SCAI Question 4

Solving Scientific Problems

How should we advance AI to solve scientific problems that are critical and beneficial to humanity as a whole?

Context and Assumptions

Throughout human history, significant advances have been made possible only by scientific progress. Scientific discoveries such as electricity, penicillin and semiconductors that have played key roles in progress have been driven by scientific discoveries.

Harnessing the power of AI systems offers a promising avenue for addressing some of society’s most difficult problems that remain unsolved. These intractable problems which significantly impact societal and individual longevity, are ultimately solvable but have persisted for decades, such as those involving climate change and complex biological systems. AI holds the potential to bring innovative self-directed approaches that will assist humanity in fundamental ways. For example, in climate change, advances in carbon sequestration will enable humanity to deal with the impact of global warming due to fossil fuels, and in complex biological systems, advances in genome sequencing and editing could help us understand human biology and develop cures for disease.

There are, however, some challenges. There is a general and systemic lack of integration of foundation models and exploratory methods that generate and examine new hypotheses. There is also a need to develop logical and causal inference mechanisms in AI neural systems, as well as adequate funding for the specialised infrastructure required for AI development.

Question

How should we advance AI to solve scientific problems that are critical and beneficial to humanity as a whole?

In considering this question, we also need to think about how scientists and researchers across the world can come together to harness AI and prioritise resources in applying these powerful systems to scientific problems.

Indicators of Progress

In the short-term, a notable indicator of progress includes the growing evidence of increasing cross-disciplinary and cross-border collaboration and cooperation. Addressing traditional boundaries between different scientific and industrial areas can play a significant role in facilitating access to a diverse and extensive range of scientific and mathematical corpus. In the longer term, international research collaboration, supported by multilateral funding, has the potential to yield results that benefit the global community. Importantly, this would allow us to integrate scientific knowledge into AI models to increase their reliability and accuracy, while using less computation.

A key indicator of progress will be the emergence of theorem-solving AI systems that are applicable to a wide variety of scientific areas of interest, and allow us to draw on cross-disciplinary scientific knowledge at a scale previously unavailable but with the potential to transform scientific development. We look forward to a time when we are able to see high impact scientific papers being written by such AI systems with application in areas such as carbon sequestration, seasonal climate prediction, deciphering the human ageing process and new materials design.