Executive summary
The unprecedented speed and scale of progress with artificial intelligence (AI) in recent years suggests society may be living through an inflection point. The virality of platforms such as ChatGPT and Midjourney, which can generate human-like text and image content, has accelerated public interest in the field and raised flags for policymakers who have concerns about how AI-based technologies may be integrated into wider society. Beyond this, comments made by prominent computer scientists and public figures regarding the risks AI poses to humanity have transformed the subject into a mainstream political issue. For scientific researchers, AI is not a novel topic and has been adopted in some form for decades. However, the increased investment, interest, and adoption within academic and industry-led research has led to a ‘deep learning revolution’ () that is transforming the landscape of scientific discovery.
Enabled by the advent of big data (for instance, large and heterogenous forms of data gathered from telescopes, satellites, and other advanced sensors), AI-based techniques are helping to identify new patterns and relationships in large datasets which would otherwise be too difficult to recognise. This offers substantial potential for scientific research and is encouraging scientists to adopt more complex techniques that outperform existing methods in their fields. The capability of AI tools to identify patterns from existing content and generate new predictions also allows scientists to run more accurate simulations and create synthetic data. These simulations, which draw data from lots of different sources (potentially in real time), can help decision-makers assess more accurately the efficacy of potential interventions and address pressing societal or environmental challenges.
The opportunities of AI for scientific research are highlighted throughout this report and explored in depth through three case studies on its application for climate science, material science, and rare disease diagnosis.
Alongside these opportunities, there are various challenges arising from the increased adoption of AI. These include reproducibility (in which other researchers cannot replicate experiments conducted using AI tools); interdisciplinarity (where limited collaboration between AI and non-AI disciplines can lead to a less rigorous uptake of AI across domains); and environmental costs (due to high energy consumption being required to operate large compute infrastructure). There are also growing barriers to the effective adoption of open science principles due to the black-box nature of AI systems and the limited transparency of commercial models that power AI-based research. Furthermore, the changing incentives across the scientific ecosystem may be increasing pressure on researchers to incorporate advanced AI techniques at the neglect of more conventional methodologies, or to be ‘good at AI’ rather than ‘good at science’ ().
These challenges, and potential solutions, are detailed throughout this report in the chapters on research integrity; skills and interdisciplinarity; innovation and the private sector; and research ethics.
As an organisation that exists to promote the use of science for the benefit of humanity, this subject is of great importance to the Royal Society. This report, Science in the Age of AI, provides an overview of key issues to address for AI to positively transform the scientific endeavour. Its recommendations, when taken together, should ensure that the application of AI in scientific research is able to reach its full potential and help maintain public trust in science and the integrity of the scientific method.
This report has been guided by a working group of leading experts in AI and applied science and informed by a series of activities undertaken by the Royal Society. These include interviews with Fellows of the Royal Society; a global patent landscape analysis; a historical literature review; a commissioned taxonomy of AI for scientific applications; and several workshops on topics ranging from large language models to immersive technologies. These activities are listed in full in the appendix. In total, more than 100 leading scientific researchers from diverse disciplines contributed to this report.
While the report covers some of the critical areas related to the role of AI in scientific research, it is not comprehensive and does not cover, for example, the provision of high-performance computing infrastructure, the potential of artificial general intelligence, nor a detailed breakdown of the new skills required across industries and academia.
Further research questions are outlined below. The Society’s two programmes of work on Mathematical Futures () and Science 2040 () will explore, in more depth, relevant challenges related to skills and universities.
Key findings
- Beyond landmark cases like AlphaFold, AI applications can be found across all STEM fields, with a concentration in fields such as medicine, materials science, robotics, agriculture, genetics, and computer science. The most prominent AI techniques across STEM fields include artificial neural networks, deep learning, natural language processing and image recognition ().
