AI Overview

Three Ways To Define AI:

• As a system.
• As a collection of computational techniques.
• As a multidisciplinary research field.

AI as a System

Agent

An agent is anything that can be viewed as perceiving its environment through sensors and acting upon this environment through actuators.

Rational

A rational agent is an agent that selects actions in order to maximize its performance measure, given evidence provided by the percepts and andy built-in knowledge.

AI as a collection of techniques

Machine Reasoning

Explicit knowledge representation + inferences to derive new knowledge.

Done by using propositional logic:

• Pros
  • Precise Specification
  • Potentially more explainable
  • Open to correction
• Cons
  • Restricted expressiveness
  • Some knowledge hard to capture
  • Difficult to obtain all relevant knowledge.

Optimization

Algorithms for finding the best solutions according to some criterion of optimality. for example number of steps, execution time...

Machine Learning

• Traditional programming takes input and logic (rules) and produces a output.
• Machine Learning takes a mapping of input and output and produces logic (rules).

null

Supervised Machine Learning

• Advantages
  • No need to model knowledge explicitly
  • Discover new relations between Input and output.
• Challenges
  • handling noise.
  • handling biases in data.
  • optimization of generalization ability.
  • validation: performance with new data.

Security and ML

Ability to automate & generalize identification of new threats.

Domain specific challenges:

• robustness against evasion attacks.
• robustness against evolving attacks.

AI as a research field

• Views AI as as a socio-technical system.
• Combines expertise from many fields:
  • cs
  • hci
  • psychology
  • law
  • etc...