How much can artificial intelligence and machine learning accelerate polymer science?

I’ve been at the annual High Polymer Research Group meeting at Pott Shrigley this week; this year it had the very timely theme “Polymers in the age of data”. Some great talks have really brought home to me both the promise of machine learning and laboratory automation in polymer science, as well as some of the practical barriers. Given the general interest in accelerated materials discovery using artificial intelligence, it’s interesting to focus on this specific class of materials to get a sense of the promise – and the pitfalls – of these techniques.

Debra Audis, from the USA’s National Institute of Standards and Technology, started the meeting off with a great talk on how to use machine learning to make predictions of polymer properties given information about molecular structure. She described three difficulties for machine learning – availability of enough reliable data, the problem of extrapolation outside the parameter space of the training set, and the problem of explainability.

A striking feature of Debra’s talk for me was its exploration of the interaction between old-fashioned theory, and new-fangled machine learning (ML). This goes in two directions – on the one hand, Debra demonstrated that incorporating knowledge from theory can greatly speed up the training of a ML model, as well as improving its ability to extrapolate beyond the training set. But given a trained ML model – essentially a black box of weights for your neural network, Debra emphasised the value of symbolic regression to convert the black box to a closed form expression of simple functional forms of the kind a theorist would hope to be able to derive from some physical principles, providing something a scientist might recognise as an explanation of the regularities that the machine learning model encapsulates.

But any machine learning model needs data – lots of data – so where does that data come from? One answer is to look at the records of experiments done in the past – the huge corpus of experimental data contained within the scientific literature. Jacqui Cole from Cambridge has developed software to extract numerical data, chemical reaction schemes, and to analyse images from the scientific data. For specific classes of (non-polymeric) materials she’s been able to create data sets with thousands of entries, using automated natural language processing to extract some of the contextual information that makes the data useful. Jacqui conceded that polymeric materials are particularly challenging for this approach; they have complex properties that are difficult to pin down to a single number, and what to the outsider may seem to be a single material (polyethylene for example) may actually be a category that encompasses molecules with a wider variety of subtle variations arising from different synthesis methods and reaction conditions. And Debra and Jacqui shared some sighs of exasperation at the horribly inconsistent naming conventions used by polymer science researchers.

My suspicion on this (informed a little by the outcomes of a large scale collaboration with a multinational materials company that I’ve been part of over the last five years) is that the limitations of existing data sets mean that the full potential of machine learning will only be unlocked by the production of new, large scale datasets designed specifically for the problem in hand. For most functional materials the parameter space to be explored is vast and multidimensional, so considerable thought needs to be given to how best to sample this parameter space to provide the training data that a good machine learning model needs. In some circumstances theory can help here – Kim Jelfs from Imperial described an approach where the outputs from very sophisticated, compute intensive theoretical models were used to train a ML model that could then interpolate properties at much lower compute cost. But we will always need to connect to the physical world and make some stuff.

This means we will need automated chemical synthesis – the ability to synthesise many different materials with systematic variation of the reactants and reaction conditions, and then rapidly determine the properties of this library of materials. How do you automate a synthetic chemistry lab? Currently, a synthesis laboratory consists of a human measuring out materials, setting up the right reaction conditions, then analysing and purifying the products, finally determining their properties. There’s a fundamental choice here – you can automate the glassware, or automate the researcher. In the UK, Lee Cronin at Glasgow (not at the meeting) has been a pioneer of the former approach, while Andy Cooper at Liverpool has championed the latter. Andy’s approach involves using commercial industrial robots to carry out the tasks a human researcher would do, while using minimally adapted synthesis and analytical equipment. His argument in favour of this approach is essentially an economic one – the world market for general purpose industrial robots is huge, leading to substantial falls in price, while custom built automated chemistry labs represent a smaller market, so one should expect slower progress and higher prices.

Some aspects of automating the equipment are already commercially available. Automatic liquid handling systems are widely available, allowing one, for example to pipette reactants into multiwell plates, so if one’s synthesis isn’t sensitive to air one can use this approach to do combinatorial chemistry. Adam Gormley from Rutgers described this approach for making a library of copolymers by an oxygen-tolerant adaptation of reversible addition−fragmentation chain-transfer polymerisation (RAFT), to produce libraries of copolymers with varying polymer molecular weight and composition. Another approach uses flow chemistry, in which reactions take place not in a fixed piece of glassware, but as the solvents containing the reactants travel down pipes, as described by Tanja Junkers from Monash, and Nick Warren from Leeds. This approach allows in-line reaction monitoring, so it’s possible to build in a feedback loop, adjusting the ingredients and reaction conditions on the fly in response to what is being produced.

It seems to me, as a non-chemist, that there is still a lot of specific work to be done to adapt the automation approach to any particular synthetic method, so we are still some way from a universal synthesis machine. Andy Cooper’s talk title perhaps alluded to this: “The mobile robotic polymer chemist: nice, but does it do RAFT?” This may be a chemist’s joke.

But whatever approach one has realised to be able to produce a library of molecules with different characteristics, and analyse their properties, there remains the question of how to sample what is likely to be a huge parameter space in order to provide the most effective training set for machine learning. We were reminded by the odd heckle from a very distinguished industrial scientist in the audience that there is a very classical body of theory to underpin this kind of experimental strategy – the Design of Experiments methodology. In these approaches, one selects the optimum set of different parameters in order most effectively to span parameter space.

But an automated laboratory offers the possibility of adapting the sampling strategy in response to the results as one gets them. Kim Jelfs set out the possible approaches very clearly. You can take the brute force approach, and just calculate everything – but this is usually prohibitively expensive in compute. You can use an evolutionary algorithm, using mutation and crossover steps to find a way through parameter space that optimises the output. Bayesian optimisation is popular, and generative models can be useful for taking a few more random leaps. Whatever the details, there needs to be a balance between optimisation and exploration – between taking a good formulation and making it better, and searching widely across parameter space for a possibly unexpected set of conditions that provides a step-change in the properties one is looking for.

It’s this combination of automated chemical synthesis and analysis, with algorithms for directing a search through parameter space, that some people call a “self-driving lab”. I think the progress we’re seeing now suggests that this isn’t an unrealistic aspiration. My somewhat tentative conclusions from all this:

  • We’re still a long way from an automated lab that can flexibly handle many different types of chemistry, so for a while its going to be a question of designing specific set-ups for particular synthetic problems (though of course there will be a lot of transferrable learning).
  • There is still lot of craft in designing algorithms to search parameter space effectively.
  • Theory still has its uses, both in accelerating the training of machine learning models, and in providing satisfactory explanations of their output.
  • It’s going to take significant effort, computing resource and money to develop these methods further, so it’s going to be important to select use cases where the value of an optimised molecule makes the investment worthwhile. Amongst the applications discussed in the meeting were drug excipients, membranes for gas separation, fuel cells and batteries, optoelectronic polymers.
  • Finally, the physical world matters – there’s value in the existing scientific literature, but it’s not going to be enough just to process words and text; for artificial intelligence to fulfil its promise for accelerating materials discovery you need to make stuff and test its properties.

Implications of Rachel Reeves’s Mais Lecture for Science & Innovation Policy

There will be a general election in the UK this year, and it is not impossible (to say the least) that the Labour opposition will form the next government. What might such a government’s policies imply for science and innovation policy? There are some important clues in a recent, lengthy speech – the 2024 Mais Lecture – given by the Shadow Chancellor of the Exchequer, Rachel Reeves, in which she sets out her economic priors.

In the speech, Reeves sets out in her view, the underlying problems of the UK economy – slow productivity growth leading to wage stagnation, low investment levels, poor skills (especially intermediate and technical) and “vast regional disparities, with all of England’s biggest cities outside London having productivity levels below the national average”. I think this analysis is now approaching being a consensus view – see, for example, this recent publication – The Productivity Agenda – from The Productivity Institute.

