Interview with Professor Alan Bundy CBE

Professor Alan Bundy, CBE, was born in May 1947 in Isleworth, Middlesex.  He is one of three children. His father worked in the print industry on the Radio Times and his mother was a shorthand typist.

Alan attended the Ealing Road Junior School where he says he did not get on very well initially. He says: “I was what they call a late starter, so I was in the B stream for most of my time there and was only put in the A stream because the teacher decided that he would do some IQ tests and I came out rather well in it and they suddenly realised that they hadn’t recognised my talents.”

Alan failed his eleven-plus exam and went to Brentford Secondary Modern School followed by Heston Secondary Modern when his family moved to Heston in 1958. He adds: “I began to do better at that stage academically. I clearly shouldn’t have been in a secondary modern school. I used to regularly come first in the class for most subjects.”

Having gained six GCSEs, Alan was on a career path towards being a lathe operator until his uncle, a career’s advisor, stepped in suggested that Alan should move to Springgrove Grammar school. Here he studied technical drawing, applied and pure maths and physics at A level.

Alan went to the University of Leicester in 1965 where he went on to gain a first in maths followed by a PhD in mathematical logic related to Gödel’s incompleteness theorem.

His interest in maths started at junior school, he says: “I’d always been interested in mathematics and certainly through secondary school, I was interested in logic particularly, I’d always had an affection for logic, so when I did my first degree in pure maths at Leicester, I wanted then to go on and do a PhD and I wanted to do one in logic and fortunately the head of department there was a logician and so he took me on as a PhD student.”

After completing his PhD, Alan found himself not “terribly happy with the state of mathematics at that point, I felt that the motivation of what sort of mathematical topics you pursue was a bit arbitrary.” It was at this stage that he was drawn to computing.   He adds: “I was attracted to computing because I could see that this was an area where logic was applied and where there was a clear motivation, and particularly given my background in mathematics, looking at automated reasoning was an attractive proposition because I thought I could use my knowledge of mathematics and my interest in logic and I could put all those things together.”

Alan’s first encounter with a computer was at school where he had lessons in Fortran this was followed by use of a computer at Leicester. Alan says: “I did some sort of voluntary extracurricular work on that and at that stage it was all cards. One would have to type up these cards and they’d be sent in a deck to the computer for batch processing and then you would find there was an error in the second card and so on. It was very frustrating and what one was able to do was very simple.”

 Alan’s interest in logic continued to grow.  He says: “At university I got interested in building some little robots; Grey Walter like autonomous robots out of Lego and various electric motors and so on, and set it running around the department. One of my lecturers pointed me to the work at Stanford University on Shakey the Robot and STRIPS which I thought was very interesting; the fact by writing effectively sets of axioms, one could actually control a robot – that seemed a pretty exciting thing.  Not that I got into robotics, as it turned out, because I felt I was more in my comfort zone doing automated theorem proving.”

In 1971, having completed his PhD, Alan joined the University of Edinburgh’s Meta-Mathematics Unit as a Research Assistant, a role that he got after approaching Bernard Meltzer from Edinburgh who visited Leicester to give a logic seminar. Alan applied and joined Bernard’s group which included Bob Kowalski, Pat Hayes, Bob Boyer and J Moore, all of whom have gone on to make major contributions to artificial intelligence and theorem proving. He says: “I was very lucky to be in that community.”

Bernard ran the Metamathematics Unit, which was adjacent to, and very strongly connected with, the Department of Machine Intelligence and Perception, run by Donald Michie, Christopher Longuet-Higgins, and originally Richard Gregory.  The professors of both groups regularly argued which led to what Alan describes as “a lot of political trouble’ which was finally resolved in 1974 when the Department of Artificial Intelligence was created, and Bernard was made head of department.

Alan says: “I remember coming out of our first general meeting and Bernard saying to me, ‘Well that was boring’, and I said, ‘Yes, thank God’, because up to then general meetings had been riotous, it had been very unpleasant with all these professors arguing amongst themselves.”

With the change in departments, Alan was promoted to lecturer. He says: “I was lucky to get a lectureship.  I was only an RA, I hadn’t really intended to stay in Edinburgh, I expected that I would have to move on at some point, but getting the first lectureship that was issued in the Department of Artificial Intelligence, (another eventually went to Gordon Plotkin), that was very distinguished.” The department was small and consisted of Bernard Meltzer, Robin Popplestone, Rob Burstall and Jim Howe, plus research fellows and PhD students.

Bernard allowed his group to work freely within the scope of the department from which Alan says he learned a lot and he has subsequently employed in his own research groups. He explains: “There was a very open atmosphere and he gave us our heads. He didn’t have a tight programme that he wanted us to contribute to, within the scope of the remit of the group, he more or less, let us get on with our own thing and I think that was very good at quite a young age.  We were mostly in our twenties and so we developed that independence and free thinking at that point which we were able to rely on later. … I think it’s very important to give autonomy to people very early in their career. I often contrast it with the German system where you don’t really get the ability to run your own group until you’re about forty and I think by then probably things have got a bit stale.”

