Evolution of artificial intelligence

The goal of creating non-biological intelligence has been with us for a long time, predating the nominal 1956 establishment of the field of artificial intelligence by centuries or, under some definitions, even by millennia. For much of this history it was reasonable to recast the goal of “creating” intelligence as that of “designing” intelligence. For example, it would have been reasonable in the 17th century, as Leibnitz was writing about reasoning as a form of calculation, to think that the process of creating artificial intelligence would have to be something like the process of creating a waterwheel or a pocket watch: first understand the principles, then use human intelligence to devise a design based on the principles, and finally build a system in accordance with the design. At the dawn of the 19th century William Paley made such assumptions explicit, arguing that intelligent designers are necessary for the production of complex adaptive systems. And then, of course, Paley was soundly refuted by Charles Darwin in 1859. Darwin showed how complex and adaptive systems can arise naturally from a process of selection acting on random variation. That is, he showed that complex and adaptive design could be created without an intelligent designer. On the basis of evidence from paleontology, molecular biology, and evolutionary theory we now understand that nearly all of the interesting features of biological agents, including intelligence, have arisen through roughly Darwinian evolutionary processes (with a few important refinements, some of which are mentioned below). But there are still some holdouts for the pre-Darwinian view. A recent survey in the United States found that 42% of respondents expressed a belief that “Life on Earth has existed in its present form since the beginning of time” [7], and these views are supported by powerful political forces including a stridently anti-science President. These shocking political realities are, however, beyond the scope of the present essay. This essay addresses a more subtle form of pre-Darwinian thinking that occurs even among the scientifically literate, and indeed even among highly trained scientists conducting advanced AI research. Those who engage in this form of pre-Darwinian thinking accept the evidence for the evolution of terrestrial life but ignore or even explicitly deny the power of evolutionary processes to produce adaptive complexity in other contexts. Within the artificial intelligence research community those who engage in this form of thinking ignore or deny the power of evolutionary processes to create machine intelligence. Before exploring this complaint further it is worth asking whether an evolved artificial intelligence would even serve the broader goals of AI as a field. Every AI text opens by defining the field, and some of the proffered definitions are explicitly oriented toward design—presumably design by intelligent humans. For example Dean et al. define AI as “the design and study of computer programs that behave intelligently” [2, p. 1]. Would the field, so defined, be served by the demonstration of an evolved artificial intelligence? It would insofar as we could study the evolved system and particularly if we could use our resulting understanding as the basis for future designs. So even the most design-oriented AI researchers should be interested in evolved artificial intelligence if it can in fact be created.