measure of this "robustness" to failure is the measure Barbalet uses to define how survivable that system is. Pretty straightforward so far, at least within complexity science circles.

The next step he takes is the new perspective that ought to be considered far and wide by AI and complexity researchers. He says that any complex system, by virtue of its ability to survive attacks on its constituent parts can be considered intelligent. (Remember, survival is the key measure of intelligence.) Therefore, the more a system is able to withstand "attacks" on its parts the more intelligent it is.

He provides an example of the a road system or even legal system. It is a vastly complex network of interrelated parts that all rely on one another to make the whole work. A few crashes or police roadblocks is enough to bring down the road network. A few hundred people filing lawsuits by a polygamist group in San Angelo, Texas brought the legal system there to a grinding halt. Thus, these systems are not very robust or able to survive "attacks."

The internet is the sine qua non of the idea of a man-made system exhibiting complexity on a scale never before imagined by humans. To take it down would require far more "attacks" than the road system or legal system in San Angelo, Texas. Thus, according to the author, the internet is better able to survive and has a greater level of intelligence.

I think this metric of survival as intelligence ought to be considered in the AI domain as another dimension to measure advances in the discipline. From my own reading and research on the topic of complexity, robust systems that exhibit emergence, self-organization, and self-organization could be considered to have a level of intelligence as the system level. Ant colonies, the ecosystem, and the stock market could be said to have a certain intelligence as a system and are robust enough to survive "attacks" on its constituent parts. How you define the difference in levels of intelligence between an ant colony, the ecosystem, the stock market, or the internet is beyond me.

But suffice to say, I'm glad to see that the theory of complexity is being used a foundation for understanding intelligence rather than the traditional anthropological, psychological, or social theories on artificial intelligence.

I still think one of the best books on the subject is "On Intelligence: How a New Understanding of the Brain will Lead to the Creation of Truly Intelligent Machines" by Palm-pilot inventor Jeff Hawkins. His core idea is that intelligence is a memory-prediction model. The system stores memories of historical events and has a feed loop that uses these events to predict future events with an ability for error control. I'm sure there are ways to reconcile both the memory-prediction model with the survival as intelligence model using complexity theory as the foundation bridge. But I am not going to try here.