Obituary: Peter DonaldsonA great opportunity in Peter Donaldson's life came in 1967 when he was invited by Giles Brindley to be the Chief Engineer in a new research group the Medical Research Council, Neurological Prostheses Unit. Brindley had impressed the medical establishment while, as an academic at Cambridge, making and implanting into a blind woman a device which stimulated the visual cortex, giving her sensations "like stars"; the first step toward prosthetic devices for artificial vision. This was at a time when pacemakers, with one stimulating electrode, were new and not reliable, so Brindley's 80-channel device was enormously ambitious but suggested what might be possible for the disabled. However, Brindley's device also showed the difficulty: it deteriorated so that little remained functional after a few months. At that time, there was no theory about how to make micro-electronic implants so that they would have an adequate service life. Indeed, it was not known whether it was possible. Some thought that the components should be encapsulated in a polymer of very low permeability "to keep the water out". Before he could design an improved visual prosthesis, Donaldson had to find a suitable encapsulant. He tried many materials and in those days his lab was full of jars of saline in which encapsulated test circuits remained until they failed. He noticed that sometimes failure occurred because the water penetrated along the interface with the wires, which passed through the encapsulant, rather than by diffusion through it. The breakthrough came, by chance, when he noticed at home that on a kettle that (for some reason) he had mended using silicone bath caulk, the silicone remained bonded to the metal despite the wet and the frequent heating. He tried silicone encapsulants for the tests and found a great improvement over the much more impermeable materials like waxes and epoxies. Making sense of this apparent paradox became one of his great contributions. He went on to produce a series of papers, describing elegant experiments, which were mostly published in Medical & Biological Engineering & Computing, in which he dealt with various aspects of silicone encapsulation, such as the important physical properties of the encapsulant, the chemical properties of the encapsulated parts, osmosis, and so on. Donaldson was born in 1927. His father was a submariner and he followed, entering Dartmouth Royal Naval College, age 13, early in the war. However, he had developed a deep interest in technical subjects at an early age, joined the Electrical Branch, and was sent to study Engineering at Cambridge. Returning to the navy, married, he contracted TB and was invalidated from the service. Later he was to say that it was the best thing that could have happened. His wife's uncle, Professor Bryan Matthews, Head of the Cambridge Department of Physiology, took him on as an engineer to make apparatus. After the war, when such huge advances had been made in electronic engineering, there was enormous scope for its application to electrophysiology. He soon felt able to write a comprehensive book Electronic Apparatus for Biological Research (1958). He was interested too in other fields, with Professor Richard Gregory in optical instruments and with Professor Horace Barlow in learning machines. He made one of the first machines that learnt from a human: the task was to balance an inverted pendulum, an exemplary application that has been used widely since. His apparatus would seem very strange to modern eyes: it was entirely hardware and used domestic watt-hour meters for the multipliers. It was described in Medical & Biological Engineering (2, p. 393, 1964), the journal of which he was the editor from 1963-7. Brindley and Donaldson collected a small team in the Neurological Prostheses Unit, which was to be very stable until their retirements in 1992. Donaldson's work on encapsulation spanned the whole of this period but he also worked on packaging, electrodes, RF links, and other aspects of neuroprosthesis technology. There were also many applications. The visual prostheses, most ambitious, were first. The basis for their constructions were flexible caps to fit on the dome of the head, between the skull and the scalp, with electrode arrays on short cables which passed through the skull to the posterior surface of the cortex. To achieve adequate yield (fraction of the outputs that were functional) he made sub-units and tested those before inserting each into a pre-formed recess in the cap and joining them with platinum wires. This required months of careful work under a microsope inside a clean air cabinet. He made two visual prostheses that were implanted into volunteers and, although they were not useful to the blind volunteers for daily life, Brindley and Rushton produced important results for the applied physiology of vision. By the late 1970s, Brindley and Craggs were interested