What is AI? Flashcards
What is the problem with defining the AI as a system that thinks like human?
We would have to spend most of our effort on studying how people’s minds operate, rather than thinking about what intelligence pug to mean in various domains.
4 definitions of Artificial Intelligence
- Systems that think like human
- Systems that act like humans
- Systems that think rationally
- Systems that act rationally
What is Turing Test?
Operational definition of intelligence behavior.
What is wrong with the definition of the AI that is a system that acts like human.
Human often think/act in ways we don’t consider intelligent.
Why would you replicate human behavior including its limitations.
The limitation may be useful sometimes. Example : supporting human learning via teachable agents.
Rationality
An abstract ideal of intelligence.
What is the problem where defining the AI as a system that thinks rationally.
A system that only thinks and doesn’t do anything is quite useless, and it is hard to measure thought.
Why using rationality to describe AI is better than using human behavior?
Rationality is more cleanly defined. And because of that:
- it’s a better design objective
- in cases where human behavior is irrational, often we’d prefer rationality.
- once we have rational agent we can always tweak it to become irrational.
Why acting rationally is the best description of AI?
Because rationality is better defined than human behavior and action is better defined that thought.
Which definition of AI do we use?
The study and design of systems that act rationally.
What is an intelligent agent?
Artifacts that act rationally in their environment.
What are 4 intelligent agent’s properties?
1) Their actions are appropriate for their goals and circumstances.
2) They are flexible to changing environments and goals/
3) They learn from their expirience.
4) They make appopriate choices given perceprtual limitations and limited resources.
What is cognitive plausibility?
A constrain in studying AI that may lead to just replicating human performance instead if understanding the mathematical principles behind the intelligence.
What is robot?
An artificial agent that physically present in the world.
What is software agent? Give an example.
An agent without a physical presence. Example: text-based tranlation agents, intelligent tutoring system.
What is the similiarity between a software agent and a robot?
They share all fundamental components of intelligent behaviour, except for the way they interact with the environment.
What are 4 inputs to an intelligent agent?
1) Prior knowledge
2) Past experiences
3) Goals/values
4) Observations
What is the output of an intelligent agent?
Actions
Representation of the input of an agent?
Knowledge
A goal of AI?
Specify how system can:
- Acquire and represent knwoledge about a domain (representation)
- Use the knowledge to solve problems in that domain. (Reasoning)
What does a representation and reasoning system contain?
- Representation language
- Computational reasoning procedures
What is a R&R system’s pipeline?
problem -> representation -> computation -> representation -> solution
What is representation language used for in R&R systems?
It is used to desribe the environment and problems to be solved.
What are computational reasoning procedures used for in R&R systems?
They are used to compute a solution to a problem.
What is the choice of an appropriate R&R system depends on?
Various dimensions: properties of the environment, type of problems, the agent, the computational resourses.
2 kinds of problems?
1) Static
2) Sequential
What problem types are ones where finding a solution does not involve reasoning into the future?
Static
What problem types are ones where finding a solution does involve reasoning into the future?
Sequential
What are fixed and indefinite horizons?
In sequentail problems, fixed horizon means that the problem require reasoning a fixed number of steps into the future.
Indefinite horizon - finate, but unknown number of steps.
3 problem types in AI?
1) Constrain Satisfaction (static)
2) Answering Query (static)
3) Planning (sequential)
Name the problem type that involves finding state that satisfies set of constrain.
Constrain satisfaction
Name the problem type that involves checking if a given proposition true/likely, given what is known.
Answering query
Name the problem type that involves finding sequence of actions to reach a goal state/maximize outcome.
Planning
Name 2 types of environment.
1) Deterministic
2) Stochastic (Uncertain)
What does sensing unsertanty mean?
It means that the agent cannot fully observe the current state of the world.
What does effect uncertainty mean?
The agent does not know for sure the effects of it’s actions.
How to know if environment is stochastic or deterministic?
An environment is stochastic if at least one of sensing uncertainty or effect uncertainty if true, otherwise it is deterministic.
Name 2 representation types.
1) Logic
2) Probability
What are explicit state, features, and relations representation schemes?
Explicit state - modeling the environment by enumerating states.
Features - modeling the environment by describing a state in terms of features.
Relations - describing states in terms of objects and relations.
Name 2 types of world representations based on the abstraction levels.
1) Flat - single level of abstraction
2) Hierarchical - multiple level of abstraction.
What is hierarchical representations are mainly used for?
Scaling up flat representations
What is the difference between using knowledge given and knowlledge learned from experience?
When knowledge is given, the agent given a model of the world once and for all. When knowledge learned from experience is used, the agent can learn how the world work based on experience, but it still starts with some prior knowledge.
What is the difference when an agent has goals or preferences?
When an agent has a goal, there is some state or set of states that the agent want to be in and there is some proposition or set of propositions that the agent wants to make true.
When an agent has preferences, agent’s task is to reach a state which makes it as happy as possible and how happy the agent is in each state is described ny a preference/utility function.
Name 3 kinds of agents that may exist in multi-agent domains.
1) Cooperative
2) Competitive
3) A bit of both
Name 5 examples of AI use.
1) Natural language understanding
2) Computer vision
3) Speech recognition
4) Robotics
5) Language generation
