Course Overview Agents acting in an environment Future and Ethics of AI Dimensions of complexity c D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 1
Course Overview
Agents acting in an environment
Future and Ethics of AI
Dimensions of complexity
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 1
What is Artificial Intelligence?
Artificial Intelligence is the synthesis and analysis ofcomputational agents that act intelligently.
An agent is something that acts in an environment.
An agent acts intelligently if:I its actions are appropriate for its goals and circumstancesI it is flexible to changing environments and goalsI it learns from experienceI it makes appropriate choices given perceptual and
computational limitations
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 2
Agents acting in an environment
Prior Knowledge
Environment
StimuliActions
Past Experiences
Goals/Preferences
Agent
Abilities
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 3
Inside Black Box
Learner Inference Engine
offline online
Prior Knowledge
Past Experiences/Data
ObservationsGoals/Preference
Actions
Abilities
KB
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 4
Controller
memories Controller
percepts commands
Body
memories
Environment
stimuli actionsAgent
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 5
Functions implemented in a controller
memories
percepts commands
memories
For discrete time, a controller implements:
belief state function returns next belief state / memory.What should it remember?
command function returns commands to body.What should it do?
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 6
Future and Ethics of AI
What will super-intelligent AI bring?
I Automation and unemployment? What if people are notlonger needed to make economy work?
I Smart weapons? Automated terrorists?
What will a super-intelligent AI be able to do better?I predict the futureI optimize (constrained optimization)
Whose values/goals will they use? (Why?)
Will we need a new ethics of AI?
Is super-human AI inevitable (wait till computers getfaster)? (Singularity)Is there fundamental research to be done?Is it easy because humans are not as intelligent as we liketo think?
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 7
Future and Ethics of AI
What will super-intelligent AI bring?I Automation and unemployment? What if people are not
longer needed to make economy work?
I Smart weapons? Automated terrorists?
What will a super-intelligent AI be able to do better?I predict the futureI optimize (constrained optimization)
Whose values/goals will they use? (Why?)
Will we need a new ethics of AI?
Is super-human AI inevitable (wait till computers getfaster)? (Singularity)Is there fundamental research to be done?Is it easy because humans are not as intelligent as we liketo think?
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 8
Future and Ethics of AI
What will super-intelligent AI bring?I Automation and unemployment? What if people are not
longer needed to make economy work?I Smart weapons? Automated terrorists?
What will a super-intelligent AI be able to do better?I predict the futureI optimize (constrained optimization)
Whose values/goals will they use? (Why?)
Will we need a new ethics of AI?
Is super-human AI inevitable (wait till computers getfaster)? (Singularity)Is there fundamental research to be done?Is it easy because humans are not as intelligent as we liketo think?
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 9
Future and Ethics of AI
What will super-intelligent AI bring?I Automation and unemployment? What if people are not
longer needed to make economy work?I Smart weapons? Automated terrorists?
What will a super-intelligent AI be able to do better?
I predict the futureI optimize (constrained optimization)
Whose values/goals will they use? (Why?)
Will we need a new ethics of AI?
Is super-human AI inevitable (wait till computers getfaster)? (Singularity)Is there fundamental research to be done?Is it easy because humans are not as intelligent as we liketo think?
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 10
Future and Ethics of AI
What will super-intelligent AI bring?I Automation and unemployment? What if people are not
longer needed to make economy work?I Smart weapons? Automated terrorists?
What will a super-intelligent AI be able to do better?I predict the futureI optimize (constrained optimization)
Whose values/goals will they use? (Why?)
Will we need a new ethics of AI?
Is super-human AI inevitable (wait till computers getfaster)? (Singularity)Is there fundamental research to be done?Is it easy because humans are not as intelligent as we liketo think?
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 11
Future and Ethics of AI
What will super-intelligent AI bring?I Automation and unemployment? What if people are not
longer needed to make economy work?I Smart weapons? Automated terrorists?
What will a super-intelligent AI be able to do better?I predict the futureI optimize (constrained optimization)
Whose values/goals will they use? (Why?)
Will we need a new ethics of AI?
Is super-human AI inevitable (wait till computers getfaster)? (Singularity)Is there fundamental research to be done?Is it easy because humans are not as intelligent as we liketo think?
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 12
Future and Ethics of AI
What will super-intelligent AI bring?I Automation and unemployment? What if people are not
longer needed to make economy work?I Smart weapons? Automated terrorists?
What will a super-intelligent AI be able to do better?I predict the futureI optimize (constrained optimization)
Whose values/goals will they use? (Why?)
Will we need a new ethics of AI?
Is super-human AI inevitable (wait till computers getfaster)? (Singularity)Is there fundamental research to be done?Is it easy because humans are not as intelligent as we liketo think?
