Artificial Intelligence and Machine Learning Sydney Carter, Thomas - PowerPoint PPT Presentation

Artificial Intelligence and Machine Learning Sydney Carter, Thomas Politopoulos, Tariq Jamal, Kian Ettehadieh

What is AI? Four schools of thought:

Acting Humanly The Turing Test - Natural language processing - Knowledge representation - Automated reasoning - Machine learning Total Turing Test adds: - Computer vision - Robotics

Thinking humanly Relies on cognitive science - Follow the same steps as humans - But how do humans think? Difference between performing a task well and performing it like a human.

Thinking rationally Thinking logically “Patterns for argument structures that always yielded correct conclusions when given correct premises” Logical notation: “Socrates is a man; all men are mortal; therefore, Socrates is mortal.” Obstacles: - Translating to logical notation - Are these premises correct? - Computational resources

Acting rationally Rational Agent: - Acts autonomously - In pursuit of the optimal outcome Inference: a conclusion reached on the basis of evidence and reasoning. - Only part of being rational - Quick vs Deliberate Limited rationality - Perfect rationality isn’t always feasible - Computational limits

The Foundations of AI

Philosophy ● Aristotle (384-322 BC) Logic ○ ● Ramon Lull (1315) ○ human made objects Thomas Hobbes (1588-1689) ● Leviathan (1651) ○ Artificial Animal ○ ● Rene Descartes (1596-1650) “if the mind is governed entirely by physical laws, then it has no more free will than a rock ○ “deciding” to fall toward the center of the earth” ○ Dualism, we are a body, and a soul (something outside of nature) ○ Materialism, the mind creates consciousness ● Logical Positivism Rationalism + Empiricism = (Logic + Observed Sensory Experiences) ○

Mathematics ● Logic ○ George Boole BOOLEAN logic ■ ○ Gottlob Frege Computation ● ○ Limits between Logic and Computation Probability ● Gerolamo Cardano, 16th Century ○

Economics ● Adam Smith (1723-1790) ○ “was the first to treat [economics] as a science” Leon Valrasse/ Frank Ramsey ● ○ UTILITY Decision Theory ● (probability + utility theory) ○

Neuroscience ● Paul Broca (1824-1880) ○ Study on Aphasia in patients with brain damage (1861) Singularity ●

Computer Engineering Hardware Software ww2 computer science ● ● ● ENIAC ● AI has recompensed + Power ● ● + Capacity - Price ●

Control Theory/ Cybernetics ● Ktesibios of Alexandria } ○ Water Clock (250 BC) (self-regulated feedback control systems) Cornelis Drebbel ● ○ Thermostat (17th Century) Norbert Wiener ● “the scientific study of control and communication in the animal and the machine” (1948) ○ ● Objective Function

Linguistics ● Noam Chomsky (b. 1928) ○ Syntactic Structures (1957) Goes against Behaviourist theory ■ ● Natural Language Processing ○ context and understanding of subject matter ● Knowledge Representation

History of AI ● The Gestation of AI (1943-1955) First recognizable work done by Warren McCulloch and Walter Pitts ○ ■ Basic physiology and functions of neurons Analysis of propositional logic ■ ■ Turing’s theory of computation Sparked many further developments in early understandings of AI ○ ■ SNARC, Turing Test The Birth of AI (1956) ● ○ John McCarthy organized AI workshop at Dartmouth ■ Minsky, Claude Shannon, Nathaniel Rochester ■ No major breakthroughs ■ Established AI as it’s own separate field from the other computer sciences

The History of AI (cont.) ● Early Enthusiasms (1952-1969) People were excited by the notion of computers solving nonlinear problems ○ ○ General Problem Solver, Newell and Simon First program to embody ‘thinking humanly’ ■ ○ Sparked development of more ‘intelligent’ programs, and AI culture Physical Symbol System (Newell and Simon) ■ ■ Geometry Theorem Prover (Herbert Gelernter) Checkers playing program (Arthur Samuel) ■ ■ LISP (McCarthy, early dominant AI programming language) However, despite all the enthusiasm, AI did not develop as quickly or impactfully as was ○ predicted

The History of AI (cont.) ● Developments in AI (1969-1979) Success in principal = practical failure ○ ■ AI systems of this time were unable to interpret variable such as ambiguity As the problems AI was intended to solve grew more complex, original theories tested ■ on simplistic problems proved not to work Many governments and educational institutions stopped or decreased funding for AI research ○ and development The DENDRAL program, Buchanan 1969, offered counter approach to the ‘weak method’ ○ problem solving of AI machines of the time First specific knowledge intensive system ■ ■ Resurfaced hope for the AI and intelligent machine industry

The History of AI (cont.) ● AI becomes an industry (1980-present) The first commercial AI system was R1, Mcdermott ○ ■ Digital Equipment Corporation, helped configure orders for new computer systems Saved to company $40million and year by 1986 ■ ■ By 1988, other companies jumped on board, incorporating AI to their systems AI industry boomed from a few million dollars in 1980 to billions of dollars in 1988 ○

The History of AI (cont.) ● Further notable developments AI adopts the scientific method ○ ■ AI has come to embrace its scientific counterparts Influenced by fields such as neuroscience, psychology, computer science ● ■ It has established itself as its own scientific field which is continuously evolving The emergence of intelligent agents ○ ■ AI has reached a level of sophistication where intelligent machines and programs have developed a sense of agency ● Siri and Alexa, Facebook and Instagram algorithms

Quick note: Common AI Analogies (Hebron) Biological systems The brain and body systems as inspiration and as a model: “a central metric ● in evaluating machine intelligence since the inception of the field.” Thermodynamics ● Balancing a multiplicity of factors to reach equilibrium Electrical systems ● If neural communication is made up of electrical signals, can we create intelligent circuits? Problem: neural networks don’t work with switches and paths, they are more ● like peer-to-peer sharing; each neuron has a piece of the larger whole.

“What can AI do today?” - Russell et al Robotic Vehicles Speech Recognition and translation ● DARPA’s ‘Stanley’: object detection ● Microsoft’s speech recognition AI + human reaction data recently hit a 5.1% margin of error. ● NVIDIA’s AI driverless car uses ● Language translation human-driven routes as data to learn from Autonomous logistics & planning AI is often most effective when dealing ★ with large complex problems and LOTS of ● NASA’s Remote Agent and accurate data*** MAPGEN, planning daily Usually, more data > smarter AI ★ operations for spacecraft, Mars rovers ● Military planning (Iraq War)

More examples! Siri, Cortana, Alexa, Bixby Chat bots ● Speech recognition, app integration (shopping, GPS, internet searches), basic scheduling and communication Facebook, Twitter, Netflix, Spotify etc. ● Behavioural analysis and content delivery Healthcare ● Personal care assistance, diagnostics ● Cybernetic limbs require learning human nerve inputs

More examples! Gaming ● Legitimate AI ○ Human vs. Computer board games (IBM chess CPU) ● The AI effect ○ Devs attempt to create the illusion that the game’s characters/opponents are thinking ○ Positive user experience doesn’t require realistic AI ● The Uncanny Valley ○ Similar to the uncanny valley of CGI, extreme fidelity can be unsettling to players

QUESTIONS!~ ● What sort of task or problem do you think would benefit from an AI application? What might be some advantages., risks, precedents etc? If a 100% accurate AI system existed, would there be any tasks that you would ● not allow it to handle? ● Consider some of media representations of AI in areas such as film, advertising, literature and digital tech. How do you think they have influenced our popular conception of AI?