Week 2: History of AI Flashcards
Where did AI originate?
-AI began at the 1956 Dartmouth Conference
-Founding fathers came from many different disciplines
What were the original goals of AI?
-make machines use language
-form abstractions and concepts
-solve kinds of problems now reserved for humans
-improve themselves
AI Boom 1
*Primarily searching a solution given fixed rules
-introduction of AI and hueristic (if I give AI some rules, what tricks does it do to complete the search)
-people very optimistic and interested
-GOFAI AI system
-Winter 1 (decreased interest) around the 70s
AI Boom 2
*-“Expert systems”: extracting knowledge from human experts to guide the solution of an AI system
-emerged in the 80s
-Progress with knowledge engineering
-Huge investment and beginning to be used in the real world
-winter towards the 2000s
AI Boom 3
-currently in this wave
-machine learning
-huge explosion of AI
-Unclear whether progress will continue increasing or if there will be another winter - so far there hasn’t been another winter
(unlikely to be another winter- the business model shows a continuous need for AI that is heavily dependent on consumers)
What does the dashed line represent in Arthur Miller’s checker player game?
Above the dashed line represents what actually happened in the gain and below the dashed line are hypothetical situations
How did Arthur Miller’s checker game function?
-Modeled as a tree
-AI is thinking about all of the possible moves it can play
-For each move, it will consider countermoves from the opponent and from that move, the other possible actions it can take
-Every leaf node is a possible future
How is simulating the future of a board game different from real life?
Simulating the future of a board game is easier than real life because you have fixed rules, limited stakes and actions, and full information
What is the myth around machine learning intelligence and what disproved it?
-Myth suggesting that a machine’s intelligence cannot surpass that of its creator
-The checker game disproves this idea by learning to play checkers better than the creator
Reasons for failure of the first wave
-Didn’t scale up to larger problems - can’t simply have a faster machine with a larger memory
-The fact that a program can find as solution in principle does not mean that the program contains the mechanisms needed to execute the solution in practice
State space definition
how many different configurations there can be on a game board
Game tree size definition
all of the possible moves based on the leaf nodes (i.e. how the game can evolve)
How do state space and game tree size differ across games?
The state space and the game tree size for chess is much larger than checkers because there are more pieces and rules for each piece – even larger in the game Go
What do the games of checkers, chess, and go represent?
Represent the three milestones in AI because AI designed for each game cannot play the other. A special type of AI had to be designed for each game.
Reasons for failure for second wave
-The reasoning methods used by the systems broke down in the face of uncertainty
-The systems could not learn from experience
What is Deep Blue?
-chess playing hardware system that beat the chess grandmaster
Hardware of Deep Blue
-Capable of evaluating 200 million positions per second, enabling it to explore numerous possible moves and counter moves in real time
-Stored 700,000 grandmaster games and more than 4000 positions
-Enumerates every possible move so you can’t beat deep blue once you reach five or fewer pieces
AI Seeing vs Doing Shortcoming
-AI can look at the probability of certain events, it can’t detect third variables influencing this data
-We intuitively understand the causal relationship between two variables. It is difficult for AI to understand this.
e.g. Crime rates and high ice cream consumption are correlated
To decrease crime rates, should we close ice scream shops – AI would say yes
How did AI Boom 3 operate?
Used Deep Learning to learn statistical patterns from data
What is AI computer vision?
the ability to detect and idenitfy an image
How does AI computer vision work?
-each pixel is assigned a number value that corresponds to the color shade (for example, large number represent white and number closer to 0 represent dark)
-intelligence then analyzes this data for patterns
Pixel
the smallest unit of an image that can be individually controlled or displayed on a digital screen or represented in an image file