Interestingly, Reeves resists the temptation to blame everything on the current government, stressing that this situation reflects long-standing weaknesses, which began in the early 1990’s, which were not sufficiently challenged by the Labour governments of the late 90’s and 00’s, and then were made much worse in the 2010’s by Austerity, Brexit, and post-pandemic policy instability. Singling out Conservative Chancellor of the Exchequer Nigel Lawson as the author of policies that were both wrong in principle and badly executed, she identifies this period as the root of “an unprecedented surge in inequality between places and people which endures today. The decline or disappearance of whole industries, leaving enduring social and economic costs and hollowing out our industrial strength. And – crucially – diminishing returns for growth and productivity.”

To add to our problems, Reeves stresses that the external environment the UK now faces is much more challenging than in previous decades, with geopolitical instability reviving the basic question of national security, uncertainties from new technologies like AI, and the challenges of climate instability and the net zero energy transition. She is blunt in saying “globalisation, as we once knew it, is dead” – “a growth model reliant on geopolitical stability is a growth model resting on increasingly shallow foundations.”

What comes next? For Reeves, the new questions are “how Britain can pay its way in the world; of our productive capacity; of how to drive innovation and diffusion throughout our economy; of the regional distribution of work and opportunity; of how to mobilise investment, develop skills and tackle inefficiencies to modernise a sclerotic economy; and of energy security”, and the answers are to be found what economist Dani Rodrik calls “productivism”.

In practise, this means an industrial strategy which, recognising the limits of central government’s information and capacity to act, works in partnership. This needs to have both a sector focus – building on the UK’s existing areas of comparative advantage and its strategic needs – and a regional focus, working with local and regional government to support the development of clusters and the realisation of agglomeration benefits.

In terms of the mechanics of the approach, Reeves anticipates that this central mission of government – restoring economic growth – will be driven from the Treasury, through a a beefed up “Enterprise and Growth” unit. To realise these ambitions, she identifies three areas of focus – recreating macroeconomic stability, investment – particularly in partnership with the private sector, and reform – of the planning system, housing, skills, the labour market and regional governance.

Innovation is a central part of Reeves’s vision for increased investment, partly through the familiar call for more capital to flow to university spin-outs. But there is also a call for more focus on the diffusion of new technologies across the whole economy, including what Reeves has long called the “everyday economy”. In my view, this is correct, but will need new institutions, or the adaptation of existing ones (as I argued, with Eoin O’Sullivan: “What’s missing in the UK’s R&D landscape – institutions to build innovation capacity”). There is a very sensible commitment to a ten year funding cycle for R&D institutions, essential not least because some confidence in the longevity of programmes is essential to give the private sector the confidence to co-invest.

This was quite a dense speech, and the commentary around it – including the pre-briefing from Labour – was particularly misleading. I think it would be a mistake to underestimate how much of a break it represents from the conventional economic wisdom of the past three decades, though the details of the policy programme remain to be filled in, and, as many have commented, its implementation in a very tough fiscal environment is going to be challenging. Our current R&D landscape isn’t ideally configured to support these aspirations and the UK’s current challenges (as I argue in my long piece “Science and innovation policy for hard times: an overview of the UK’s Research and Development landscape”); I’d anticipate some reshaping to support the “missions” that are intended to give some structure to the Labour programme. And, as Reeves says unequivocally, of these missions, the goal of restoring productivity and economic growth is foundational.

Optical fibres and the paradox of innovation

Here is one of the foundational papers for the modern world – in effect, reporting the invention of optical fibres. Without optical fibres, there would be no internet, no on-demand video – and no globalisation, in the form we know it, with the highly dispersed supply chains that cheap and reliable information transmission between nations and continents that optical fibres make possible. This won a Nobel Prize for Charles Kao, a HK Chinese scientist then working in STL in Essex, a now defunct corporate laboratory.

Optical fibres are made of glass – so, ultimately, they come from sand – as Ed Conway’s excellent recent book, “Material World” explains. To make optical fibres a practical proposition needed lots of materials science to make glass pure enough to be transparent over huge distances. Much of this was done by Corning in the USA.

Who benefitted from optical fibres? The value of optical fibres to the world economy isn’t fully captured by their monetary value. Like all manufactured goods, productivity gains have driven their price down to almost negligible levels.

At the moment, the whole world is being wired with optical fibres, connecting people, offices, factories to superfast broadband. Yet, the the world trade in optical fibres is worth just $11 bn, less than 0.05% of total world trade. This is characteristic of that most misunderstood phenomenon in economics, Baumol’s so-called “cost disease”.

New inventions successively transform the economy, while innovation makes their price fall so far that, ultimately, in money terms they are barely detectable in GDP figures. Nonetheless,society benefits from innovations, taken for granted through ubiquity & low cost. (An earlier blog post of mine illustrates how Baumol’s “cost disease” works through a toy model)

To have continued economic growth, we need to have repeated cycles of invention & innovation like this. 30 years ago, corporate labs like STL were the driving force behind innovations like these. What happened to them?

Standard Telecommunication Laboratories in Harlow was the corporate lab of STC, Standard Telephones and Cables, a subsidiary of ITT, with a long history of innovation in electronics, telephony, radio coms & TV broadcasting in the UK. After a brief period of independence from 1982, STC was bought by Nortel, Canadian descendent of the North American Bell System. Nortel needed a massive restructuring after late 90’s internet bubble, & went bankrupt in 2009. The STL labs were demolished & are now a business park

The demise of Standard Communication Laboratories just one example of the slow death of UK corporate laboratories through the 90’s & 00’s, driven by changing norms in corporate governance and growing short-termism. These were well described in the 2012 Kay review of UK Equity Markets and Long-Term Decision Making. This has led, in my opinion, to a huge weakening of the UK’s innovation capacity, whose economic effects are now becoming apparent.

Deep decarbonisation is still a huge challenge

In 2019 I wrote a blogpost called The challenge of deep decarbonisation, stressing the scale of the economic and technological transition implied by a transition to net zero by 2050. I think the piece bears re-reading, but I wanted to update the numbers to see how much progress we had made in 4 years (the piece used the statistics for 2018; the most up-to-date current figures are for 2022). Of course, in the intervening four years we have had a pandemic and global energy price spike.

The headline figure is that the fossil fuel share of our primary consumption has fallen, but not by much. In 2018, 79.8% of our energy came from oil, gas and coal. In 2022, this share was 77.8%.

There is good news – if we look solely at electrical power generation, generation from hydro, wind and solar was up 32% 2018-2022, from 75 TWh to 99 TWh. Now 30.5% of our electricity production comes from renewables (excluding biomass, which I will come to later).

The less good news is that electrical power generation from nuclear is down 27%, from 65 TWh to 48 TWh, and this now represents just 14.7% of our electricity production. The increase in wind & solar is a real achievement – but it is largely offset by the decline in nuclear power production. This is the entirely predictable result of the AGR fleet reaching the end of its life, and the slow-motion debacle of the new nuclear build program.

The UK had 5.9 GW of nominal nuclear generation capacity in 2022. Of this, all but Sizewell B (1.2 GW) will close by 2030. In the early 2010’s, 17 GW of new nuclear capacity was planned – with the potential to produce more than 140 TWh per year. But, of these ambitious plans, the only project that is currently proceeding is Hinkley Point, late and over budget. The best we can hope for is that in 2030 we’ll have Hinkley’s 3.2 GW, which together with Sizewell B’s continuing operation could produce at best 38 TWh a year.

In 2022, another 36 TWh of electrical power – 11% – came from thermal renewables – largely burning imported wood chips. This supports a claim that more than half (56%) of our electricity is currently low carbon. It’s not clear, though, that imported biomass is truly sustainable or scaleable.

It’s easy to focus on electrical power generation. But – and this can’t be stressed too much – most of the energy we use is in the form of directly burnt gas (to heat our homes) and oil (to propel our cars and lorries).

The total primary energy we used in 2022 was 2055 TWh; and of this 1600 TWh was oil, gas and coal. 280 TWh (mostly gas) was converted into electricity (to produce 133 TWh of electricity), and 60 TWh’s worth of fossil fuel (mostly oil) was diverted into non-energy uses – mostly feedstocks for the petrochemical industry – leaving 1260 TWh to be directly burnt.

