Last month the UK House of Lords Artificial Intelligence Committee published a report, ‘Artificial Intelligence, AI in the UK: ready, willing and able?’ In the report, which concluded that the UK is in a strong position to be a world leader in the development, use and ethics of artificial intelligence (AI), the Lords committee revisited the evaluation of the Alvey Programme in the 1980s, as a means of considering how the UK had previously managed AI technology.
University of Manchester Deputy President, Luke Georghiou, was one of the authors of the Alvey evaluation. He reflects on the lessons learnt from Alvey, and the prospects for the renewed focus on AI research today.
- AI is one of the flagship areas of research for national research strategies.
- The Alvey programme of the 1980s was conceived as a response to Japanese strengths and investment in computer science.
- The review of Alvey by the University of Manchester and the University of Sussex concluded that the programme had delivered mixed results.
- In order to develop a sustainable innovation ecosystem for AI today, it will be crucial to develop complementary initiatives in training and skills, substantial user-engagement, greater focus in key centres and institutes, and a systematic attention to ethical issues through a responsible research and innovation approach.
Artificial intelligence (AI) has emerged as a new competitive battleground in nations’ search for technological and economic advantage. Excitement about the pervasiveness of opportunities associated with machine learning, deep learning and other core concepts of AI has been matched by concerns about ethical dimensions and social consequences. National responses include initiatives such as the UK’s Turing Institute (named for Manchester’s iconic pioneer) and the Vector Institute in Toronto. Scenarios of an AI-dominated future are regularly presented but for those with long memories the hopes and the hype being expressed have a familiar ring. This historic echo was noted in the recent House of Lords Report on AI which took the unusual step of commissioning a full appendix on “Historic Government Policy on Artificial Intelligence in the United Kingdom”, focused on events of over 30 years ago.
For this reader it brought to life seven years, from 1983 to 1990, spent with colleagues from the Universities of Manchester and Sussex carrying out a real-time evaluation of the Alvey Programme for Advanced Information Technology – a £350m (approximately £940m at today’s value) industry-academic initiative that remains the largest of its kind in the UK. Alvey was conceived as a response to Japan’s 5th Generation Computer Programme, whose focus on parallel computing was seen to pose a threat to UK industry. (Alvey was not an acronym – it was named after the BT executive John Alvey who led the inquiry that led to its formation. He took no substantive part in the programme itself.) The UK national programme was a multi-dimensional initiative with sub-programmes on Software Engineering, Very Large Scale Integration (VLSI) in microelectronics, with a significant defence interest, Man-Machine Interface (gender-laden terminology that was rightly left behind in that decade) and Intelligent Knowledge-Based Systems – which for reasons we will discuss was a deliberately coined synonym for AI.
The Alvey Programme produced mixed results. Our evaluation judged that its technological objectives had largely been met. To some extent, its structural objectives were also achieved in that it grew a large community of researchers and formed a template for academic-industry cooperation that was replicated both nationally and in the EU’s archetypal ‘ESPRIT programme’. ESPRIT started shortly after Alvey and instituted the project structures that still run through the Framework Programmes, including the current Horizon 2020. Despite this, the overall conclusion of the evaluation was downbeat. The leading UK participating firms such GEC and Plessey were losing ground and embarking upon a series of mergers and acquisitions that in turn did little to halt their decline. A series of headlines and leaders picked up on the evaluation’s final report, with an editorial in The Independent concluding: ‘Our industrial policy can provide, especially in education and training, essential support for industry, but it should not try to conjure, from “a weak and fragmented UK ability”, a new pillar of our prosperity.’ (Editorial, The Independent, May 1991)
With the benefit of hindsight, some long-term benefits can be seen. Alvey’s focus on object-oriented programming was being implemented by Microsoft a quarter of a century later. One of its core projects, a large-scale demonstrator on Mobile Information Systems, from the military radio communications company Racal Electronics, was an important step towards the company’s ‘new cellular radio service’ launched in 1985 as Vodafone. However, Alvey had been boxed into the single mode of pre-competitive research by a Thatcherite aversion to any measures seen to be near market. In that era, it was remarkable that an intervention on this scale was supported at all. (Many wondered how Alvey had managed to pass Prime Ministerial scrutiny. Allegedly Sir Arnold Weinstock, Chief Executive of GEC had persuaded her to back the initiative against her natural instincts.)
As a result, little was done to complement it with what, today, would be seen as essential parts of an innovation ecosystem, including user engagement, enhancing the supply of trained people and fostering patient capital.
Much of the debate, then as now, was on Alvey’s engagement with AI. Since 1973 the field had been blighted by a highly negative review of its prospects by Sir James Lighthill, Lucasian Professor of Mathematics at Cambridge University. Lighthill saw real benefits as being at least 25 years away. By 1981 a group of leading AI academics, dubbed by Brian Oakley, the Alvey Programme Director, as the ‘artificial intelligentsia’, were ready to rehabilitate the subject. Wishing to avoid being tainted by the Lighthill review and the then current looser label of expert systems they came up with the name for IKBS – possibly one of the most obviously committee-designed terms ever coined. In turn, this was sub-divided into knowledge-based systems, image understanding and vision, speech and natural language, and logic programming. (System Architectures were also located in the IKBS division but were effectively a separate sub-programme.)
The evaluation acknowledged that the academic research base had been strengthened and that a new competence had been established in UK industry. Less encouragingly it reported that academic participants were particularly critical of their industrial partners. There was a dearth of firms willing or capable of exploiting the results. Contractual obligations were also a barrier with one project held up by an obligation to use UK hardware in preference to its US counterpart (described as a Trabant compared to a Ferrari). Staff recruitment and retention also hindered progress. A general lack of commercial follow-up left a sense of under-achievement and contributed to a second ‘winter’ albeit less explicitly than that induced by Lighthill.
In the current AI Spring there is reason to hope that the ‘seasonal’ cycle will not be as exaggerated even if there is an inevitable drawback from some of the hype. Crucially, there is a far higher engagement by business and much potential in the combination of machine learning, robotics and data analytics. With the Industrial Strategy presenting AI as one of its four Grand Challenges, the key will be to avoid the over-reach of Alvey in trying to progress innovation and industrial development solely with an R&D initiative. Crucial complementary initiatives in training and skills, substantial user-engagement, greater focus in key centres and institutes and systematic attention to ethical issues through a responsible research and innovation approach are all necessary components of a sustainable innovation ecosystem for AI.
