ECM2423 – Artificial Intelligence and Applications Lecture Notes
Defining
- A field of research in the design and study of computer systems that behave intelligently.
- Focusses on:
o Non-trivial problems requiring reasoning.
o Beyond numerical computations and manipulations.
- Benefits:
o Engineering aspect – solving hard problems.
o Cognitive aspect – understanding nature of human intelligence.
- Application to games well – clear, well-defined rules and no changing environment.
- Belief AI program only as intelligent as those programming them.
- Some applications:
o Physical assistants
o Space exploration
o Route planning
o Web search
o Speech/handwriting recognition
o Machine translation/NLP
Thinking/Acting Humanly/Rationally
Retain key functionalities when going to artificial equivalent – mimics, functionality retained, some
thing hard to change – emotions, depending on what retained – different fields, difficult to translate
humans.
- Acting humanly – artificial intelligence that can act human.
o Turing test: if response of computer to unrestricted textual natural-language
conversation cannot be distinguished from human being then computer can be said
to be intelligent.
o Eliza – an early chatbot.
o Loebner Prize: contest for restricted form of Turing test.
- Thinking humanly – cognitive modelling
,ECM2423 – Artificial Intelligence and Applications Lecture Notes
o Cognitive science: Philosophy, AI, neural science, psychology, linguistics,
anthropology, trying to create computational theories of human cognition.
o Machines must exhibit behaviour sufficient to fool human judge + must do in way
demonstrably analogous to human cognition (way it thinks).
o Requires detailed matching of computer behaviour + timing to detailed
measurements of human subjects in psychological experiments.
- Thinking rationally – laws of thought
o Formalise right reasoning using mathematical model (deductive reasoning).
o Logicist school: encoding knowledge in formal logical statements + use
mathematical deduction to perform reasoning.
o Problems:
▪ Formalizing common sense knowledge is difficult.
▪ General deductive inference is computationally intractable.
- Acting rationally – rational agents
o Agent: entity perceives its environment + able to execute actions to change.
o Agents have inherent goals to achieve (e.g., survive).
o Rational agent acts in way to maximise achievement of goals.
o True maximisation of goals requires omniscience + unlimited computational abilities.
o Limited rationality involves maximising goals within computational + information
available.
Foundations of AI
- Philosophers – AI conceivable by considering ideas that mind is like a machine + operates on
knowledge encoded in internal language + that can be used to choose what actions to take,
treating thoughts as logic.
- Mathematicians – formalisation of deductive reasoning, tools to manipulate statements of
logical certainty/un, probabilistic statements + set groundwork for understanding
computation + reasoning about algorithms.
- Economists – rationality + max outcome, formalised problem of making decisions that
maximise expected outcome to decision maker.
- Neuroscientists – discovered some facts about how brain works + connected + ways in
similar + different to computers, e.g., linking to neural networks.
- Psychologists – adopted idea that humans + animals can be considered information
processing machines, linguists showed language use fits into model.
- Computer engineers – provided more powerful machines making AI applications possible.
History
- AI is a young field 60+ years, cycles of success misplaced optimism, resulting cutbacks in
enthusiasm and funding, cycling of introducing new creative approaches + systematically
refining best ones.
- Early ai focused on reasoning, later scientific method.
, ECM2423 – Artificial Intelligence and Applications Lecture Notes
- 1956: birth of AI, workshop at Dartmouth college (John McCarthy, Marvin Minsky, Claude
Shannon), name AI created along with foundational principle:
o Every aspect of learning or any other feature of intelligence can be so precisely
described that a machine can be made to simulate it.
- 1952: Samuel’s Checkers program learned to play at a strong amateur level.
- 1955: Newell & Simon’s Logic Theorist proved theorems in Principia Mathematica (people
believed that their theorems were more elegant than human solved).
- Micro-worlds – original focus, limited unchanging environment, limited objects, limited
interactions, search space very small, understanding ambiguous words.
- Lisp invented, previously people coded in 1s and 0s.
- Great expectations for AI, but had underwhelming results:
o Natural language processing failed to consider context, translation word by word
does not work.
- 1966: ALPAC report cut off government funding for MT.
-
- Implications of early era:
- Problems:
o Limited computation – search space grew exponentially, outpacing hardware.
o Limited information – complexity of AI problems (number of words, objects,
concepts in the world), humans are not perfectly efficient, however make best use
of limited knowledge.
- Contributions:
o Lisp, garbage collection, time-sharing (John McCarthy).
o Key paradigm: separate modelling (declarative), + algorithms (procedural).
- Knowledge-based systems/expert-based systems (70-80s):
o Eliciting specific domain knowledge form experts in the form of rules, put into
computer using inference engine, using knowledge has to derive new knowledge.
o If [premises] then [conclusion].
o DENDRAL: infer molecular structure from mass spectrometry.
o MYCIN: diagnose blood infections, recommend antibiotics.
o XCON: convert customer orders into parts specification, save DEC $40 million a year
by 1986.
- Contributions:
o First real applications in industry.
o Knowledge helped curb exponential growth.
- Problems:
o Knowledge is not deterministic rules, needs to model uncertainty.
