AI Flashcards
Explain the difference between weak AI and strong AI. What are their assumptions and what does that
mean for AI?
Weak:
- non-sentient
- focused on narrow task
- based on rule-based manipulation (eg. if… then…)
- not ‘truly intelligent’
- Searle: chinese room argument= doesn’t reflect true intelligence
strong: - can think - requires connectionist architecture - assumes human mind to be an info processing system = thinking= form of computing
Explain the difference between symbolic AI and connectionist AI.
symb:
- rule-based (symbol manipulation)
- no pattern recognition
- cannot classify objects or deal w noisy input
c:
- based on networks similar to neurobiology (parrallel processing)
- can solve complex non-linear problems
- can generalise novel input (learn)
What was Turing’s imitation game about and how does this relate to intelligent systems?
game:
- if machine indistinguishable from human,
= must be intelligent
- doesn’t question underlying mechanism
- is it actually intelligent?
How does Searle’s Chinese room argument relate to Turing’s imitation game? What conclusions would
Searle draw out of a computer passing the imitation game?
- even when programmed to answer all questions correctly, it doesn’t mean intelligence
- doesn’t understand essence of chinese
- simply programmed to answer
Explain the criticism on symbolic AI. Give arguments against and for.
for:
- MYCIN was v successful (diagnosing blood infections)
against:
- highly inflexible (relies on programmed symbols
- cant deal w noise
- can’t reach human-like intelligence
What are the similarities between connectionist AI architectures and the human brain?
- based on the structure of a human brain
- parallel processing
- can generalise (learn)
What does it mean for human memory to be content-addressable?
- means that memory can be retrieved by using content as cue
- possible due to memory not stored in neurons but in the connections b/n them
What is an expert system?
representation of an expert’s knowledge of a subject
what are expert system’s components and their functions?
knowledge base:
- knowledge about the world
- how concepts relate to each other
inference engine:
- manipulate the symbols
eg . predicate logic
Explain how a conclusion is deduced in an expert system using forward and backward chaining. What are
the algorithms of both chaining directions?
forward: - sequential until solution found - if then rules applied - facts are matched = deduce conclusion
backward:
- have a hypothesis
- use if then rule backwards
- check if antecedents be true until hypo accepted/ rejected
- if all antecedents accepted= conclusion
(skips unneccesary qustions)
When should you use forward chaining, and when backward chaining?
- f when want to know everything you can from a set of facts
- b when not all facts are known
Sudoku:
Explain the backtracking
algorithm and give an algorithmic implementation of backtracking in solving a Sudoku.
- iteratively generating possible solutions
- when realise that it doesn’t lead to a complete solution
- goes back to previous choice and creates a new solution
what is the difference between depth-frst search and breadth-frst search?
DFS:
- goes to 1 child and expand on all its children before expanding on the next child and all its children
- alternatives are ignored
- backtracks when deadend met
BFS:
- expands on everything on first layer before moving on to the next layer
Which type (BFS/ DFS) should be used under what circumstances? Explain.
BFS
- used when v deep trees
- not when all paths would reach goal about the same time
DFS
- used when trees not that deep
- goal expected to be found after a reasonable amount of steps
How does a heuristically informed algorithm differ from its naive counterpart?
has more info than naive counterparts
- allows for better decisions
We can heuristically inform depth-frst and breadth-frst search using, for example, a distance-to-target
metric. Name the two algorithms that result from this, and explain how they work.
GREEDY ALGORITHMS
(choose locally best choice at each stage)
hill-climbing
- moves similar to DFS
- uses shortest distance to target as heuristic
beam search
- modified BFS
- moves thru the best nodes
In optimal search, the branch-and-bound algorithm works by discarding some branches. Explain which
branches are discarded by the algorithm and why.
(uniform-cost search)
- it ignores path w distances that are equal or greater than the distance for that solution
- benefits from the having the estimated distance left
- discards redundant paths for more efficiency
The A* algorithm combines three techniques that were discussed in the lectures. Name these 3
techniques, and how they contribute to efficient path finding.
- branch-and-bound
- distance estimates
- redundant path removal
- efficient thru evaluating the cost to the node and the cost from the node to the goal
Give a definition and an example of combinatorial explosion.
- rapid growth of the complexity and possible solutions to a problem
- eg. trying to solve sudoku by using naive brute force
What does it mean when a game (e.g. checkers, chess) is solved?
when optimal steps for each player can be calculated
Explain why robots need sensors and effectors, and include their defnitions.
sensors:
- allows to perceive environment
effectors:
- so it can move
What are degrees of freedom in robotics? Give an example.
- independent parameters that define the system’s configuration
- eg. drone has 6 df
3 for location and 3 for angular orientation
How does the behavior of two Braitenberg Vehicles differ between straight vs. crossed excitatory
connections? Describe their behavior exhaustively
straight: fear
- turns away from signal and slows down as it goes further away
crossed: aggression
- turns toward signal
- faster as closer to signal
L3
What happens to these two vehicles if we make the connections inhibitory?
LOVE
straight: love
- race toward signal
- stops in front of it
crossed: unfaithful lover
- orient itself away from signal
- speed up as further from signal
What happens when we implement a threshold function?
- will appear to make its own decision
= have own values
Explain the concept of evolutionary robotics using pseudocode.
(genetic algorithm)
- choose population
- evaluate each individuals’ fitness
- only keep best performing robots
- replenish population w well-performing robots
- terminate when fitness reached a plateau
Name two robot learning techniques
motor babbling:
reinforcement learning:
motor babbling
robot learning technique
- random movements initially (M)
- until movement changes body configuration and environment (K)
- co-activation b/n M&K= bidirectional associations
- could also learn by human guiding movements (imitation learning)
- build model of required movement
reinforcement learning
robot learning technique
- send reinforcement signal to the control system
- signal usually binary (altho other values possible)
- tells machine whether it passed/ failed
