The talk will first specify what is human intelligence according to current psychological research and theory. We will then discuss how intelligence is conceived and implemented by artificial intelligence systems.

Human intelligence is the result of evolution over millions of years, implemented by the brain. It is both modular (e.g., perceptual systems, domain-specific relational processing, such as quantitative, causal, and spatial relations) and general (Alignment, Abstraction, and Cognizance, relational integration processes, AACog) dealing with increasing efficiency with change and the unexpected. Reasoning and logic are the developmental product of AACog. AACog develops through cycles of increasing precision of representations, increasing experience-guided interdependence of representations, increasing awareness and regulation of current representations and mental processes, increasing precision of value-related regulation of choices and interpretations to fill in missing information and act proactively. This is how human general intelligence, g, develops.

Artificial intelligence simulates various aspects of human intelligence with impressive success. Two domains are prominent. Perception-based categorization (accurately recognize real world objects, specify their relations, and predict category-related membership, such as face recognition or mapping of how proteins are structured in the DNA from information about amino acids involved) and algorithmic-based problem-solving (rule-based complex activity in specific domains, such as chess). Despite these accomplishments, AI systems cannot yet deal with the unexpected. Their “intelligence” is the intelligence of their programmer rather than their own intelligence. To approach human intelligence, AI systems need to develop their own general intelligence (AGI), the holly grail of dealing with the unexpected. This has not been possible so far because AI uses wrong learning algorithms implementing ready-made regression equations matching the rules of the domain targeted.

This is not how human general intelligence, g, develops. To approach the development of human g artificial systems must go through the developmental cycles outlined above. In principle, this is possible, but we do not yet know how digital hardware may implement the development of human g as implemented by the human brain. Ideas about how this may happen will be discussed in the talk.