To achieve our net-zero target, we need to stop burning gas and oil, and instead use electricity. This implies a considerable increase in the amount of electricity we generate – and this increase all needs to come from low-carbon sources. There is good news, though – thanks to the second law of thermodynamics, we can convert electricity more efficiently into useful work than we can by burning fuels. So the increase in electrical generation capacity in principle can be a lot less than this 1260 TWh per year.

Projecting energy demand into the future is uncertain. On the one hand, we can rely on continuing improvements in energy efficiency from incremental technological advances; on the other, new demands on electrical power are likely to emerge (the huge energy hunger of the data centres needed to implement artificial intelligence being one example). To illustrate the scale of the problem, let’s consider the orders of magnitude involved in converting the current major uses of directly burnt fossil fuels to electrical power.

In 2022, 554 TWh of oil were used, in the form of petrol and diesel, to propel our cars and lorries. We do use some electricity directly for transport – currently just 8.4 TWh. A little of this is for trains (and, of course, we should long ago have electrified all intercity and suburban lines), but the biggest growth is for battery electrical vehicles. Internal combustion engines are heat engines, whose efficiency is limited by Carnot, whereas electric motors can in principle convert all inputted electrical energy into useful work. Very roughly, to replace the energy demands of current cars and lorries with electric vehicles would need another 165 TWh/year of electrical power.

The other major application of directly burnt fossil fuels is for heating houses and offices. This used 334 TWh/year in 2022, mostly in the form of natural gas. It’s increasingly clear that the most effective way of decarbonising this sector is through the installation of heat pumps. A heat pump is essentially a refrigerator run backwards, cooling the outside air or ground, and heating up the interior. Here the second law of thermodynamics is on our side; one ends up with more heat out than energy put in, because rather than directly converting electricity into heat, one is using it to move heat from one place to another.

Using a reasonable guess for the attainable, seasonally adjusted “coefficient of performance” for heat pumps, one might be able to achieve the same heating effect as we currently get from gas boilers with another 100 TWh of low carbon electricity. This figure could be substantially reduced if we had a serious programme of insulating old houses and commercial buildings, and were serious about imposing modern energy efficiency standards for new ones.

So, as an order of magnitude, we probably need to roughly double our current electricity generation capacity from its current value of 320 TWh/year, to more than 600 TWh/year. This will take big increases in generation from wind and solar, currently running around 100 TWh/year. In addition to intermittent renewables, we need a significant fraction of firm power, which can always be relied on, whatever the state of wind and sunshine. Nuclear would be my favoured source for this, so that would need a big increase from the 40 TWh/year we’ll have in place by 2030. The alternative would be to continue to generate electricity from gas, but to capture and store the carbon dioxide produce. For why I think this is less desirable for power generation (though possibly necessary for some industrial processes), see my earlier piece: Carbon Capture and Storage: technically possible, but politically and economically a bad idea.

Industrial uses of energy, which currently amount to 266 TWh, are a mix of gas, electricity and some oil. Some of these applications (e.g. making cement and fertiliser) are going to be rather hard to electrify, so, in addition to requiring carbon capture and storage, this may provide a demand for hydrogen, produced from renewable electricity, or conceivably process heat from high temperature nuclear reactors.

It’s also important to remember that a true reckoning of our national contribution to climate change would include taking account of the carbon dioxide produced in the goods and commodities we import, and our share of air travel. This is very significant, though hard to quantify – in my 2019 piece, I estimated that this could add as much as 60% to our personal carbon budget.

To conclude, we know what we have to do:

  • Electrify everything we can (heat pumps for houses, electric cars), and reduce demand where possible (especially by insulating houses and offices);
  • Use green hydrogen for energy intensive industry & hard to electrify sectors;
  • Hugely increase zero carbon electrical generation, through a mix of wind, solar and nuclear.

In each case, we’re going to need innovation, focused on reducing cost and increasing scale.

There’s a long way to go!

All figures are taken from the UK Government’s Digest of UK Energy Statistics, with some simplification and rounding.

The shifting sands of UK Government technology prioritisation

In the last decade, the UK has had four significantly different sets of technology priorities, and a short, but disruptive, period, where such prioritisation was opposed on principle. This 3500 word piece looks at this history of instability in UK innovation policy, and suggests some principles of consistency and clarity which might give us some more stability in the decade to come. A PDF version can be downloaded here.

Introduction

The problem of policy churn has been identified in a number of policy areas as a barrier to productivity growth in the UK, and science and innovation policy is no exception to this. The UK can’t do everything – it represents less than 3% of the world’s R&D resources, so it needs to specialise. But recent governments have not found it easy to decide where the UK should put its focus, and then stick to those decisions.

In 2012 this the then Science Minister, David Willetts, launched an initiative which identified 8 priority technologies – the “Eight Great Technologies”. Willetts reflected on the fate of this initiative in a very interesting paper published last year. In short, while there has been consensus on the need for the UK to focus (with the exception of one short period), the areas of focus have been subject to frequent change.

Substantial changes in direction for technology policy have occurred despite the fact that we’ve had a single political party in power since 2010, with particular instability since 2015, in the period of Conservative majority government. Since 2012, the average life-span of an innovation policy has been about 2.5 years. Underneath the headline changes, it is true that there have been some continuities. But given the long time-scales needed to establish research programmes and to carry them through to their outcomes, this instability makes it different to implement any kind of coherent strategy.

Shifting Priorities: from “Eight Great Technologies”, through “Seven Technology Families”, to “Five Critical Technologies”

Table 1 summarises the various priority technologies identified in government policy since 2012, grouped in a way which best brings out the continuities (click to enlarge).

The “Eight Great Technologies” were introduced in 2012 in a speech to the Royal Society by the then Chancellor of the Exchequer, George Osborne; a paper by David Willetts expanded on the rationale for the choices . The 2014 Science and Innovation Policy endorsed the “Eight Great Technologies”, with the addition of quantum technology, which, following an extensive lobbying exercise, had been added to the list in the 2013 Autumn Statement.

2015 brought a majority Conservative government, but continuity in the offices of Prime Minister and Chancellor of the Exchequer didn’t translate into continuity in innovation policy. The new Secretary of State in the Department of Business, Innovation and Skills was Sajid Javid, who brought to the post a Thatcherite distrust of anything that smacked of industrial strategy. The main victim of this world-view was the innovation agency Innovate UK, which was subjected to significant cut-backs, causing lasting damage.

This interlude didn’t last very long – after the Brexit referendum, David Cameron’s resignation and the premiership of Theresa May, there was an increased appetite for intervention in the economy, coupled with a growing consciousness and acknowledgement of the UK’s productivity problem. Greg Clark (a former Science Minister) took over at a renamed and expanded Department of Business, Energy and Industrial Strategy.

A White Paper outlining a “modern industrial strategy” was published in 2017. Although it nodded to the “Eight Great Technologies”, the focus shifted to four “missions”. Money had already been set aside in the 2016 Autumn Statement for an “Industrial Strategy Challenge Fund” which would support R&D in support of the priorities that emerged from the Industrial Strategy.

2019 saw another change of Prime Minister – and another election, which brought another Conservative government, with a much greater majority, and a rather interventionist manifesto that promised substantial increases in science funding, including a new agency modelled on the USA’s ARPA, and a promise to “focus our efforts on areas where the UK can generate a commanding lead in the industries of the future – life sciences, clean energy, space, design, computing, robotics and artificial intelligence.”

But the “modern industrial strategy” didn’t survive long into the new administration. The new Secretary of State was Kwasi Kwarteng, from the wing of the party with an ideological aversion to industrial strategy. In 2021, the industrial strategy was superseded by a Treasury document, the Plan for Growth, which, while placing strong emphasis on the importance of innovation, took a much more sector and technology agnostic approach to its support. The Plan for Growth was supported by a new Innovation Strategy, published later in 2021. This did identify a new set of priority technologies – “Seven Technology Families”.