Alan adds that it comes down to selecting the right people at which Bernard was very good. He adds: “If you pick the right people they will rise to that occasion and they will make mistakes, but I think they will make something of that opportunity.”

After being promoted to lecturer in 1974, Alan went on to become a Reader and then Professor in the Department. In 1998 the computer science, artificial intelligence and cognitive science departments merged into the School of Informatics, and Alan became the first head of that school.

Commenting on the change of name to ‘informatics’, Alan says: “What was driving us academically was the feeling that computing was now pervasive, that it was influencing other disciplines, particularly our understanding of human systems and human society as well as having practical applications in building hi-tech kit. We didn’t want to call ourselves the department of computer science because we didn’t feel that that reflected the breadth of our vision. We were looking for a term which would not just mean we’re looking at artificial systems, but that we’re looking at natural systems from a computational viewpoint. The feeling was because there were subjects like health informatics and so on, demonstrating a variety of ways in which informatics was used to emphasise the application side, that informatics was a better name.”

The change was not easy for some people, but I was delighted to see that alliances formed that crossed the boundaries of the old departments. One was of people with a focus on teaching, but there were others. Alan adds: “I think the time was ripe for it and it was a very successful move.”

Asked about projects that he is particularly proud of, Alan says: “The work on inductive theorem, proving and particularly the rippling technique, looking at ways in which one can manipulate formulae to turn them from one shape into another so that they’re useful to you.  I think that was a major achievement.  I think some of my early work too; the first project that I got funded was the MECHO project where we were looking at mechanics problems stated in English, typically from the sort of A level syllabus, in which we translated the English into logical formulae and then extracted equations which we then solved to find the answer.  At the time it was extremely novel, nobody else was really working on that.  There had been some work on understanding mathematical English and solving problems, but I think we got deeper into it than anybody else had done previously. I’ve gone back to that area a bit, I’ve moved away from theorem proving and am now looking at the way that information is represented and particularly, at the moment, I’m interested in what happens when the representation is faulty and how one can automatically repair a faulty representation to make it more accurate.”

 Alan believes that questions of representation are one of the main ways that computing can contribute to the other sciences. He says: “If you’re going to automate anything or provide any kind of computational aid to other sciences then you’re going to have to try to represent their knowledge by using logic, for instance, and using computational tools is one way to look at that.  There’s also statistical tools that one can bring to bear and that’s currently the major fashion, but I think the future lies in combining statistical and logical approaches and I think that’s going to be the complete answer, I think there’s a role for more symbolic reasoning in combination and fortunately I think the field’s moving in that direction now.”

He continues: “There’s this lovely quote I got about the nature of science that it’s not just finding out new facts, it’s finding out new ways to think about the world. That’s really one of the things that’s driven my whole career, actually trying to think about new ways to think about the world, and I think computing has served as a very good vehicle for that. If you go back to the work that I did on rippling, it’s a way of thinking about manipulation of formulae at a much higher level, at the meta level, and thinking about the shapes that you see there and how you can convert one shape into another and the kind of shapes that you recognise are novel.  I think maybe a lot of mathematicians have an implicit intuitive idea, but computation forces you to make that explicit and that’s then very valuable because you can then not only make better theorem provers but you can teach better, because you have a handle on the concept that you’re trying to get over. It’s not just learn from me in an apprenticeship relationship, but here’s the concept you need to be consciously aware of in thinking about this problem.”

Alan has always worked at the cutting edge, and this has meant his work has largely been focused on more long-term strategy rather than closer to application and therefore he has not regularly worked with industry. He says: “I always want to discover new techniques and new ways of thinking about the world. Therefore, a lot of my work has been more strategic and long term and it’s not always been close to application.  That’s changed a bit recently, I’m actually working with a couple of industrial projects, a small start-up called Brainnwave was interested in the work on correcting faulty representations and I have another project which is drawing conclusions from the web, doing inference from web knowledge and at the moment Huawei are funding that.”

He points to the longer term investment in research by commercial organisations in the US which is different from the model in the UK and says: “I think there has always been that attitude in the States. If you think about the history of AI, companies like Xerox and IBM and so on have always taken a long view, and you can see that now with Google and their funding of DeepMind.  If you look at the ecology of research in Stanford, for instance, or in Palo Alto, you see traffic of people doing pretty much pure or strategic research moving between academia and industry and back again.  There hasn’t been that tradition in the UK, and I think we’ve suffered for that, I think that’s been a negative in UK industry in relationships with academia … It’s been more difficult in the UK unless you were doing stuff that was very near to market, which has never really interested me. I want to show that one can understand things in computational detail that would surprise people. Looking back, that’s what really drives me.  I describe it as viewing computing as art. One of the functions of art is to surprise you, to make you think about things that you previously didn’t even consider, and I think that’s what really drives me; to surprise people.”

