Born 11 November, Rangoon, Burma.
Educated at Rugby School, UK.


Open Classical Scholarship to Balliol College, Oxford.

1942 - 1945

War service in the Foreign Office, Bletchley Park.


Pre-clinical medical school, Oxford.


Balliol College War Memorial Studentship.
M A (Oxon) in Human Anatomy and Physiology.


D Phil (Oxon) in Mammalian Genetics.

1952 - 1958

Research Associate, Department of Zoology, University of London.


Scientific Fellow of the Zoological Society of London.


Senior Lecturer in Surgical Science, University of Edinburgh.


Reader in Surgical Science, Edinburgh.


Author (jointly): An Introduction to Molecular Biology.


Founder and Director, Experimental Programming Unit, Edinburgh.
Royal Society Lecturer in the USSR.


Personal Chair of Machine Intelligence, Edinburgh.
Founder and first Chairman,
Department of Machine Intelligence and Perception, Edinburgh.


Author: Computer Programming for Schools:first steps in Algol,
Edinburgh: Oliver & Boyd (with A. Ortony and R.M. Burstall).


Fellow of the Royal Society of Edinburgh.
Co-founder and first Board Chairman,
Edinburgh University Centre for Industrial Consultancy and Liaison.


D Sc (Oxon) in Biological Sciences.
Fellow of the British Computer Society.


William Withering Lectures, University of Birmingham, UK.


Visiting Lecturer, USSR Academy of Sciences.

1974 - 1984

Director, Machine Intelligence Research Unit, Edinburgh.


Author: On Machine Intelligence (2nd ed. 1986).
G A Miller Lectures, University of Illinois, USA.

1975 - 1984

Chairman of the Board of Trustees, A M Turing Trust.


Herbert Spencer Lecture, Oxford University.


Samuel Wilks Memorial Lecture, Princeton University, USA.


Editor: Expert Systems and the Micro-electronic Age.


Founder, British Computer Society Specialist Group in Expert Systems.


Author: Machine Intelligence and Related Topics.
Editor: Introductory Readings in Expert Systems.
Evening Discourse, The Royal Institution, UK.


Editor (jointly): Intelligent Systems: the Unprecedented Opportunity.
G A Miller Lecture, University of Illinois, USA.


Author (jointly): The Creative Computer (US title: The Knowledge Machine).
Technology Lecture, The Royal Society of London, UK.
Professor Emeritus of Machine Intelligence, Edinburgh.
Founder and Director of Research, the Turing Institute, Glasgow, UK.


Visiting Lecturer, USSR Academy of Sciences.


Chief Scientist, the Turing Institute.


Pioneer Award of the International Embryo Transfer Society
(jointly, for work in the 1950's with Dr Anne McLaren).
Honorary Life Membership of the BCS Specialist Group on Expert Systems.


S L A Marshall Lecture, US Army Inst. for the Behav. and Social Sciences, USA.
Elected a Founding Fellow of the American Association for Artificial Intelligence.


Hon D Sc, National Council for Academic Awards, UK.


C C Garvin Lecture, Virginia Polytechnic Institute and State University, USA.
Hon D Sc, University of Salford, UK.


Editor (jointly). Machine Learning, Neural and Statistical Classification, Ellis Horwood Ltd.


Visitor, AI Applications Institute, University of Edinburgh, UK.
Founder and Treasurer of the Human-Computer Learning Foundation.
Associate Member of the Josef Stefan Institute of Slovenia.
Achievement Medal of the Institute of Electrical Engineers.


Feigenbaum Medal of the World Congress on Expert Systems.
Hon D Univ, University of Stirling, UK.


Hon. D.Sc. University of Aberdeen, UK.


Hon. D.Univ. University of York, UK.


Foreign Honorary Member, American Academy of Arts and Sciences.
IJCAI Award for Research Excellence.


Foreign Honorary Member, Slovenian Academy of Sciences.

Lifetime Achievement Award – British Computer Society Specialist Group on Artificial Intelligence.