2022 was the year of three Prime Ministers. Liz Truss’s hard-line free market position was certainly unfriendly to the concept of industrial strategy, but in her 44 day tenure as Prime Minister there was not enough time to make any significant changes in direction to innovation policy.

Rishi Sunak’s Premiership brought another significant development, in the form of a machinery of government change reflecting the new Prime Minister’s enthusiasm for technology. A new department – the Department for Innovation, Science and Technology – meant that there was now a cabinet level Secretary of State focused on science. Another significant evolution in the profile of science and technology in government was the increasing prominence of national security as a driver of science policy.

This had begun in the 2021 Integrated Review , which was an attempt to set a single vision for the UK’s place in the world, covering security, defence, development and foreign policy. This elevated “Sustaining strategic advantage through science and technology” as one of four overarching principles. The disruptions to international supply chains during the covid pandemic, and the 2022 invasion of Ukraine by Russia and the subsequent large scale European land war, raised the issue of national security even higher up the political agenda.

A new department, and a modified set of priorities, produced a new 2023 strategy – the Science & Technology Framework – taking a systems approach to UK science & technology . This included a new set of technology priorities – the “Five critical technologies”.

Thus in a single decade, we’ve had four significantly different sets of technology priorities, and a short, but disruptive, period, where such prioritisation was opposed on principle.

Continuities and discontinuities

There are some continuities in substance in these technology priorities. Quantum technology appeared around 2013 as an addendum to the “Eight Great Technologies”, and survives into the current “Five Critical Technologies”. Issues of national security are a big driver here, as they are for much larger scale programmes in the USA and China.

In a couple of other areas, name changes conceal substantial continuity. What was called synthetic biology in 2012 is now encompassed in the field of engineering biology. Artificial Intelligence has come to high public prominence today, but it is a natural evolution of what used to be called big data, driven by technical advances in machine learning, more computer power, and bigger data sets.

Priorities in 2017 were defined as Grand Challenges, not Technologies. The language of challenges is taken up in the 2021 Innovation Strategy, which proposes a suite of Innovation Missions, distinct from the priority technology families, to address major societal challenges, in areas such as climate change, public health, and intractable diseases. The 2023 Science and Technology Framework, however, describes investments in three of the Five Critical Technologies, engineering biology, artificial intelligence, and quantum technologies, as “technology missions”, which seems to use the term in a somewhat different sense. There is room for more clarity about what is meant by a grand challenge, a mission, or a technology, which I will return to below.

Another distinction that is not always clear is between technologies and industry sectors. Both the Coalition and the May governments had industrial strategies that explicitly singled out particular sectors for support, including through support for innovation. These are listed in table 2. But it is arguable that at least two of the Eight Great Technologies – agritech, and space & satellites – would be better thought of as industry sectors rather than technologies.

Table 2 – industrial strategy sectors, as defined by the Coalition, and the May government.

The sector approach did underpin major applied public/private R&D programmes (such as the Aerospace Technology Institute, and the Advanced Propulsion Centre), and new R&D institutions, such as the Offshore Renewable Catapult Centre, designed to support specific industry sectors. Meanwhile, under the banner of Life Sciences, there is continued explicit support from the pharmaceutical and biotech industry, though here there is a lack of clarity about whether the primary goal is to promote the health of citizens through innovation support to the health and social care system, or to support pharma and biotech as high value, exporting, industrial sectors.

But two of the 2023 “five critical technologies” – semiconductors and future telecoms – are substantially new. Again, these look more like industrial sectors than technologies, and while no one can doubt their strategic importance in the global economy it isn’t obvious that the UK has a particularly strong comparative advantage in them, either in the size of the existing business base or the scale of the UK market (see my earlier discussion of the background to a UK Semiconductor Strategy).

The story of the last ten years, then, is a lack of consistency, not just in the priorities themselves, but in the conceptual basis for making the prioritisation – whether challenges or missions, industry sectors, or technologies.

From strategy to implementation

How does one turn from strategy to implementation: given a set of priority sectors, what needs to happen to turn these into research programmes, and then translate that research into commercial outcomes? An obvious point that nonetheless needs stressing, is that this process has long lead times, and this isn’t compatible with innovation strategies that have an average lifetime of 2.5 years.

To quote the recent Willetts review of the business case process for scientific programmes: “One senior official estimated the time from an original idea, arising in Research Councils, to execution of a programme at over two and a half years with 13 specific approvals required.” It would obviously be desirable to cut some of the bureaucracy that causes such delays, but it is striking that the time taken to design and initiate a research programme is of the same order as the average lifetime of an innovation strategy.

One data point here is the fate of the Industrial Strategy Challenge Fund. This was announced in the 2016 Autumn Statement, anticipating the 2017 Industrial Strategy White Paper, and exists to support translational research programmes in support of that Industrial Strategy. As we have seen, this strategy was de-emphasised in 2019, and formally scrapped in 2021. Yet the research programmes set up to support it are still going, with money still in the budget to be spent in FY 24/25.

Of course, much worthwhile research will be being done in these programmes, so the money isn’t wasted; the problem is that such orphan programmes may not have any follow-up, as new programmes on different topics are designed to support the latest strategy to emerge from central government.

Sometimes the timescales are such that there isn’t even a chance to operationalise one strategy before another one arrives. The major public funder of R&D, UKRI, produced a five year strategy in March 2022 , which was underpinned by the seven technology families. To operationalise this strategy, UKRI’s constituent research councils produced a set of delivery plans . These were published in September 2022, giving them a run of six months before the arrival of the 2023 Science and Innovation Framework, with its new set of critical technologies.

A natural response of funding agencies to this instability would be to decide themselves what best to do, and then do their best to retro-fit their ongoing programmes to new government strategies as they emerge. But this would defeat the point of making a strategy in the first place.

The next ten years

How can we do better over the next decade? We need to focus on consistency and clarity.

Consistency means having one strategy that we stick to. If we have this, investors can have confidence in the UK, research institutions can make informed decisions about their own investments, and individual researchers can plan their careers with more confidence.

Of course, the strategy should evolve, as unexpected developments in science and technology appear, and as the external environment changes. And it should build on what has gone before – for example, there is much of value in the systems approach of the 2023 Science and Innovation Framework.

There should be clarity on the basis for prioritisation. I think it is important to be much clearer about what we mean by Grand Challenges, Missions, Industry Sectors, and Technologies, and how they differ from each other. With sharper definitions, we might find it easier to establish clear criteria for prioritisation.

For me, Grand Challenges establish the conditions we are operating under. Some grand challenges might include:

  • How to move our energy economy to a zero-carbon basis by 2050;
  • How to create an affordable and humane health and social care system for an ageing population;
  • How to restore productivity growth to the UK economy and reduce the UK’s regional disparities in economic performance;
  • How to keep the UK safe and secure in an increasingly unstable and hostile world.

One would hope that there was a wide consensus about the scale of these problems, though not everyone will agree, nor will it always be obvious, what the best way of tackling them is.

Some might refer to these overarching issues as missions, using the term popularised by Mariana Mazzacuto , but I would prefer to refer to a mission as something more specific, with a sense of timescale and a definite target. The 1960’s Moonshot programme is often taken as an exemplar, though I think the more significant mission from that period was to create the ability for the USA to land a half tonne payload anywhere on the earth’s surface, with an accuracy of a few hundred meters or better.

The crucial feature of a mission, then, is that it is a targeted program to achieve a strategic goal of the state, that requires both the integration and refinement of existing technologies and the development of new ones. Defining and prioritising missions requires working across the whole of government, to identify the problems that the state needs to be solved, and that are tractable enough given reasonable technology foresight to be worth trying, and prioritising them.

The key questions for a judging missions, then, are, how much does the government want this to happen, how feasible is it given foreseeable technology, how well equipped is the UK to deliver it given its industrial and research capabilities, and how affordable is it?

