Intelligence is the cognitive ability to understand the world; to help achieve a wide variety of goals; and to integrate and adapt such knowledge and skills in ongoing learning. It needs to function in real time, in the real world, and with limited knowledge and time.
Human intelligence is special in that it features the ability to form and use highly abstract concepts, and to think and reason using symbols (language).
Here’s a list of distinctive and essential features of human intelligence, and thus advanced AI (also see AGI Checklist):
-
It must be general: An individual intelligence must be able to learn and do a wide range of cognitive tasks
-
Real-time interactive learning utilizing environmental feedback
-
Learning, reasoning, and acting under knowledge and resource constraints
-
Unsupervised and self-supervised learning. One-shot/ instance learning
-
Short and long-term memory of interactions, plus prediction
-
Selective unlearning and re-learning (i.e. correcting errors)
-
Learning, recognition, and reasoning utilizing dynamic context and goals
-
Real-time autonomous concept formation, including highly abstract ones
-
Natural language ability, including deep understanding
-
Real-time adaptive focus and selection
-
Has concept ontology grounded in experience, including cause and affect relationships
-
Inherent transfer-learning: knowledge and cognitive skills can be applied to any domain
-
Deep integration and synergistic interaction of knowledge, memory, cognition and skills
-
Introspection, and meta-cognitive control
-
Cognition/ reasoning that includes a theory-of-mind (other agent’s mental states)
Artificial intelligence has these additional desirable characteristics:
-
Data and algorithms are scrutable (not black box)
-
Externally generated ontologies and skills can be integrated
-
External conventional databases and programs can be deeply connected
-
‘Brains’ can be merged
Interestingly, the current focus in AI research is on machine learning, which address few of these requirements of intelligence.
So one may ask, why are so few people working on ‘real AI’ — why don’t we have AGI yet?