Introduction
Overview -> Introduction -> lecture
Briefing
This module is more than hard facts. While Artificial Intelligence solves a lot of important problems very well, it is also a technology full of surprises, ambiguities, and moral dilemmas. For this reason you have to piece together the material yourselves and look for a meaning whih is relevant to you. It is important to read and hear different perspectives on the subject, and therefore I provide a range of video links and a long reading list.
(1) Practical Information
- Wiki + BlackBoard
- Practical Information at the Front Page
- Compulsory Assignments
- Exam Format
Sessions
- Typical session format.
- Briefing 8.15-9.15
- Break
- Exercise 9.30-11.30
- Lunch
- Status 12.15-12.45
- Exercise 12.45-13.15
- Break
- Status 13.30-14.00
- Today, the briefing will take longer.
- I’ll try to stick with 12.15 and 13.30 as the times to gather, especially in digital sessions.
- I will be available most of the exercise time to help and discuss, and I may make plenary addresses if important questions arise.
(2) Debate: Key recurring questions
- What is intelligence?
- What is AI?
- Is AI possible?
(3) Lecture: Background and Introduction
Reading Russel & Norvig Chapter 1
What is AI?
- Rational versus Human
- the Turing Test
- Six areas of AI
- value alignment problem
Reason
- Theoretical versus Practical reason
- instrumental reason
- Hume’s is-ought problem
- Deduction, Induction, Abduction, Analogy
Foundations
- Philosophy - Mathematics - Economics
- Neuroscience - Psychology
- Computer Engineering - Control Theory - Cybernetics
- Linguistics
Parallel development.
- Computerisation of scientific disciplines.
- Disciplines provide models for AI
- Development of AI leads to better Algorithms
Philosophy
Epistemology
- How do we know? How do we know that we know?
- Hume’s is-ought problem
Goals
We deliberate not about ends, but about means. For a doctor does not deliberate whether he shall heal, (Aristotle, Ethics)
Simon & Newell implemented Aristotle’s principle as the General Problem Solver.
Schön in The Reflective Practitioner observed a doctor who had to deliberate. The treatment might help one disease and cause another.
Mathematics and Computing
- Logic
- Gödel’s incompleteness theorem
- Computability (NP-completeness)
Neuro-science
- Modelling the brain
- Simulating the brain
Psychology
- Cognition
- Computer Models explain Psychology
Computer Engineering
- Computing power
Control Theory
- Cybernetics - navigator/helmsman
Epochs
- 19th C: Ada Lovelace
- the analytical engine can only do what we tell it to do
- Dawn
- McCulloch & Pitt 1943: Neural Networks
- Hebb 1949: Hebbian Learning
- Minsky & Edmonds 1950: first computer running ANN - 40 neurons
- Turing 1950: the imitation game
- Birth of AI as a separate field 1956
- Simon & Newell: Logic Theorist
- Enthusiasm and Expectation 1952-59
- coincides with Scientification
- Lisp 1958, Eliza, GPL
- Machine Evolution 1958
- Reality 1966-73
- optimistic ten-year predictions fail
- three challenges
- AI has no domain knowledge
- intractible problems
- linmited representations (two perceptrons do not suffice)
- Knowledge-based systems 1969-79
- Previous limitation: weak methods which do not scale. Brute force?
- Solution: knowledge bases
- From 1980: Industrialisation
- first commercial expert system 1982 (DEC)
- From 1986: ANN returns - connectionist opposed to symbolic and logicist
- From 1987: Scientific method - experiments and statistics
- machine learning and probabilistic reasoning
- From 1995: Intelligent Agents - rebirth of the vision of the whole agent
- From 2001: Big data
- From 2011: Vector Processors and Computing Power
- reinforcement learning
- deep learning
Applications
- Autonomous Ships: path planning/path selection (Ottar, Robin, etc.)
- also autonomous flying drones (Erlend)
- Medical Image Processing (Hans Georg, Kjell-Inge)
- Diagnosis, e.g. cancer detection
- Image segmentation to make 3D models of the patient
- Cosmology: detect dark matter (Hans Georg, Ben David)
- Information Security (previously - Hans Georg)
- detect clandestine communication (steganalysis)
- intrusion detection
Concerns
- Lethal autonomous weapons.
- Surveillance and persuasion.
- Biased decision making.
- Job redundancy.
- Safety-critical applications.
- Cybersecurity.
Norbert Wiener: machines which strive for human objectives, without a pre-programmed goal. [R&N:52]
Discussion could the machine take control?