For supporting an industry sector, though, the questions are different. Sector support is part of an active industrial strategy, and given the tendency of industry sectors to cluster in space, this has a strong regional dimension. The goals of industrial strategy are largely economic – to raise the economic productivity of a region or the nation – so the criteria for selecting sectors should be based on their importance to the economy in terms of the fraction of GVA that they supply, and their potential to improve productivity.

In the past industrial strategy has often been driven by the need to create jobs, but our current problem is productivity, rather than unemployment, so I think the key criteria for selecting sectors should be their potential to create more value through the application of innovation and the development of skills in their workforces.

In addition to the economic dimension, there may also be a security aspect to the choice, if there is a reason to suppose that maintaining capability in a particular sector is vital to national security. The 2021 nationalisation of the steel forging company, Sheffield Forgemasters, to secure the capability to manufacture critical components for the Royal Navy’s submarine fleet, would have been unthinkable a decade ago.

Industrial strategy may involve support for innovation, for example through collaborative programmes of pre-competitive research. But it needs to be broader than just research and development; it may involve developing institutions and programmes for innovation diffusion, the harnessing of public procurement, the development of specialist skills provision, and at a regional level, the provision of infrastructure.

Finally, on what basis should we choose a technology to focus on? By a technology priority, we refer to an emerging capability arising from new science, that could be adopted by existing industry sectors, or could create new, disruptive sectors. Here an understanding of the international research landscape, and the UK’s part of that, is a crucial starting point. Even the newest technology, to be implemented, depends on existing industrial capability, so the shape of the existing UK industrial base does need to be taken account. Finally, one shouldn’t underplay the importance of the vision of talented and driven individuals.

This isn’t to say that priorities for the whole of the science and innovation landscape need to be defined in terms of challenges, missions, and industry sectors.
A general framework for skills, finance, regulation, international collaboration, and infrastructure – as set out by the recent Science & Innovation Framework – needs to underlie more specific prioritisation. Maintaining the health of the basic disciplines is important to provide resilience in the face of the unanticipated, and it is important to be open to new developments and maintain agility in responding to them.

The starting point for a science and innovation strategy should be to realise that, very often, science and innovation shouldn’t be the starting point. Science policy is not the same as industrial strategy, even though it’s often used as a (much cheaper) substitute for it. For challenges and missions, defining the goals must come first; only then can one decide what advances in science and technology are needed to bring those in reach. Likewise, in a successful industrial strategy, close engagement with the existing capabilities of industry and the demands of the market are needed to define the areas of science and innovation that will support the development of a particular industry sector.

As I stressed in my earlier, comprehensive, survey of the UK Research and Development landscape, underlying any lasting strategy needs to be a settled, long-term view of what kind of country the UK aspires to be, what kind of economy it should have, and how it sees its place in the world.

Science and Innovation in the 2023 Autumn Statement

On the 22nd November, the Government published its Autumn Statement. This piece, published in Research Professional under the title Economic clouds cast gloom over the UK’s ambitions for R&D, offers my somewhat gloomy perspective on the implications of the statement for science and innovation.

This government has always placed a strong rhetorical emphasis on the centrality of science and innovation in its plans for the nation, though with three different Prime Ministers, there’ve been some changes in emphasis.

This continues in the Autumn Statement: a whole section is devoted to “Supporting the UK’s scientists and innovators”, building on the March 2023 publication of a “UK Science and Technology Framework”, which recommitted to increasing total public spending on research to £20 billion in FY 2024/25. But before going into detail on the new science-related announcements in the Autumn Statement, let’s step back to look at the wider economic context in which innovation strategy is being made.

There are two giant clouds in the economic backdrop the Autumn Statement. One is inflation; the other is economic growth – or, to be more precise, the lack of it.

Inflation, in some senses, is good for governments. It allows them to raise taxes without the need for embarrassing announcements, as people’s cost-of-living wage rises take them into higher tax brackets. And by simply failing to raise budgets in line with inflation, public spending cuts can be imposed by default. But if it’s good for governments, it’s bad for politicians, because people notice rising prices, and they don’t like it. And the real effect of stealth public spending cuts do, nonetheless, materialise.

The effect of the inflation we’ve seen since 2021 is a rise in price levels of around 20%; while the inflation rate peak has surely passed, prices will continue to rise. We can already see the effect on the science budget. Back in 2021, the Comprehensive Spending Review announced a significant increase in the overall government research budget, from £15 billion to £20 billion in 24/25. By next year, though, the effect of inflation will have been to erode that increase in real terms, from £5 billion to less than £2 billion in 2021 money. The effect on Core Research is even more dramatic; in effect inflation will have almost totally wiped out the increase promised in 2021.

Our other problem is persistent slow economic growth, as I discussed here. The underlying cause of this is the dramatic decrease in productivity growth since the financial crisis of 2008. The consequence is the prospect of two full decades without any real growth in wages, and, for the government, the need to simultaneously increase the tax burden and squeeze public services in an attempt to stabilise public debt.

The detailed causes of the productivity slowdown are much debated, but the root of it seems to be the UK’s persistent lack of investment, both public and private (see The Productivity Agenda for a broad discussion). Relatively low levels of R&D are part of this. The most significant policy change in the Autumn Statement does recognise this – it is a tax break allowing companies to set the full cost of new plant and machinery against corporation tax. On the government side, though, the plans are essentially for overall flat capital spending – i.e., taking into account inflation, a real terms cut. Government R&D spending falls in this overall envelope, so is likely to be under pressure.

Instead, the government is putting their hopes on the private sector stepping up to fill the gap, with a continuing emphasis on measures such as R&D tax credits to incentivise private sector R&D, and reforms to the pension system – including the “Long-term Investment for Technology and Science (LIFTS)” initiative – to bring more private money into the research system. The ambition for the UK to be a “Science Superpower” remains, but the government would prefer not to have to pay for it.

One significant set of announcements – on the “Advanced Manufacturing Plan” – marks the next phase in the Conservatives’ off-again, on-again relationship with industrial strategy. Commitments to support advanced manufacturing sectors such as aerospace, automobiles and pharmaceuticals, as well as the “Made Smarter” programme for innovation diffusion, are very welcome. The sums themselves perhaps shouldn’t be taken too seriously; the current government can’t bind its successor, whatever its colour, and anyway this money will have to be found within the overall spending envelope produced by the next Comprehensive Spending Review. But it is very welcome that, after the split-up of the Department of Business, Energy and Industrial Strategy, that the successor Department of Business and International Trade still maintains an interest in research and innovation in support of mainstream business sectors, rather than assuming that is all now to be left to its sister Department of Science, Innovation and Technology.

For all the efforts to create a tax-cutting headline, the economic backdrop for this Autumn statement is truly grim. There is no rosy scenario for the research community to benefit from; the question we face instead is how to fulfil the promises we have been making that R&D can indeed lead to productivity growth and economic benefit.

Productivity and artificial intelligence

To scientists, machine learning is a relatively old technology. The last decade has seen considerable progress, both as a result of new techniques – back propagation & deep learning, and the transformers algorithm – and massive investment of private sector resources, especially computing power. The result has been the striking and hugely publicised success of large language models.

But this rapid progress poses a paradox – for all the technical advances over the last decade, the impact on productivity growth has been undetectable. The productivity stagnation that has been such a feature of the last decade and a half continues, with all the deleterious effects that produces in flat-lining living standards and challenging public finances. The situation is reminiscent of an earlier, 1987, comment by the economist Robert Solow: “You can see the computer age everywhere but in the productivity statistics.”

There are two possible resolutions of this new Solow paradox – one optimistic, one pessimistic. The pessimist’s view is that, in terms of innovation, the low-hanging fruit has already been taken. In this perspective – most famously stated by Robert Gordon – today’s innovations are actually less economically significant than innovations of previous eras. Compared to electricity, Fordist manufacturing systems, mass personal mobility, antibiotics, and telecoms, to give just a few examples, even artificial intelligence is only of second order significance.

To add further to the pessimism, there is a growing sense that the process of innovation itself is suffering from diminishing returns – in the words of a famous recent paper: “Are ideas getting harder to find?”.

