Searching for the principles of a less artificial A.I.

Abstract

What would it take to build a computer physician that can take its place amongst human peers? Currently, Neural Nets, especially as so-called “Deep Learning” nets, dominate what is popularly called “Artificial Intelligence”, but to many critics they seem to be little more than powerful data-analytic tools inspired by some of the more basic functions and regions of the human brain such as those involved in early processes in biological vision, classification, and categorization. The deeper nature of human intelligence as the term is normally meant, including relating to consciousness, has been the domain of philosophers, psychologists, and some neuroscientists. Now, attention is turning to neuronal mechanisms in humans and simpler organisms as a basis of a truer AI with far greater potential. Arguably, the approach required should be rooted in information theory and algorithmic science. But as discussed in this paper, caution is required: “just any old information” might not do. The information might need to be of a particular dynamical and actioning nature, and that might significantly impact the kind of computation and computer hardware required. Overall, however, the authors do not favor emergent properties such as those based on complexity and quantum effects. Despite the possible difficulties, such studies could, in return, have substantial benefits for biology and medicine beyond the computational tools that they produce to serve those disciplines.

Keywords

• AI
• Algorithms
• X factor
• Emergent properties
• Consciousness
• Quantum effects