Some Deeply Biased Parting Thoughts

• Our models are not likely to work in the real world or have anything to do with human cognition unless they are grounded.
  • Grounding intelligence in perception, action, and The World
    To do this, we will need to build vertical, rather than horizontal, models.
  • Grounding intelligence in brains
    Unless our theorizing is geared to mesh with the neurobiological data, we risk wasting our time exploring some impossibly remote, if temporarily fashionable, corner of computational space. (Churchland and Sejnowski)
• We need to take learning and evolution seriously. We can't get very far by studying the end-state. The way the system got there can constrain our models.
• The answers are not likely to be easy ones. Intelligence will require a number of different mechanisms. The challenge will be to figure out which ones are used where and how they interact with each other.
• Human cognitive processing is continuous, distributed, and robust. The right mathematics for studying this is probably dynamical systems, not symbolic logic.
• But is there still a symbolic level at which it is convenient (or correct) to describe human behavior in symbolic terms?