Conclusions

Questions
- will the agent be intelligent, thinking or conscious?
  - see Chapter 26, Russell and Norvig (rn26)
- what will happen if we succeed
  - see Chapter 27, Russell and Norvig (rn27)
- what goes into a general-purpose intelligent agent?
  - is agent-based architecture the best?
- what is the ideal platform for the agent?
  - hardware/software
  - object-oriented design
  - parallelism

Intelligent Agents
- functionality
  - problem-solving and search
  - knowledge and reasoning
  - planning
  - uncertainty reasoning
  - learning
  - perception
    - natural language processing
    - vision
    - speech processing
  - other functionality
    - taste
    - smell
    - touch
    - manipulation of environment
      - robotics
- computational constraints
  - rationality
  - bounded optimality
- consider example from first class
  - can we build an intelligent TAGER student?
  - knowledge? (CYC?)
- non-agent-based approach
  - when is percept/action interface violated?
    - embedded AI
  - isolation requires ability to communicate arbitrary knowledge via
    percepts (KIF?)

Platform
- C
  - need symbol processing (library)
  - need inference engine (library)
  - only procedural
  + low-level control
  + fast
- Lisp
  + symbol processing
  - still need inference engine (library)
  - still only procedural
  + reasonably low-level control
  - slow
  + can be converted to C
- Prolog
  + symbol processing
  + built-in inference engine
  + declarative and procedural
  - low-level control difficult
  - slow
  + can be converted to C

Object-Oriented Design
- suffers from lack of foundational AI
  - knowledge representation (FOPC?)
  - reasoning (deduction?)
  - problem solving (search?)
  - planning (POP?)
  - uncertainty (belief networks and Bayes rule?)
  - learning (decision tree induction?)
  - perception (edge detection, ... ?)
+ possible within technique
+ supports multi-agent systems
  - complex systems (e.g., economy, ecology, sociology, brain?)

Parallelism
- massive parallelism
  - to big and expensive
+ distributed processing
+ analog computing
  - massive parallelism (like brain)
  - specialized computations