Lec 1-2 Flashcards
(21 cards)
Machines with minds, in the full and literal sense
Thinking - human/antropomorphic
Study of mental faculties
Thinking - rationality
Creating machines that perform functions that require intelligence
Acting - human/antromorphic
Explain and emulate intelligent behavior in terms of computational processes
Acting - rationality
Early examples AI
- Greek mythology, artificial beings
- Aristotle (BC) and Leibniz (1700) reduce logical thinking to formal calculus
- Babbage (1800) analytical engine
- Lovelace (1800) programs for engine
- Fiction
- Turing Test
- Shannon Computer Chess
Early computer science
- 1955 Logic theory Newell and Simon
- Machine learning computer checkers mid 1950’s
Here is where it took off with the advent of modern computers
Why are games useful for AI algorithms
- Clearly structured
- Clear goals/tasks
- Limited enough to be tractable
- Complex enough to be interesting
What is a microworld?
- Bounded domain
- Unambiguous and simple/simplified structure
- Clear and measurable goals
- As opposed to games; not necessarily competitive
Problems with game tree
It grows exponentially, with m moves/turn and d turns/game O(m^d) terminal states. Becomes computationally intractable
What is the idea behind a Heuristic Evaluation Function?
Somehow estimate “value” of state without expanding whole tree. It assigns a numerical score to any game state.
Might be handcrafted or learned with NNs
What is the basic idea of MiniMax algorithm?
- Optimise short-term tactical play using game tree expansion
- Score leaves with evaluation function at non-terminal depth
What are different names for Connectionist AI
- Connectionism
- Parallel Distributed Processing (PDP)
- Artificial neural networks
- Deep learning
What is the basic idea of connectionist AI:
Inspired by the human brain (not simulation).
What is the McCulloch & Pitts artificial neuron? (1943)
Binary neuron output. Different types of connections (excitatory, weights are positive vs inhibitory, weights are negative)
How does BTAN (Binary Threshold Artificial Neuron) work?
It receives n inputs gives 1 output O
All inputs are binary and have either a negative or positive weight
Activation is weighted sum of inputs
Threshold activation function with threshold
What are synonyms for Symbolic AI?
GOFAI (Good old Fashioned AI)
What are the features of Symbolic AI?
Explicit knowledge representation
- roughly human readable, knowledge is represented by symbols
It can encode relations between symbols with expressions/symbol structures
Such expressions are also knowledge.
Reasoning is resolving expressions into conclusions using rules
This reduces reasoning to a search problem
Separating knowledge and processing
Symbolic AI (in contrast to Connectionist)
What were initial uses of Automatic Computers
Numerical and for military purposes. (Code-breaking, ballistic trajectories, computations for hydrogen bomb)
Chess (Shannon), and Logic Reasoning were cautiously explored, non-numerical problems
What was the Dartmouth Workshop and the goal of it?
2 month, 10 man study of AI. Proceeding the basis of the conjecture that every aspect of learning or any other feature of intelligence can be described such that a machine can be made to simulate it.
What was the initial idea the researchers had with the Dartmouth Proposal?
An attempt to find how machines are able to use language, form abstractions and concepts, solve kinds of problems now reserved for humans and improve. They thought to make significant advance in one or more of these problems.