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 13
Future and Ethics of AI
What will super-intelligent AI bring?I Automation and unemployment? What if people are not
longer needed to make economy work?I Smart weapons? Automated terrorists?
What will a super-intelligent AI be able to do better?I predict the futureI optimize (constrained optimization)
Whose values/goals will they use? (Why?)
Will we need a new ethics of AI?
Is super-human AI inevitable (wait till computers getfaster)? (Singularity)Is there fundamental research to be done?Is it easy because humans are not as intelligent as we liketo think?
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 14
Dimensions of Complexity
Flat or modular or hierarchical
Explicit states or features or individuals and relations
Static or finite stage or indefinite stage or infinite stage
Fully observable or partially observable
Deterministic or stochastic dynamics
Goals or complex preferences
Single-agent or multiple agents
Knowledge is given or knowledge is learned fromexperience
Reason offline or reason while interacting withenvironment
Perfect rationality or bounded rationality
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 15
State-space Search
flat or modular or hierarchical
explicit states or features or individuals and relations
static or finite stage or indefinite stage or infinite stage
fully observable or partially observable
deterministic or stochastic dynamics
goals or complex preferences
single agent or multiple agents
knowledge is given or knowledge is learned
reason offline or reason while interacting with environment
perfect rationality or bounded rationality
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 16
Classical Planning
flat or modular or hierarchical
explicit states or features or individuals and relations
static or finite stage or indefinite stage or infinite stage
fully observable or partially observable
deterministic or stochastic dynamics
goals or complex preferences
single agent or multiple agents
knowledge is given or knowledge is learned
reason offline or reason while interacting with environment
perfect rationality or bounded rationality
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 17
Decision Networks
flat or modular or hierarchical
explicit states or features or individuals and relations
static or finite stage or indefinite stage or infinite stage
fully observable or partially observable
deterministic or stochastic dynamics
goals or complex preferences
single agent or multiple agents
knowledge is given or knowledge is learned
reason offline or reason while interacting with environment
perfect rationality or bounded rationality
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 18
Markov Decision Processes (MDPs)
flat or modular or hierarchical
explicit states or features or individuals and relations
static or finite stage or indefinite stage or infinite stage
fully observable or partially observable
deterministic or stochastic dynamics
goals or complex preferences
single agent or multiple agents
knowledge is given or knowledge is learned
reason offline or reason while interacting with environment
perfect rationality or bounded rationality
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 19
Decision-theoretic Planning
flat or modular or hierarchical
explicit states or features or individuals and relations
static or finite stage or indefinite stage or infinite stage
fully observable or partially observable
deterministic or stochastic dynamics
goals or complex preferences
single agent or multiple agents
knowledge is given or knowledge is learned
reason offline or reason while interacting with environment
perfect rationality or bounded rationality
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 20
Reinforcement Learning
flat or modular or hierarchical
explicit states or features or individuals and relations
static or finite stage or indefinite stage or infinite stage
fully observable or partially observable
deterministic or stochastic dynamics
goals or complex preferences
single agent or multiple agents
knowledge is given or knowledge is learned
reason offline or reason while interacting with environment
perfect rationality or bounded rationality
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 21
Relational Reinforcement Learning
flat or modular or hierarchical
explicit states or features or individuals and relations
static or finite stage or indefinite stage or infinite stage
fully observable or partially observable
deterministic or stochastic dynamics
goals or complex preferences
single agent or multiple agents
knowledge is given or knowledge is learned
reason offline or reason while interacting with environment
perfect rationality or bounded rationality
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 22
Classical Game Theory
flat or modular or hierarchical
explicit states or features or individuals and relations
static or finite stage or indefinite stage or infinite stage
fully observable or partially observable
deterministic or stochastic dynamics
goals or complex preferences
single agent or multiple agents
knowledge is given or knowledge is learned
reason offline or reason while interacting with environment
perfect rationality or bounded rationality
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 23
Humans
flat or modular or hierarchical
explicit states or features or individuals and relations
static or finite stage or indefinite stage or infinite stage
fully observable or partially observable
deterministic or stochastic dynamics
goals or complex preferences
single agent or multiple agents
knowledge is given or knowledge is learned
reason offline or reason while interacting with environment
perfect rationality or bounded rationality
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 24
Comparison of Some Representations
CP MDPs IDs RL POMDPs GThierarchical 4
properties 4 4 4
relational 4
indefinite stage 4 4 4 4
stochastic dynamics 4 4 4 4 4
partially observable 4 4 4
values 4 4 4 4 4
dynamics not given 4
multiple agents 4
bounded rationality
c©D. Poole and A. Mackworth 2016 Artificial Intelligence, Lecture 16.1, Page 25