Defining
- A field of research in the design and study of computer systems that behave intelligently.
- Focusses on:
o Non-trivial problems requiring reasoning.
o Beyond numerical computations and manipulations.
- Benefits:
o Engineering aspect – solving hard problems.
o Cognitive aspect – understanding nature of human intelligence.
- Application to games well – clear, well-defined rules and no changing environment.
- Belief AI program only as intelligent as those programming them.
- Some applications:
o Physical assistants
o Space exploration
o Route planning
o Web search
o Speech/handwriting recognition
o Machine translation/NLP
Thinking/Acting Humanly/Rationally
Retain key functionalities when going to artificial equivalent – mimics, functionality retained, some
thing hard to change – emotions, depending on what retained – different fields, difficult to translate
humans.
- Acting humanly – artificial intelligence that can act human.
o Turing test: if response of computer to unrestricted textual natural-language
conversation cannot be distinguished from human being then computer can be said
to be intelligent.
o Eliza – an early chatbot.
o Loebner Prize: contest for restricted form of Turing test.
- Thinking humanly – cognitive modelling
,ECM2423 – Artificial Intelligence and Applications Lecture Notes
o Cognitive science: Philosophy, AI, neural science, psychology, linguistics,
anthropology, trying to create computational theories of human cognition.
o Machines must exhibit behaviour sufficient to fool human judge + must do in way
demonstrably analogous to human cognition (way it thinks).
o Requires detailed matching of computer behaviour + timing to detailed
measurements of human subjects in psychological experiments.
- Thinking rationally – laws of thought
o Formalise right reasoning using mathematical model (deductive reasoning).
o Logicist school: encoding knowledge in formal logical statements + use
mathematical deduction to perform reasoning.
o Problems:
▪ Formalizing common sense knowledge is difficult.
▪ General deductive inference is computationally intractable.
- Acting rationally – rational agents
o Agent: entity perceives its environment + able to execute actions to change.
o Agents have inherent goals to achieve (e.g., survive).
o Rational agent acts in way to maximise achievement of goals.
o True maximisation of goals requires omniscience + unlimited computational abilities.
o Limited rationality involves maximising goals within computational + information
available.
Foundations of AI
- Philosophers – AI conceivable by considering ideas that mind is like a machine + operates on
knowledge encoded in internal language + that can be used to choose what actions to take,
treating thoughts as logic.
- Mathematicians – formalisation of deductive reasoning, tools to manipulate statements of
logical certainty/un, probabilistic statements + set groundwork for understanding
computation + reasoning about algorithms.
- Economists – rationality + max outcome, formalised problem of making decisions that
maximise expected outcome to decision maker.
- Neuroscientists – discovered some facts about how brain works + connected + ways in
similar + different to computers, e.g., linking to neural networks.
- Psychologists – adopted idea that humans + animals can be considered information
processing machines, linguists showed language use fits into model.
- Computer engineers – provided more powerful machines making AI applications possible.
History
- AI is a young field 60+ years, cycles of success misplaced optimism, resulting cutbacks in
enthusiasm and funding, cycling of introducing new creative approaches + systematically
refining best ones.
- Early ai focused on reasoning, later scientific method.
, ECM2423 – Artificial Intelligence and Applications Lecture Notes
- 1956: birth of AI, workshop at Dartmouth college (John McCarthy, Marvin Minsky, Claude
Shannon), name AI created along with foundational principle:
o Every aspect of learning or any other feature of intelligence can be so precisely
described that a machine can be made to simulate it.
- 1952: Samuel’s Checkers program learned to play at a strong amateur level.
- 1955: Newell & Simon’s Logic Theorist proved theorems in Principia Mathematica (people
believed that their theorems were more elegant than human solved).
- Micro-worlds – original focus, limited unchanging environment, limited objects, limited
interactions, search space very small, understanding ambiguous words.
- Lisp invented, previously people coded in 1s and 0s.
- Great expectations for AI, but had underwhelming results:
o Natural language processing failed to consider context, translation word by word
does not work.
- 1966: ALPAC report cut off government funding for MT.
-
- Implications of early era:
- Problems:
o Limited computation – search space grew exponentially, outpacing hardware.
o Limited information – complexity of AI problems (number of words, objects,
concepts in the world), humans are not perfectly efficient, however make best use
of limited knowledge.
- Contributions:
o Lisp, garbage collection, time-sharing (John McCarthy).
o Key paradigm: separate modelling (declarative), + algorithms (procedural).
- Knowledge-based systems/expert-based systems (70-80s):
o Eliciting specific domain knowledge form experts in the form of rules, put into
computer using inference engine, using knowledge has to derive new knowledge.
o If [premises] then [conclusion].
o DENDRAL: infer molecular structure from mass spectrometry.
o MYCIN: diagnose blood infections, recommend antibiotics.
o XCON: convert customer orders into parts specification, save DEC $40 million a year
by 1986.
- Contributions:
o First real applications in industry.
o Knowledge helped curb exponential growth.
- Problems:
o Knowledge is not deterministic rules, needs to model uncertainty.