Scientific publications:

Professor Michie's publications include four books and about 170 papers in experimental biology, AI and computing. He was Editor in Chief of the Machine Intelligence series (nineteen volumes since 1967).

Origins and directions of contributions to machine learning:

Donald Michie's computing and AI career developed in 1942 from codebreaking discussions with Alan Turing at Bletchley Park. Their friendship infected Michie with enthusiasm for the possibilities of machine intelligence, influencing his work to this day. His main contributions to wartime codebreaking became available only recently when the work was declassified. He conceived a modification and extension to the logic of the Colossus, the world's first high-speed electronic computer, improving discovery of the daily German code patterns from several person-weeks at best to a few person-hours, an achievement beyond the original goal of Colossus. This sparked a "crash programme" that yielded nine successively enhanced Colossus machines working round the clock by the war's end.

His innovations in computer learning had been sparked by discussions with Alan Turing during and immediately after the war. In his February 1947 lecture to the London Mathematical Society, delivered before the first stored-program machines were operational, Turing sketched the basic idea. He argued that a computer program, when such a thing existed, could be enabled to improve its performance through its own accumulating experience. Michie began his first experiments in 1960. His tic-tac-toe machine MENACE demonstrated the basic principle of a self-reinforcing learning mechanism. MENACE employed Michie's conceptually simple general-purpose learning algorithm BOXES which could also discover robust control strategies for the pole-and-cart problem, but was soon employed industrially to evolve strategies for automatic control, such as controlling a steel mill.

Michie's next move in computer learning was to develop and systematize techniques for inductively extracting machine-executable concepts from example data. Michie showed how to circumvent the "Feigenbaum bottleneck" in expert systems development - the elicitation through dialog of (often intuitive) human expertise, arduous and often near-impossible to articulate.

Stimulated by Michie's guidance at Stanford University, Quinlan showed the promise of his own ID3 algorithm for processing complex data. Its disadvantage was that it yielded massively unwieldy and incomprehensible decision-rules. Michie and his student Alen Shapiro overcame this with "structured induction", an interactive regime for generating machine-executable decision rules and configuring them into transparent concept-hierarchies.

Using this more user-oriented approach, Michie and his students repeatedly reduced previously intractable problem-domains (e.g. constructing large bodies of chess endgame knowledge previously unknown to grandmasters), including numerous industrial problems. The resulting machine-generated solution-programs explained the rationale of their own decisions, providing more comprehensive, more precise and deeper answers than those from human experts. Notable examples were:

  1. a 1984 rule-based program for controlling a uranium refining plant for Westinghouse Research, in Pennsylvania, USA, improving yield-efficiency from less than 85% to about 95%. The resulting annual savings were in excess of $10 million, and
  2. with Sammut, using desktop aircraft simulators to demonstrate the automatic synthesis of skilled and versatile autopilots by machine imitation of laboratory-trained student pilots.

In 1968 Michie had already encapsulated rote-learning as a generic library facility, drawing together the seemingly disconnected principles and techniques of machine learning, and embedding them in the defined procedures of a programming language. With these "memo functions" , also referred to as "function caching", Michie made novel use of the fact that every computable function has both a tabular and an algorithmic representation. Under the memo-function regime, executable function procedures automatically and incrementally speeded themselves up by storing the results of functions for later reuse, rather than recomputing them. Memo functions were added to the library of the POP-2 programming system and later worked into POP-2 itself and into LISP. Thus self-optimising capabilities were made available to programmers for any integer-values function procedure that they cared to implement.

Memo functions were also employed to crucial effect in the early 1970s Edinburgh University robot FREDDY, the world's first versatile and teachable assembly robot, FREDDY. Hardware speeds of that time fell short by orders of magnitude of those required for the robot's perceive-interpret-plan-act-cycle. Incorporation of memo-functions in all time-critical functions conferred the needed speedup. Some specific capabilities of this 1973 hardware-software device have yet to be surpassed, in particular the ability to sort out a disorderly heap of parts and re-assemble them to build a specified object. Rapid re-instruction by a human operator in the visual recognition of objects in its task environment was made possible by means of a "teaching by showing" technique.

Prof. Donald Michie