The optimistic view, by contrast, is that the productivity gains will come, but they will take time. History tells us that economies need time to adapt to new general purpose technologies – infrastructures & business models need to be adapted, and the skills to use them need to be spread through the working population. This was the experience with the introduction of electricity to industrial processes – factories had been configured around the need to transmit mechanical power from central steam engines through elaborate systems of belts and pulleys to the individual machines, so it took time to introduce systems where each machine had its own electric motor, and the period of adaptation might even involve a temporary reduction in productivity. Hence, one might expect a new technology to follow a J-shaped curve.

Whether one is an optimist or a pessimist, there are a number of common research questions that the rise of artificial intelligence raises:

  • Are we measuring productivity right? How do we measure value in a world of fast moving technologies?
  • How do firms of different sizes adapt to new technologies like AI?
  • How important – and how rate-limiting – is the development of new business models in reaping the benefits of AI?
  • How do we drive productivity improvements in the public sector?
  • What will be the role of AI in health and social care?
  • How do national economies make system-wide transitions? When economies need to make simultaneous transitions – for example net zero and digitalisation – how do they interact?
  • What institutions are needed to support the faster and wider diffusion of new technologies like AI, & the development of the skills needed to implement them?
  • Given the UK’s economic imbalances, how can regional innovation systems be developed to increase absorptive capacity for new technologies like AI?

A finer-grained analysis of the origins of our productivity slowdown actually deepens the new Solow paradox. It turns out that the productivity slowdown has been most marked in the most tech-intensive sectors. In the UK, the most careful decomposition similarly finds that it’s the sectors normally thought of as most tech intensive that have contributed to the slowdown – transport equipment (i.e., automobiles and aerospace), pharmaceuticals, computer software and telecoms.

It’s worth looking in more detail at the case of pharmaceuticals to see how the promise of AI might play out. The decline in productivity of the pharmaceutical industry follows several decades in which, globally, the productivity of R&D – expressed as the number of new drugs brought to market per $billion of R&D – has been falling exponentially.

There’s no clearer signal of the promise of AI in the life sciences than the effective solution of one of the most important fundamental problems in biology – the protein folding problem – by Deepmind’s programme AlphaFold. Many proteins fold into a unique three dimensional structure, whose precise details determine its function – for example in catalysing chemical reactions. This three-dimensional structure is determined by the (one-dimensional) sequence of different amino acids along the protein chain. Given the sequence, can one predict the structure? This problem had resisted theoretical solution for decades, but AlphaFold, using deep learning to establish the correlations between sequence and many experimentally determined structures, can now predict unknown structures from sequence data with great accuracy and reliability.

Given this success in an important problem from biology, it’s natural to ask whether AI can be used to speed up the process of developing new drugs – and not surprising that this has prompted a rush of money from venture capitalists. One of the most high profile start-ups in the UK pursuing this is BenevolentAI, floated on the Amsterdam Euronext market in 2021 with €1.5 billion valuation.

Earlier this year, it was reported that BenevolentAI was laying off 180 staff after one of its drug candidates failed in phase 2 clinical trials. Its share price has plunged, and its market cap now stands at €90 million. I’ve no reason to think that BenevolentAI is anything but a well run company employing many excellent scientists, and I hope it recovers from these setbacks. But what lessons can be learnt from this disappointment? Given that AlphaFold was so successful, why has it been harder than expected to use AI to boost R&D productivity in the pharma industry?

Two factors made the success of AlphaFold possible. Firstly, the problem it was trying to solve was very well defined – given a certain linear sequence of amino acids, what is the three dimensional structure of the folded protein? Secondly, it had a huge corpus of well-curated public domain data to work on, in the form of experimentally determined protein structures, generated through decades of work in academia using x-ray diffraction and other techniques.

What’s been the problem in pharma? AI has been valuable in generating new drug candidates – for example, by identifying molecules that will fit into particular parts of a target protein molecule. But, according to pharma analyst Jack Scannell [1], it isn’t identifying candidate molecules that is the rate limiting step in drug development. Instead, the problem is the lack of screening techniques and disease models that have good predictive power.

The lesson here, then, is that AI is very good at the solving the problems that it is well adapted for – well posed problems, where there exist big and well-curated datasets that span the problem space. Its contribution to overall productivity growth, though, will depend on whether those AI-susceptible parts of the overall problem are in fact the rate-limiting steps.

So how is the situation changed by the massive impact of large language models? This new technology – “generative pre-trained transformers” – consists of text prediction models based on establishing statistical relationships between the words found in a massively multi-parameter regression over a very large corpus of text [3]. This has, in effect, automated the production of plausible, though derivative and not wholly reliable, prose.

Naturally, sectors for which this is the stock-in-trade feel threatened by this development. What’s absolutely clear is that this technology has essentially solved the problem of machine translation; it also raises some fascinating fundamental issues about the deep structure of language.

What areas of economic life will be most affected by large language models? It’s already clear that these tools can significantly speed up writing computer code. Any sector in which it is necessary to generate boiler-plate prose, in marketing, routine legal services, and management consultancy is likely to be affected. Similarly, the assimilation of large documents will be assisted by the capabilities of LLMs to provide synopses of complex texts.

What does the future hold? There is a very interesting discussion to be had, at the intersection of technology, biology and eschatology, about the prospects for “artificial general intelligence”, but I’m not going to take that on here, so I will focus on the near term.

We can expect further improvements in large language models. There will undoubtedly be improvements in efficiencies as techniques are refined and the fundamental understanding of how they work is improved. We’ll see more specialised training sets, that might improve the (currently somewhat shaky) reliability of the outputs.

There is one issue that might prove limiting. The rapid improvement we’ve seen in the performance of large language models has been driven by exponential increases in the amount of computer resource used to train the models, with empirical scaling laws emerging to allow extrapolations. The cost of training these models is now measured in $100 millions – with associated energy consumption starting to be a significant contribution to global carbon emissions. So it’s important to understand the extent to which the cost of computer resources will be a limiting factor on the further development of this technology.

As I’ve discussed before, the exponential increases in computer power given to us by Moore’s law, and the corresponding decreases in cost, began to slow in the mid-2000’s. A recent comprehensive study of the cost of computing by Diane Coyle and Lucy Hampton puts this in context [2]. This is summarised in the figure below:

The cost of computing with time. The solid lines represent best fits to a very extensive data set collected by Diane Coyle and Lucy Hampton; the figure is taken from their paper [2]; the annotations are my own.

The highly specialised integrated circuits that are used in huge numbers to train LLMs – such as the H100 graphics processing units designed by NVIdia and manufactured by TSMC that are the mainstay of the AI industry – are in a regime where performance improvements come less from the increasing transistor densities that gave us the golden age of Moore’s law, and more from incremental improvements in task-specific architecture design, together with simply multiplying the number of units.

For more than two millennia, human cultures in both east and west have used capabilities in language as a signal for wider abilities. So it’s not surprising that large language models have seized the imagination. But it’s important not to mistake the map for the territory.

Language and text are hugely important for how we organise and collaborate to collectively achieve common goals, and for the way we preserve, transmit and build on the sum of human knowledge and culture. So we shouldn’t underestimate the power of tools which facilitate that. But equally, many of the constraints we face require direct engagement with the physical world – whether that is through the need to get the better understanding of biology that will allow us to develop new medicines more effectively, or the ability to generate abundant zero carbon energy. This is where those other areas of machine learning – pattern recognition, finding relationships within large data sets – may have a bigger contribution.

Fluency with the written word is an important skill in itself, so the improvements in productivity that will come from the new technology of large language models will arise in places where speed in generating and assimilating prose are the rate limiting step in the process of producing economic value. For machine learning and artificial intelligence more widely, the rate at which productivity growth will be boosted will depend, not just on developments in the technology itself, but on the rate at which other technologies and other business processes are adapted to take advantage of AI.