- High quality data is foundational for AI applications, but researchers face barriers related to the volume, heterogeneity, sensitivity, and bias of available data. The large volume of some scientific data (eg collected from telescopes and satellites) can total petabytes, making objectives such as data sharing and interoperability difficult to achieve. The heterogeneity of data collected from sensor data also presents difficulties for human annotation and standardisation, while the training of AI models on biased inputs can likely lead to biased outputs. Given these challenges, data curators and information managers are essential to maintain quality and address risks linked to artificial data generation, such as data fabrication, poisoning, or contamination.
- Industry and academic institutions are advancing AI innovation for scientific research (). The past decade has seen a surge in patent applications related to AI for science, with China, the United States, Japan, and South Korea dominating the number of patents filed in these territories. A review commissioned for this report suggests the valuation of the global AI market (as of 2022) is approximately £106.99 billion ().
- China contributes approximately 62% of the patent landscape. Within Europe, the UK has the second largest share of AI patents related to life sciences after Germany, with academic institutions such as the University of Oxford, Imperial College, and Cambridge University featuring prominently among the top patent filers in the UK. Companies such as Alphabet, Siemens, IBM, and Samsung appear to exhibit considerable influence across scientific and engineering fields.
- The black-box, and potentially proprietary, nature of AI tools is limiting the reproducibility of AI-based research. Barriers such as insufficient documentation, limited access to essential infrastructures (eg code, data, and computing power) and a lack of understanding of how AI tools reach their conclusions (explainability) make it difficult for independent researchers to scrutinise, verify and replicate experiments. The significant potential to advance discoveries using complex deep learning models may also encourage scientists or funders to prioritise AI use over rigour. The adoption of open science principles and practices could help address these challenges and enhance scientific integrity ().
- Interdisciplinary collaboration is essential to bridge skill gaps and optimise the benefits of AI in scientific research. By sharing knowledge and skills from each other’s fields, collaboration between AI and domain subject experts (including researchers from the arts, humanities, and social sciences) can help produce more effective and accurate AI models. This is being prevented, however, by siloed research environments and an incentive structure that does not reward interdisciplinary collaboration in terms of contribution towards career progression.
- Generative AI tools can assist the advancement of scientific research. They hold promise for expediting routine scientific tasks, such as processing unstructured data, solving complex coding challenges, or supporting the multilingual translation of academic articles. In addition, there may be a place for text-generation models to be used for academic and non-academic written tasks, with potential implications for scholarly communications and research assessment. In response, funders and academic institutions are setting norms to prevent non-desirable uses (), ().
Future research questions
The following topics emerged in research activities as key considerations for the future of AI in science:
- AI and computing infrastructures for science: How can AI workloads be optimised to harness the full potential of heterogeneous computing infrastructures in scientific research, considering the diverse requirements of different scientific domains?
- AI and small data: What are the implications of the growing use of AI for researchers in which only small data is available? How can AI techniques be effectively used to augment small datasets for training purposes? What trade-offs exist between model size reduction and preservation of performance when applied to small data scenarios?
- AI and inequities in the scientific system: What barriers exist in providing equitable access to AI technologies in underrepresented communities? How can AI be used to broaden participation among scientific and expert communities, including underrepresented scholars and non-scientist publics?
- AI and intellectual property: What inputs of AI systems (datasets, algorithms, or outputs) are crucial for intellectual property protection, and in what ways does it interact with the application of open science principles in science?
- AI and the future of skills for science: How are the skill requirements in scientific research changing with the increasing integration of AI? What competencies will be essential for researchers in the future and what efforts are needed to promote AI literacy across diverse scientific disciplines?
- AI and the future of scholarly communication: How is the landscape of scholarly and science communication evolving with the integration of AI technologies? How can AI be leveraged to improve knowledge translation, multilingualism, and multimodality in scholarly outputs?
- AI and environmental sustainability: What role can AI play in promoting sustainable practices within the scientific community? How can AI algorithms
be optimised to enhance the energy efficiency of environmental modelling, and contribute to sustainable practices in fields such as climate science, ecology, and environmental monitoring?
- AI standards and scientific research: How can AI standards help address the challenges of reproducibility or interoperability in AI-based scientific research? How can the scientific community contribute to the establishment of AI standards?
