Week 1: AI & Machine Learning
When does a machine learn?
- Computer autonomously "discovers" (learns) rules, strategies, etc - May or may not be programmed by humans to apply rules
Machine learning
- Learning is central to AI; if machine can't learn, we don't consider it true AI - Simplest conception of AI: machine is doing machine learning if it's learning - Conceptual fuzziness in usage of AI - Often used as interchangeable synonym for AI - Sometimes conceptualised as a subset or an application of AI
Definition of AI: Whatever computers can't do yet?
- Outdated definition (e.g. computers can now respond to voice commands); limits definition of AI to 'next big thing' - A joke; criticises overuse of the term and disagreements about and alleged coherence of AI
Deep learning
- Subset of machine learning - Complex model of learning - Rough mathematical simulation of how brain's neural networks process data - Layers of processors like layers of neurons
AI and related terms, according to Rodney Brooks
- Suitcase words; they pack in a variety of meanings, different contents get packed in over time - AI definition has evolved - Then: machines imitating intelligence - Now: includes learning
AI, as defined by David Poole and Alan Mackworth (cont.)
An agent acts intelligently when - What it does is appropriate for its circumstances and its goals - It is flexible to changing environments and changing goals - It learns from experience - It makes appropriate choices given its perceptual and computational limitations - An agent: - Typically cannot observe the state of the world directly - Has only finite memory - Does not have unlimited time to act
When does a machine not learn?
Computer only applies rules that humans programmed it to apply
Machine learning, as defined by Arthur Samuel
Computers with "the ability to learn without being explicitly programmed"
Example of machine learning
DeepMind - Demonstrates machine learning (deep learning) through superhuman performance of classic Atari video games - Machine in detail - One instruction: improve your score - No programming specific to games - Only input: pixels on screen
AI, as defined by Patrick Winston
Machine-based "architectures... that support models of thinking, perception and action" More specific definition of AI as compared to Malone and Google; machines simulate, imitate or even replicate intelligent activities
Simplest definition of AI
Machines acting in ways that seem intelligent -Tom Malone, MIT What is intelligence, then?
AI, as defined by David Poole and Alan Mackworth
The field that studies... computational agents that act intelligently... - Agent: something that acts in an environment; it does something - Computational agent: agent whose decisions about its actions can be explained in terms of computation
AI, as defined by Google
Theory and development of computer systems able to perform tasks that require human intelligence E.g. visual perception, speech recognition, decision making and translation between languages