During his time at Edinburgh Alan has seen the department grow and shrink depending on the appetite for AI. It began not long after he joined with the Lighthill report in the seventies which resulted in Edinburgh’s department shrinking to four people.

He says: “One thing that drives it is over-expectation. In the very earliest days people like Newell and Simon came up with this famous list of goals which they thought were going to be solved in ten years and was far too ambitious.  Interestingly, some of them have been solved more recently, an order of magnitude longer than they anticipated. What I’ve observed is it’s often not the academics now that come up with these over-expectations, it’s the user community, particularly from industry. Captains of industry get carried away by the potential for AI and completely exaggerate what is possible and then of course if they’re not met and then there’s a lot of disappointment.”

He continues: “People ask me could it happen again, and I think to some extent that it could. I think one of the dangers that I see with the AI successes is that people over-generalise from them. People’s knowledge of intelligent agents is from the animal kingdom, humans and other animals, and the animal kingdom generally is generalist.  We can deal with novel situations, we have a wide range of abilities, not necessarily highly expert, but we can deal with a wide range of different environments.  If you look at the successes of AI, for example, AlphaGo, it plays a brilliant world class game of Go, but it can’t do anything else, it can’t even move the pieces on the board, the human has to move the pieces.  Self-driving cars, very successful at driving cars, but they can’t play chess or Go, and they can’t answer general knowledge questions like Watson does. So, what we’re looking at is what I call idiot savants, they’re really brilliant in a very, very narrow area but are not able to deal with anything outside that area.”

Alan believes that what will make a difference this time if the passion for AI drops off is that the community has industrial successes to point to. Alan adds: “There have been applications of machine learning which have been transformative in medicine, in self-driving cars, in image recognition, speech recognition and so on. We now have those successes we can point to, so I don’t think there’s a danger that suddenly all the funding will be removed because people will still want to fund those successes, but there will be a period of disillusion when people realise the limitations of these systems and how easily they can be fooled.”

Alan believes that academia in the UK is in a very healthy place, particularly in computing. He says: “I think there are some very interesting research projects going on.” He believes that to some extent universities have become a victim of their own success as class numbers continue to rise but he would rather be in that situation than the opposite. Funding is also variable; he points to the fact that it has not kept up with the growth in the number of researchers across the globe.  He says: “I think we’re all finding it more difficult to get funding but there’s much more interest from industry now in funding.  My funding has moved pretty much from EPSRC to industry in the last few years and I think that’s not uncommon.” The threat of Brexit has also created uncertainty in both funding and around international staff.

He also points to the record number of spin-outs that have grown out of the School of Informatics at Edinburgh, more than in Cambridge, saying: “When I was head of the school in 2001, a Scottish government report said that a quarter of all the spin-outs [not just computer-orientated ones] in Scotland across all disciplines came from our school.”

 In preparing students for industry, Alan says: “It’s certainly a driver and that’s why so many people want to do data mining and data science and so on here, our machine learning classes are huge and also the robotics group, and that’s driven by student perception of where the jobs are, which I think is probably an accurate perception at the moment.”

He highlights that students from Edinburgh are in high demand in industry, with the best going to work in the City on financial algorithms or that used to be the case a couple of decades ago. He imagines it’s broader now, with so many companies wanting to use ML.  He adds: “I think generally it looks like we’re certainly preparing people for the jobs that are out there.  There always used to be a complaint that we didn’t prepare people with the tools that industry wanted, but that’s a very short-term view, because those tools will only be popular for a few years and then what you really need is students who are future-proofed by having a deep understanding of the field and the theory and so on so that they can learn new stuff.  I think a lot of people in industry didn’t used to appreciate that.  I don’t hear those complaints so much, so whether people are taking a broader view now, I don’t know, but I never thought that was a valid criticism; it was a criticism, but I didn’t think it was a valid one.”

Alan was awarded a CBE in 2012 for his services to computing science. He is a fellow of many societies including the Royal Society of Edinburgh (FRSE), the Royal Society (FRS), BCS, The Chartered Institute for IT (FBCS), the Institution of Electrical Engineers (FIET), the Royal Academy of Engineering (FREng), the Association for Computer Machinery (ACM).

Alan is a founding Fellow of the Fellowship for the Association for the Advancement of Artificial Intelligence (AAAI) and the European Coordinating Committee for Artificial Intelligence (ECCAI). He also helped establish the UK Computing Research Committee.

Interviewed By: Troy Astarte on the 18th February 2020 at the School of Informatics, University of Edinburgh

Transcribed By: Susan Nicholls

Abstracted By: Lynda Feeley