I don’t think we can expect large language models, or AI in general, to be a magic bullet to instantly solve our productivity malaise. It’s a powerful new technology, but as for all new technologies, we have to find the places in our economic system where they can add the most value, and the system itself will take time to adapt, to take advantage of the possibilities the new technologies offer.

These notes are based on an informal talk I gave on behalf of the Productivity Institute. It benefitted a lot from discussions with Bart van Ark. The opinions, though, are entirely my own and I wouldn’t necessarily expect him to agree with me.

[1] J.W. Scannell, Eroom’s Law and the decline in the productivity of biopharmaceutical R&D,
in Artificial Intelligence in Science Challenges, Opportunities and the Future of Research.

[2] Diane Coyle & Lucy Hampton, Twenty-first century progress in computing.

[3] For a semi-technical account of how large language models work, I found this piece by Stephen Wolfram very helpful: What is ChatGPT doing … and why does it work?

Robert Cecil Jones (1932 – 2023)

My father, Robert Cecil Jones, died on Friday 8 September 2023. His 90 year life spanned a childhood and youth in Aberystwyth, 16 years service in the Royal Air Force, a spell as an adult educator in Birmingham and a second career as a priest in the Church in Wales. He is survived by his wife of 67 years, Sheila, and me, his only child, Richard.

Robbie’s mother, Gladys Jones (née Williams), was from Yspyty Ifan, in the Conwy valley; Len Jones was a journeyman butcher from Conwy. Robbie was born in Colwyn Bay in 1932, but the family soon moved to Aberystwyth, where Len carried out his trade for Dewhurst the Butchers until his death in 1971. The family lived without interruption in the same house on Dinas Terrace, where Gladys took in visitors for Bed and Breakfast.

Robbie was an only child, and for much of the war he was brought up by Gladys alone – Len was conscripted in the Royal Armoured Corps. He was a tank crewman who was captured at the Battle of Tobruk, and spent the rest of the war as a prisoner of war, first in Italy, then Germany. Robbie attributed his close bond with his mother (and, less plausibly, his love of ice cream) to this wartime period.

He was educated at Ardwyn Grammar School – there, in 1949, he first met Sheila Howell Lewis, the love of his life. They married in 1956. He did well at school (except at maths, which in those days didn’t seem to matter so much), and was offered a place at Oxford. He couldn’t afford to take that up, but instead studied geography at the University College of Wales Aberystwyth, living at home, alongside the other students Gladys took in as lodgers.

He’d talk about his university years with great affection – he was active in student politics, in amateur drama, in entertaining visiting dignitaries to the university (a rowdy night out with Dylan Thomas being something he recalled with particular pleasure). He was involved in many high spirited pranks at the expense of the town and university authorities. He’d recall these with all the more amusement for the fact that some of his co-conspirators went on to become pillars of the Welsh legal and political establishment.

Robbie always spoke with enormous gratitude for both his parents, who had had very little education themselves, but never hesitated in their support for his studies. After his BA, he trained as a teacher, and then started research for a Masters in historical geography, working with the great Welsh geographer Emrys Bowen, while supporting himself working as a journalist on the Cambrian News. But he concluded that the academic life wasn’t for him.

Faced with the need to do National Service, he chose to apply for a short-service commission in the RAF instead. After officer training in the Isle of Man, he was posted to the V-bomber base, RAF Wittering, near Stamford, moving there with Sheila. At the end of the commission, he and Sheila stayed in Stamford, where Sheila had a teaching job. Robbie got a job teaching in the same school. This didn’t really work out – it didn’t do a lot for his self-confidence that his wife had to intervene to maintain discipline in his classes. But it was in their short stay in Stamford that their only son, Richard arrived.

The solution was to rejoin the RAF, which accepted Robbie for a full 16 year commission. His first posting was back to the Isle of Man, turning from trainee to trainer at the Officer Training Unit at Jurby. This was a happy time for Robbie and Sheila, who made some lifelong friends there.

Robbie’s RAF service continued, with postings to Feltwell and Brampton in East Anglia. A less happy time followed in 1967, when Robbie was posted to Aden, to assist with the UK’s chaotic and violent withdrawal from this outpost of Empire. Aden, in the grip of a bitter insurgency, was too dangerous for families, so he had to leave his behind, to live in Pwllheli close to Sheila’s parents. Better times came with a move from Aden to Sharjah, in what’s now the United Arab Emirates; still without his family, but without Aden’s bloody conflict. There he enjoyed roving across the deserts and mountains of Arabia, training with the RAF’s Mountain and Desert Rescue Team.

Back in the UK, and back with the family, postings followed to Cosford, near Wolverhampton, and finally to Bicester. Then the time came to leave the RAF, and adjust to civilian life. He took up a position in adult education in Birmingham, a rewarding time.

Christianity was always important to Robbie. He’d considered entering the ministry as a young man, but had been discouraged by his mother (though she herself was always a devout churchgoer), and had followed the advice of an important mentor that he should spend some time in a secular career first. While he was in Birmingham, he became a lay reader. Finally he decided that the time was ripe for him to enter the ministry full time, and he was accepted as a trainee priest by the Diocese of St Davids, in the (Anglican) Church in Wales.

After 2 years in theological college in Birmingham, he returned to Aberystwyth as a curate in the beautiful and historic church of Llanbadarn Fawr. His mother Gladys was thrilled to have him back close, and he spoke with huge admiration and respect for the Vicar he served under and learnt so much from, Hywel Jones. With his return to Wales came a refreshing of his Welsh language. Welsh was Robbie’s mother tongue, but as was very common in those times, his education was exclusively in English. Expressing the subtleties of theology in Welsh was a challenge at first, but he soon found his feet for preaching in his native tongue. Whether in Welsh or English, some who knew his quiet-spoken everyday manner could be surprised by the passion of his public speaking – real hwyl.

His first parish was the Ceredigion coastal resort of New Quay/ Cei Newydd. Before long he was offered the opportunity to move to Newport/ Trefdraeth, in Pembrokeshire. This he gladly seized; Sheila was born and spent her early childhood in North Pembrokeshire, so this was a return to the family homeland for her. The position of Rector of Newport wholly suited Robbie, who took very seriously the idea that an Anglican priest should serve the whole of the community whenever there was a need, whether they identified as churchgoers or not. He could relate equally to the Welsh speaking rural communities of the Gwaun Valley, with their deep traditions, as to those who had retired to this charming seaside village. He was proud to be appointed to the traditional position of Mayor of Newport.

After retirement, Robbie and Sheila moved a few miles down the coast to Fishguard. But the arrival of grandchildren persuaded them to relocate once again, to Stoney Middleton, Derbyshire, to be closer to their son Richard, daughter-in-law Dulcie, and grandchildren Rosie and Thomas.

But retirement, for Anglican priests, is a relative concept, and Robbie was soon drawn in to frequently helping out with services in the local churches, supporting the overstretched rural vicar. He continued serving the community in this way well into his 80s, until deteriorating health finally forced him to retire for good.

Should Cambridge double in size?

The UK’s economic geography, outside London, is marked by small, prosperous cities in the south and east, and large, poor cities everywhere else. This leads to a dilemma for policy makers – should we try and make the small, successful, cities, bigger, or do the work needed to make our big cities more successful? The government’s emphasis seems to have swung back to expanding successful places in the South and East, with a particular focus on Cambridge.

Cambridge is undoubtedly a great success story for the UK, and potentially a huge national asset. Decades of investment by the state in research has resulted in an exemplary knowledge-based economy, where that investment in public R&D attracts in private sector R&D in even greater proportion. Cambridge has expanded recently, developing a substantial life science campus around the south of the city, moving engineering and physical sciences research to the West Cambridge site, and developing a cluster of digital businesses around the station. But its growth is constrained by poor infrastructure (water being a particular problem), aesthetic considerations in a historic city centre (which effectively rule out high rise buildings), and the political barriers posed by wealthy and influential communities who oppose growth.

We need an economic reality check too. How much economic difference would it make, on a national scale, if Cambridge did manage to double in size – and what are the alternatives? Here’s a very rough stab at some numbers.

