Science Incentives

Summary of the third round of discussions

May 5th, 2024 - see archive.org for webpage history.

Summaries of the older rounds of discussions can be found here.

The analyses of the beginning of modern science (in the 16th and 17th centuries), of the early phases of molecular biology (in the 1950s) and of contemporary science (see interview with Joshua Graff Zivin), lead to a common conclusion:

A combination of incentives addressing career concerns and ideal aspirations is needed to advance science and to motivate scientists.

Analogies

As pointed out by historian Carlo Ginzburg in his interview , "all analogies are simplifications" but they might suggest avenues of intellectual exploration.

Evolutionary biologists (starting with Darwin in the Origin of Species (Darwin, 1859)) have shown that the evolution of complex biological organs can take place if simpler intermediate forms also have utility for their possessors. A similar approach might be considered when developing collective human intelligence.

Ideal motivations

Almost all survey responders suggested that scientists should discuss the control of AI used in fundamental science. Scientist can decide to interact with AI systems that have features they value, including more transparency, and they can contribute to train them, as explained by Sofia Serrano, Zander Brumbaugh and Noah Smith . Complex scientific problems that seem intractable using conventional approaches might be solved using novel AI methods, as discussed by Andrew McCulloch and Bjoern Peters. Deciding to train an AI model collectively is a type of problem that is most effectively addressed by a joint effort of the scientific community, and it implies a discussion of how AI can be best used in science. A major concern is that AI methods employed by large companies are not fully disclosed, and therefore cannot be independently replicated by academic scientists; this will diminish trust in science, with possible large societal impacts .

Short-term, career-promoting incentives

Many scientists have reported that their ideas and careers benefited from interactions with mentors and collaborators.

An example of the interest in finding mentors and of the willingness of senior scientists to mentor comes from the field of economics, where an initiative called "Adopt a paper" (www.adoptapaper.org) has matched more than 300 junior scholars to expert senior scholars. In this case mentors provide feedback to junior scholars about a paper before formal submission.

Mentors do not only benefit early career scientists. For example, Charles Kennel told us how he expanded his scientific interests after retiring from a major leadership position as Director of the Scripps Institution of Oceanography. He acknowledges benefiting greatly from his discussions with a mentor, Martin Rees, about potential planetary-level crises facing humanity (Kennel & Rees, 2022). Scientists with complementary expertise can certainly mentor each other at any stage of their careers.

One of the most striking examples of a fruitful collaboration, originating from a chance encounter in front of a photocopy machine, is that of Katalin Karikó and Drew Weissman (Karikó, 2023). After this chance meeting they started working together; their work was one of the foundations of the Covid vaccines and was recognized by the award of the 2023 Nobel Prize in Physiology or Medicine.

A short-term, career-promoting incentive for collective intelligence could be provided by rewarding the public sharing of ideas about the control and use of AI in science with a one-to-one video session with a potential mentor or collaborator.

Advantages

- A one-to-one session with a potential mentor or collaborator could not only be helpful in promoting collective intelligence but it is also immediately useful and enjoyable for all parties involved.

- At full scale, a one-to-one session could be offered to all participants, but some potential mentors are more widely admired, for their contributions to science or for starting successful science-based companies. The prospect of meeting them would therefore be an incentive to spend time developing the best possible ideas.

- Scientists working outside major research centers and outside developed countries face difficulties in meeting potential mentors and collaborators. Offering this opportunity to scientists worldwide would promote justice and benefit humanity by allowing more brilliant minds to demonstrate their value and participate in advanced science. AI models have shown markedly increased performance associated with larger sizes and this might also apply to human collective intelligence.

- The experience of Katalin Karikó, who was unsuccessful in obtaining academic grants and positions (Karikó, 2023), even after the work for which she later obtained the Nobel Prize was published, demonstrates a common phenomenon in the history of innovative ideas, which are often initially not supported by most scientists in a field. The recognition provided by potential mentors does not require a consensus, but only the appreciation by an individual and might better promote a diverse range of approaches.

- The one-to-one sessions are both rewards and part of the production of knowledge. The video meetings will help assure that participants are humans and not simulated by AI in a way that might distort science. The model is scalable because each participant is also a potential reward.

Potential problems and solutions

- With a small number of participants, the matching can be done by humans but beyond a certain size an AI system will be needed. Both humans and AI systems could be biased. Potential mentors could be provided with shortlists rather than with an individual indication, and they might also be able to choose among multiple independent shortlist-producing methods or do their own selection, if they prefer. The development of these methods should be open to all and separate from the public repository of ideas.

- The ideas originating from individuals and from the one-to-one meetings should be integrated. Both surveys of scientists and AI models might help, in multiple rounds. The surveys might also help to define the questions for discussion, as already shown.

- Not all scientific problems are suitable for a broad collective effort, traditional academic science or private companies have also important roles. If requested, mentors could provide advice about these careers during the one-to-one sessions, even if the participants would be selected according to the quality of their ideas about the control and use of AI in science.

Effects on society

The role of science is to evaluate and present evidence, offering, when possible, an estimate of the confidence of key statements. Major decisions about priorities and about allocation of resources should however be taken by society in general. Mentors from social sciences and humanities could help our reflection and motivate scientists that have an interest in bridging intellectual divides. Society will support science even more if all ideas are transparently shared and scientists can show to have pursued the broadest possible collective effort. The example of particle physics supports this expectation.

Practical steps

We have started collecting the names of potential mentors and we have already encountered interest from several well-known scientists. We will update this page with their names, and we will start publicizing widely this initiative when the list of potential mentors is sufficiently large.

REFERENCES

- Darwin, Charles. On the Origin of Species. (Chapter VI. Difficulties on theory) John Murray, London. 1859.

https://darwin-online.org.uk/content/frameset?itemID=F373&viewtype=text&pageseq=1

- Karikó, Katalin. Breaking Through: My Life in Science. Crown, 2023.

- Kennel, Charles F. & Rees, Martin. Two Distinguished Scientists on How to Rescue Humanity. The Anthropocene demands a massive realignment of priorities. Nautilus. 2022.

https://nautil.us/two-distinguished-scientists-on-how-to-rescue-humanity-238535/