Strategies for Landing an Oracle DBA Job as a Fresher
Playing games with_cc
1. PLAYING GAMES WITH
COGNITIVE COMPUTING
Cognitive Systems Institute Group
Speaker Series
Simon Ellis
Department of Computer Science ◇ Tetherless World Constellation
Rensselaer Polytechnic Institute, Troy, NY 12180
ELLISS5@RPI.EDU
Thursday, 30th April, 2015
2. Q10
v Lots of games!…
v AI agents play some games well and some badly, but why?
Games
3. Q10
Game complexity
v Arises from multiple aspects of the game design; e.g.:
v Data structure (amount of game state information)
v Rule structure and complexity
v Level of bluffing and inference
v Degree of openness
v …
v Game theory defines other parameters for games; e.g.:
v Zero-/Non zero-sum
v Deterministic/Stochastic
v Impartial/Partisan
v Perfect/Imperfect information
5. Q10
IBM Watson
v Designed to play – and win – at a “humans-only” game
v Consider the search space of Jeopardy!:
v English language (including borrows, loan words, calques…)
v Proper nouns
v Foreign words
v Phrases
v …
v How did Watson manage it – with 3 seconds per question?
7. Q10
“Cognitive game-playing”
v Drawing inspiration from two sources
v DeepQA (Watson) architecture
v Human approaches to playing games
v How do humans play games?
v Questions (“Where can I play?”, “What if I/they play/move…?”)
v Intuition or instinct based on past play experience
v Logic (inductive, deductive, abductive or analogical)
v Mood
v Strategy
v Self-evaluation (“Am I winning?”, “Should I change strategy?”…)
v …
8. Q10
Architecture model
v Architecture…
v … was inspired by the design of the DeepQA pipeline
v … is informed by consideration of how people play games
v … uses numerous tools (“evaluators”) to judge game state
u Evaluators correspond to the sections and subsections of the pipeline
PRIMARY GENERAL ANALYSIS
Where can I play?
Where can I not play?
What can I play?
SECONDARY GENERAL ANALYSIS
What is my score?
Can I win this turn?
Do I have any valuable tiles?
What is my position like?
MOVE GENERATION
What moves exist?
Do chains of moves exist?
PRIMARY MOVE SCORING
Will this advance my position?
What would my new score(s) be?
GENERAL META-ANALYSIS
Who is winning?
What tile might come up next?
Can I disrupt a player’s game?
What happens if I play tile M?
INPUT
STATE
OUTPUT
STATE
TACTICS
Can I control more of the board?
How many tiles can I play now?
Can I swap hands? Should I do so?
Should I retain tile Q for later?
TILE-SPECIFIC META-ANALYSIS
How can I use tile X best?
Does tile Y give me any benefit?
Can I perform combo move Z?
FINAL SCORING AND RANKING
Which move has the highest score?
What other moves score highly?
Which move gives me the highest score?
“DEEP THINKING”
How well does this move fit my tactics?
Should I change my gameplay?
Is it worth playing a lesser move now?
9. Q10
“Deep thinking”
v Could also be termed described as “meta-reasoning”
v Reasoning over meta-data
v Meta-data come from various sources
u Data derived from information about the game (current & past states)
u Self-analysis of agent’s performance
u …
v “Deep thinking” in strategy
v Agent has some pre-programmed “strategies”
v Analysis of agent’s own performance using one of these strategies
can be analysed (i.e. the agent has a degree of reflection)
v The agent can decide to change its strategy if it determines it would
be advantageous to do so
12. Q10
“Cognitive game-playing”
v Development of evaluators
v Mostly for gameplay decisions
v Some are already conceptualised
u “Where can I play?”, “What can I play?”
u “What is player X’s score?”, “What is the likelihood of drawing tile Y?”
v Many others will be required
u Some will emerge during development; i.e. to do P, Q, R and S are needed
u Others have emerged during research involving human players (e.g. RPI
Games Club, undergraduate volunteers)
v What about more general strategy?
v Without branch-and-bound, how can we make sure the agent plays
its best?
13. Q10
“Deep thinking”
v Strategy is a major component in an AI for a complex game
v Definition of strategy: “an overall methodology for playing a game”
v Simple strategies will be developed
v Goal will generally be to maximise score
v Several different methodologies possible in Infinite City
u Get single largest block of tiles
u Get highest number of scoring tiles
u Acquire score through tile bonus points
u …
14. Q10
“Deep thinking”
v “Deep thinking” system will evaluate performance
v Perform analysis over a set of evaluators
u Which evaluators work well or badly will be a matter of research
u Different strategies may well have some different inputs
v Heuristics will be necessarily simple
u May be as simple as a set of if (...) statements
u Again, a matter of research to see what works well
v Based on results, the agent may change its strategy
v Aim is to provide the agent with a degree of self-reflection
v Ability to judge its own performance using provided criteria
15. Q10
Conclusion
v Watson demonstrated the efficacy of ‘cognitive computing’
v “Cognitive game-playing” is a development of this technique
v Many tabletop games have extremely large search spaces
v Traditional A.I. search techniques do not work well for such games
v This is a powerful and flexible approach to game-playing
v Provides a solution to problems of extreme game complexity
v Self-analysis injected into system through “deep thinking”
v Makes possible very powerful, flexible, interesting artificial gamers
u … which might, one day, take on the ultimate gaming challenge…
16.
17. Q10
Acknowledgements
I would like to thank my supervisor, Professor Jim Hendler, for his continued support and advice, and for taking a chance on a stranger with some crazy ideas and offering me the initial
opportunity to work with Watson. I would also like to thank Dr Chris Welty and Dr Siddharth Patwardhan for their assistance and insights which led semi-directly to this work, Dr Bijan Parsia
(University of Manchester, UK) for his timely intervention in asking difficult questions which I had been avoiding, and Professor Selmer Bringsjord (RPI) for his consistently insightful comments
and observations.Additionally, sincere thanks are due to Dr Jonathan Dordic and Mr John Kolb (RPI) for their support, and to my other friends and colleagues at RPI likewise for theirs.