The gross value added per person in Cambridge was £49,000 in 2018, well above the UK average of £29,000 [1]. In Greater Manchester, by contrast, GVA per person was about £25,000, well below the UK average. This illustrates the’s UK unusual and sub-optimal economic geography – in most countries, it’s the big cities that drive the economy. In contrast, in the UK, big second tier cities, like Manchester, Birmingham, Leeds and Glasgow, underperform economically and in effect drag the economy down.

Let’s do the thought experiment where we imagine Cambridge doubles its population, from 126,000 to 252,000, taking those people from Greater Manchester’s population of 2.8 million, and assuming that they are able to add the same average GVA per person to the Cambridge economy. Since the GVA per head in Cambridge is so much higher than in GM, this would raise national GVA by about £3 billion.

In the overall context of the UK’s economy, with a total GVA of £1,900 billion, £3 billion doesn’t make a material difference. The trouble with small cities is that they are small – so, no matter how successful economically they are, even doubling their size doesn’t make much of an impact at a national scale.

As an alternative to doubling the size of Cambridge, we could raise the productivity of Greater Manchester. To achieve a £3 billion increase in GM’s output, we’d need to raise the GVA per person by just over 4.2%, to a bit more than £26,000 – still below the UK average.

That’s the importance of trying to raise the productivity of big cities – they are big. Relatively marginal improvements in productivity in Greater Manchester, Leeds, Birmingham and the West Midlands, Sheffield, Glasgow and Cardiff could cumulatively start to make a material difference to the economy on a national scale. And we know where those improvements need to be made – for example in better public transport, more R&D and support for innovative businesses, providing the skills that innovative businesses need, by addressing poor housing and public health.

I do think Cambridge should be encouraged and supported to expand, to accommodate the private sector businesses that want to take advantage of the public investment in R&D that’s happened there, and to give the people they need to work for them somewhere affordable to live.

But, as Tom Forth and I have argued in detail elsewhere, we need more centres of R&D and innovation outside the Greater Southeast, particularly in those places where the private sector already makes big investments in R&D that aren’t supported by the public sector. The government has already made a commitment, in the Levelling Up White Paper, to increase public investment in R&D outside the Greater Southeast by a third by 2025. That commitment needs to be delivered, and built on by the next government.

Finally, we should ask ourselves whether we are fully exploiting the great assets that have been built in Cambridge, not just to support the economy of a small city in East Anglia, but to drive the economy of the whole nation. How could we make sure that if a Cambridge semiconductor spin-out is expanding, it builds its factory in Newport, Gwent, rather than Saxony or Hsinchu? How can we use the huge wealth of experience in the Cambridge venture capital community to support nascent VC sectors in places like Leeds? How could we make sure a Cambridge biotech spin-out does its clinical trials in Greater Manchester [2], and then then manufactures its medicine in Cheshire or on Merseyside?

Two things are needed to make this happen. Firstly, we need place-based industrial strategies to build the innovation, skills and manufacturing capacity in relevant sectors in other parts of the UK, so these places have the absorptive capacity to make the most of innovations emerging from Cambridge. Then, we need to build institutional links between the key organisations in Cambridge and those in other emerging regional centres. In this way, we could take full advantage of Cambridge’s position as a unique national asset.

[1]. Data here is taken from the ONS’s Regional Gross Value Added (balanced) dataset and mid-year population estimates, in both cases using 2018 data. The data for local authority areas on a workplace basis, but populations are for residents. This probably flatters the productivity number for Cambridge, as it doesn’t take account of people who live in neighbouring areas and commute into the city.

At another limit, one could ask what would happen if you doubled the population of the whole county of Cambridgeshire, 650,000. As the GVA per head at the county level is £31.5k, quite a lot less than the figure for Cambridge city, this makes surprisingly little difference to the overall result – this would increase GVA by £3.15 bn, the same as a 4.2% increase in GM’s productivity.

Of course, this poses another question – why the prosperity of Cambridge city doesn’t spill over very far into the rest of the county. Anyone who regularly uses the train from Cambridge via Ely and March to Peterborough might have a theory about that.

[2]. The recent government report on commercial clinical trials in the UK, by Lord O’Shaughnessy, highlighted a drop in patients enrolled in commercial clinical trials in the UK of 36% over the last six years. This national trend has been bucked in Greater Manchester, where there has been an increase of 19% in patient recruitment, driven by effective partnership between the NIHR Greater Manchester Clinical Research Network, the GM devolved health and social care system, industry and academia.

The UK’s crisis of economic growth

Everyone now agrees that the UK has a serious problem of economic growth – or lack of it – even if opinions differ about its causes, and what we should do about it. Here I’d like to set out the scale of the problem with plots of the key data.

My first plot shows real GDP since 1955. The break in the curve at the global financial crisis around 2007 is obvious. Before 2007 there were booms and busts – but the whole curve is well fit by a trend line representing 2.4% a year real growth. But after the 2008 recession, there was no return to the trend line. Growth was further interrupted by the covid pandemic, and the recovery from the pandemic has been slow. The UK’s GDP is now about 18% lower than it would have been if the economy had returned to its pre-recession trend line.


UK real GDP. Chained volume measure, base year 2019. ONS: 30 June 2023 release.

Total GDP is of particular interest to HM Treasury, as it is the overall size of the economy that determines the sustainability of the national debt. But you can grow an economy by increasing the size of the population, and, from the point of view of the sustainability of public services and a wider sense of prosperity, GDP per capita is a better measure.

My second plot shows real GDP capita. GDP per person has risen less fast than total GDP, both before and after the global financial crisis, reflecting the fact that the UK’s population has been growing. Trend growth before the break was 2.1% per annum; once again, contrary to all previous experience in the post-war period, per capita GDP growth has never recovered to the pre-crisis trend line. The gap with the previous trend, 25%, or £10,900 per person, is perhaps the best measure of the UK’s lost prosperity.


UK real GDP per capita. Chained volume measure, base year 2019. ONS: 12 May 2023 release.

The most fundamental measure of the productive capacity of the economy is, perhaps, labour productivity, defined as the GDP per hour worked. One can make GDP per capita grow by people working more hours, or by having more people enter the labour market. In the late 2010s, this was a significant factor in the growth of GDP per capita, but since the pandemic this effect has gone into reverse, with more people leaving the labour market, often due to long-term ill-health.

My third plot shows UK labour productivity. This shows the fundamental and obvious break in productivity performance that, in my view, underlies pretty much everything that’s wrong with the UK’s economy – and indeed its politics. As I discussed in more detail in my previous post,“When did the UK’s productivity slowdown begin?”, I increasingly suspect that this break predates the financial crisis – and indeed that crisis is probably better thought as an effect, rather than a cause, of a more fundamental downward shift in the UK’s capacity to generate economic growth.


UK labour productivity, whole economy. Chained volume measure, index (2019=100). ONS: 7 July 2023 release.

Talk of GDP growth and labour productivity may seem remote to many voters, but this economic stagnation has direct effects, not just on the affordability of public services, but on people’s wages. My final plot shows average weekly earnings, corrected for inflation. The picture is dismal – there has essentially been no rise in real wages for more than a decade. This, at root, is why the UK”s lack of economic growth is only going to grow in political salience.


UK Average weekly earnings, 2015 £s, corrected for inflation with CPI. ONS: 11 July 2023 release.

I’ve written a lot about the causes of the productivity slowdown and possible policy options to address it, reflecting my own perspectives on the importance of innovation and on redressing the UK’s regional economic imbalances. Here I just make two points.

On diagnosis, I think it’s really important to note the mid-2000s timing of the break in the productivity curve. Undoubtedly subsequent policy mistakes have made things worse, but I believe a fundamental analysis of the UK’s problems must recognise that the roots of the crisis go back a couple of decades.

On remedies, I think it should be obvious that if we carry on doing the same sorts of things in the same way, we can expect the same results. Token, sub-scale interventions will make no difference without a serious rethinking of the UK’s fundamental economic model.