At present, the impact of AI on the economy, business, labor market and society has become the focus of attention for enterprises, relevant professional bodies, governments and non-governmental organizations, and most OECD members have developed their own national AI strategies. However, with the exception of specialized journals, these studies rarely address the role of AI in research, and for all the uses of AI, improving scientific efficiency may be the most valuable. AI can help uncover more scientific knowledge and improve the efficiency of scientific research, thereby strengthening the critical foundation for addressing global challenges. Against this backdrop, on 26 June 2023, the OECD published Artificial Intelligence in Scientific Research: Challenges, Opportunities and the Future, which explores topics such as: current, emerging and future scientific uses of AI; where AI needs to make progress to better serve science; changes in scientific productivity; Measures to accelerate the adoption of AI in research in developing countries.
The difficulty of scientific research has increased
The researchers point out that the reasons for the decline in the efficiency of scientific research may be the following:
(1) changes in the scientific incentive mechanism;
(2) limited private sector involvement in basic sciences;
(3) the economic cost of scientific discovery is high;
(4) to make new breakthroughs, it is necessary to absorb more previous and diverse scientific results, so larger teams are needed, but large teams seem to make fundamental discoveries less easily than small teams;
(5) A large number of papers have "exploded", exceeding the reading limit of review experts;
(6) the large scale of scientific literature in different fields;
(7) New disciplines emerge, and some breakthroughs require interdisciplinarity, but there is also friction between disciplines;
(8) The number of scientific laws is limited.
Artificial intelligence in science today
AI is being applied in all fields and stages of science, from hypothesis generation to experimental design, monitoring and simulation, all the way to scientific publishing and communication. In the future, AI may optimize many scientific workflows end-to-end: from data collection to final statistical analysis.
Still, AI's potential impact on science is still a long way from actually making a difference. The researchers point out that AI is very helpful for tasks such as hypothesis generation, experimental monitoring, and accurate measurements, such as deep learning is very effective in processing unstructured data, and causal model innovation will bring great benefits to medicine and social sciences. In addition to this, AI can also be used to summarize research papers, such as peer review.
At the same time, with the help of artificial intelligence, scientific research, especially experimental scientific research, will gradually be automated. In the future, human scientists will decide how to work with AI scientists and to what extent AI can identify its own problems and solutions, scientists note. In this case, AI may find biases in research by human scientists, or areas of research that human scientists have failed to explore, creating synergies.
The researchers also point out that machine learning has spread to all areas of physics, and its applications can be divided into three broad categories: data analysis, modeling, and model analysis. In addition, AI has been an integral part of the drug development process for decades, and advances in AI have allowed it to enter the relevant fields of drug discovery.
Artificial intelligence, science and developing countries
It is unclear the impact of AI on developing countries or whether AI will widen the scientific capacity gap between rich and poor countries. However, researchers in Europe, North America, and China clearly dominate AI research and its application in science. In 2020, East Asia and the Pacific accounted for 27% of all conference publications, North America 22%, and Europe and Central Asia 19%. In contrast, sub-Saharan Africa accounted for only 0.03 per cent. Previous studies have found that the computing resources needed for cutting-edge AI research tend to favor well-resourced universities, big tech companies, and wealthy countries.
Opportunities and challenges for the application of artificial intelligence to scientific research
The rational reasoning, abstract modeling, and logical reasoning (deductive and inductive) abilities of human scientists are at the heart of scientific discovery, and machine learning performs poorly in these areas, especially when machine learning requires large amounts of data, which are often unavailable in certain scientific fields. In addition, there are issues related to data annotation and annotation, for example, manually labeling large databases requires time and resources, and the ability level of annotators may vary; Differences in data characteristics in certain scientific fields, which may make it impossible to generalize in various fields; and the black-box nature of machine learning.
Currently, AI is still unable to ask interesting research questions, design appropriate experiments, and understand and describe their limitations.
At present, natural language processing technology can not meet the high requirements of comprehension of machine reading tasks, natural language processing technology lacks rich world models as the basis of language, they cannot contact the entities, relationships, events, experiences, etc. involved in the text. As a result, even the most complex models often produce fabrication or nonsense.
While AI has made great progress, capabilities such as intuition, contextualization, and abstraction are still unique to humans, so science should be advanced by combining collective human intelligence and machine intelligence, such as coding and discovering knowledge, connecting and building knowledge, supervision and quality control.
Policy recommendations
Develop a long-term multidisciplinary plan to use artificial intelligence to promote scientific and technological progress
(1) Countries should undertake a broad multidisciplinary program that brings together computer scientists and other scientists, engineers, statisticians, mathematicians, etc., to use AI to solve challenges;
(2) facilitating interaction between roboticists and domain experts, laboratory robots can revolutionize certain areas of science, reducing costs and dramatically speeding up experiments;
(3) governments can encourage and support visionary and far-reaching initiatives;
(4) In order to promote the development of artificial intelligence and science, high-performance computing and software must be promoted and popularized;
(5) use artificial intelligence technology to assist academic training; Governments can take steps to improve the availability of open research data, harnessing the power of data in everything from health to climate.
Second, use public research and development to promote the application of artificial intelligence in the field of science and technology
(1) Public R&D can deepen the application of AI in science and engineering for research areas that need breakthroughs;
(2) should support the development of open platforms (such as OpenML and DynaBench) to track which AI models are most effective at solving various problems;
(3) Given that public R&D can help foster new, interdisciplinary top-level thinking, governments should support a broad program to create a knowledge base that is critical to AI in science;
(4) policymakers can support research that examines and quantifies the loss of technological resilience, creativity and inclusion of reduced scope of AI research, as well as the potential impact of the growing dominance of related industries in AI research;
(5) Funders can also help develop specialized tools to enhance collaboration between human scientists and AI teams and integrate these tools into mainstream science.
3. Study management matters
(1) Policy bodies should systematically assess the impact of AI on day-to-day scientific practice, including on human-scientist-AI teamwork, work, career trajectory and training;
(2) funders and policymakers should establish response mechanisms to act on the insights gathered;
(3) funders and policymakers can support and establish new independent forums for ongoing dialogue on the changing nature of scientific work and its impact on research productivity and culture;
(4) policymakers should focus on the deployment of large language models;
(5) Policies should address the potential dangers posed by the dual-use use of AI-driven drug discovery.
Fourth, policymakers should enhance their professionalism to facilitate scientific decision-making
(1) existing social networks and platforms can be used to help disseminate emerging practices;
(2) take steps to improve the reproducibility of AI research, where public funding agencies can require free sharing of code, data, and metadata with third parties, allowing them to run experiments on their own hardware;
(3) Sub-Saharan Africa, as well as other developing regions, has every reason to get more AI funding in science.
Disclaimer: This article is transferred from Meta Strategy, original author Allen Wang. The content of the article is the personal views of the original author, this public account compilation/reprint is only to share, convey different views, if you have any objections, welcome to contact us!
Transferred from 丨 Meta Strategy
Written by Allen Wang
About the Institute
Founded in November 1985, the International Institute of Technology and Economics (IITE) is a non-profit research institution affiliated to the Development Research Center of the State Council, whose main functions are to study major policy, strategic and forward-looking issues in the development of the mainland's economy, science and technology, and economy, track and analyze the development trend of world science and technology and economy, and provide decision-making consulting services for the central government and relevant ministries and commissions. "Global Technology Map" is the official WeChat account of the International Institute of Technology and Economics, dedicated to conveying cutting-edge technology information and scientific and technological innovation insights to the public.
Address: Building A, Building 20, Xiaonanzhuang, Haidian District, Beijing
Tel: 010-82635522
WeChat: iite_er