in what could be done with electrodes on the peripheral nerve roots and members of the Unit, including the late John Cooper and Tim Perkins, developed the Sacral Anterior Root Stimulator (SARS), a device which could be used to empty the bladder of paraplegics made incontinent by spinal cord injury. The principles of the encapsulation were understood during the design, and this device has been found to be very reliable despite its apparent simplicity. Over 2000 have now been implanted worldwide. Many other, more experimental devices were made by Donaldson, or under his guidance, and sometimes in collaboration with other research groups: e.g. for treating epilepsy; for artificial hearing for the deaf (before cochlear stimulators became the success that they have); and for standing and stepping for paraplegics. The simplicity of some devices, made using elementary electronic circuits, with components soldered together and then encapsulated in silicone adhesive, has the advantage that an implant can be made in a few days. This was important to Brindley whose modus operandi was to treat individual patients, who had been referred to him, if possible, and this meant that sometimes-novel devices were required with only a few days notice. In his retirement, Donaldson contributed to the Lumbar Anterior Root Stimulator project at UCL and helped the company that makes the SARS, Finetech (Medical) Ltd, in their successful application for a CE mark. He also improved his workshop at home, and there produced a number of unusual working models. He became interested in electrostatic machines and made models of a motor, attributed to Benjamin Franklin, and a generator (mechanical voltage doubler) of Nicolson, among other things. These he described in articles in the IEE journal, Science & Education. He also wrote thought-provoking articles on historical or fictional topics. His fascination with electrical technology went beyond the mere functional value: he was delighted by the extraordinariness of the physical world and intrigued by counter-intuitive facts. He was as happy developing illuminating theory as carrying out meticulous experiments or making apparatus.
Nick Donaldson Obituary: Peter DonaldsonIt is very unusual for a fundamental advance to be contributed to a field with which the author of that advance has no prior experience or further professional contact. Yet that is what happened with the publication in 1960 of Donaldson's remarkable paper "Error decorrelation: a technique for matching a class of functions." (Proceedings of III International Conference on Medical Electronics, pp. 173-178). Every specialist field develops its own terminology. Peter Donaldson's paper uses the language of control theory and mathematical statistics, and was motivated entirely independently of its crucial bearing on a set of questions, almost certainly quite unknown to him at the time, with which the developing new field of artificial intelligence (AI) was becoming pre-occupied. These concerned machine emulation of human processes of learning. At the time of Donaldson's paper the possibility that a machine, in the form of a computer program, might be so designed as to learn a complex skill from its own experience of a given task had just been reported in a symbolic task domain (playing the game of checkers) in Arthur Samuel's 1957 "Some studies of machine learning using the game of checkers" IBM J. Res. Dev., 3, 210-229. Samuel's paper is generally regarded as the foundational paper of the subject of Machine Learning. What Donaldson's demonstration added were two findings: (1) that machine learning of a complex task was possible for a real-time sensorimotor skill, and (2) that it could be achieved by a form of "learning by imitation" from the real-time behaviour of a human mentor skilled in the given control task. It was not until 1968 that Roger Chambers and I were able to repeat Samuels' feat, using Donaldson's real-time control problem, in a way that in AI jargon was "rule-based" and hence articulately describable by the system itself. In the following year we followed up with a rule-based solution to (2) above, in which the machine learned the same skill by imitating a skilled human pole-balancer. In the forty years of Machine Learning work that followed, Donaldson's pole-and-cart task became a standard benchmark problem for a wealth of variants and refinements on the theme of rule-based adaptive control. It was also used in computer simulation to study processes of human "re-inforcement learning", as recently reported, for example, in "Simulator-mediated acquisition of a dynamic control skill" by Jean Hayes Michie and Donald Michie (Proceedings of 6th IFAC Symposium on Automated Systems Based on Human Skill, Sept 1997). In the annals of Machine Learning, Donaldson's 1960 paper is assured of an enduring place as one of the earliest and most seminal contributions.
Professor Donald Michie
|