Science, technology and innovation policy

OECD workshop on AI and the productivity of science






The OECD project on “AI and the Productivity of Science” addressed the critically important issue of the rate of scientific progress, whether this is stagnating, as recently argued by a number of scholars, and how AI could raise the pace of progress in science and discovery. This OECD workshop, a part of the project, brought together technical and policy experts to examine the evidence on a purported productivity decline in science as well as the ways that AI is currently used across different fields of science – from neuroscience to materials science - and across all stages in the scientific process. The workshop advanced the debate on what governments can do to maximise the positive impacts of AI on science, today and in the decades to come. Workshop presentations and expert discussions will become part of a comprehensive publication on the topic, to be released in early 2022. 


DAY 1 | Friday 29 Oct.


DAY 2 | Tuesday 2 Nov.

DAY 3 | Wednesday 3 Nov.

DAY 4 | Thursday 4 Nov.

DAY 5 | Friday 5 Nov.

Topic 1

The productivity of science: is there a slowdown and if so why?

  Topic 2

The current uses of AI in science.

Topic 4

Systemic conditions affecting the productivity of science.

Topic 5

AI and the implications for science in the developing world.


Topic 6 (continued)

Policy priorities to increase the impact of AI on science. 


Topic 3

The current limits of AI in science.


Topic 6

Policy priorities to increase the impact of AI on science.

Topic 7

The future: what could AI achieve in science in the next 10 years?


Might the productivity of science be stagnating?

Claims of a slowdown in science, with many alleged causes, are not new. However, such claims have been given new prominence by Bloom et al., (2020) and others. Various metrics have been cited: the number of researchers needed to maintain Moore’s Law has risen sharply; the number of researchers needed to maintain improvements in crop yields, and lower mortality due to cancer and heart disease, has grown; the real cost of developing a new drug doubles about every nine years; and the share of breakthrough patents may be falling.
The evidence is disputed. But, if true, any slowdown could lengthen timeframes for essential scientific progress. And governments, already under acute budgetary pressures, might have to spend more just to support existing rates of growth of useful science.

Can Artificial Intelligence (AI) help?
Spurred by advances in machine learning, and fed by vast realms of scientific data, AI is being adopted across most stages and fields of science. AI is helping to choose, design and plan experiments; improve measurement and observation (converting low-resolution images – for instance of mitochondria in cells - into high-resolution, low-noise images); discover meaningful relationships in colossal data sets (data flows in some science projects far exceed those of the entire Internet); generate hypotheses; learn scientific rules, such as the rules of chemistry to predict how to make medicines; identify the most suitable patients for clinical trials of drugs; create new capabilities in laboratory robots; summarize research; predict the replicability of research, and even suggest experts to review research proposals.


‌To examine all of these issues, the past OECD workshop on AI and the productivity of science (October 29th – November 5th), and later publication, will gather scientists, AI researchers, policy analysts and scholars of the economics of science. From multiple vantage points, in presentations and in essays, the experts will assess the evidence on progress in science and examine how AI is contributing to all stages of the scientific processes. The current limitations of AI in science will be considered in detail, along with the impacts of AI on science in the developing world, and the policy implications of current and possible future developments. The workshop will be livestreamed and is open to the public. An OECD publication on these topics will be launched in the second quarter of 2022.

 DAY 1 | Friday 29 October 2021
  • 00:00:06 Welcome and scene-setting
  • 00:23:10 - Session 1.1: What can bibliometrics contribute to understanding research productivity? (Giovanni Abramo)
  • 00:45:40 - Session 1.2: Economy-wide and cross-country studies
  • 00:45:40 - Declining R&D efficiency – evidence from Japan (Tsutomu Miyagawa)
  • 01:11:48 - Evidence from China and Germany (Paul Hünermund and Philipp Boeing)
  • 01:32:01 - Are ‘Flows of Ideas’ and ‘Research Productivity’ in secular decline? (Didier Sornette)
  • 02:10:05 - Session 1.3: Evidence from individual domains of technology and science
  • 02:10:05 - Micro-electronics (Henry Kressel)
  • 02:34:57 - Theoretical physics (Sabine Hossenfelder)
  • 02:57:17 - ‘Eroom’s Law’ (Jack W.Scannell)
  • 03:17:15 - Agriculture (Matt Clancy)
  • 03:39:03 - Machine learning (Tamay Besiroglu)
  • 04:07:59 - Session 1.4: Other forms and issues of measurement
  • 04:07:59 - Quantifying the ‘cognitive extent’ of science and how it has changed over time (Staša Milojević)
  • 04:37:29 - What does Total Factor Productivity indicate about research productivity? (Ben Southwood)
  • 04:55:55 - A quantitative and qualitative approach to productivity in science (Hector Zenil and Ross King)
 DAY 2 | Tuesday 2 November 2021
  • 00:00:04 Welcome and scene-setting Session 2.1: Uses of AI by stage in the scientific process
  • 00:01:26 - A typology and overview of AI in science (Aishik Ghosh)
  • 00:24:52 - AI-enhanced laboratory robots (Ross King and Patrick Courtney)
  • 00:52:43 - Tackling data privacy, security and policy challenges in healthcare with Federated Learning (Sezai Taski)
  • 01:15:28 - Artificial intelligence in citizen science: an overview of opportunities and risks (Luigi Ceccaron)
  • 01:34:31 - Open discussion session 2.1 (Bart Selman, Kenneth D.Forbus, Sašo Džeroski, Jeremy G.Frey, and Utpal Mangla)
  • 01:50:33 - Using ML to estimate the replicability of scientific findings (Yang Yang)
  • 02:10:54 - Machine reading: successes, challenges, and implications for science (Jesse Dunietz)
  • 02:33:35 - Using ML to verify scientific claims (Lucy Wang)
  • 02:57:53 - Generating new knowledge from disparate data sets (Neil Smalheiser)
  • 03:13:24 - Open discussion session 2.1 (Bart Selman, Kenneth D.Forbus, Sašo Džeroski, Jeremy G.Frey, and Utpal Mangla)
  • 03:25:30 - 2.2 Uses of AI in selected fields of science
  • 03:25:30 - AI and drug discovery (Kristóf Szalay)
  • 03:48:37 - AI and materials science (Tonio Buonassisi)
  • 04:07:07 - AI and advances in cognitive science (Justine Cassell and Abdellah Fourtassi)
  • 04:28:11 - Open discussion session 2.2 (Bart Selman, Sašo Džeroski, Jeremy G.Frey, and Utpal Mangla)
 DAY 3 | Wednesday 3 November 2021 (forthcoming)
 DAY 4 | Thursday 4 November 2021 (forthcoming)
 DAY 5 | Friday 5 November 2021 (forthcoming)




Related Documents