SCIENTIFIC COLLECTIVE and ARTIFICIAL INTELLIGENCE
AI and Collective Intelligence
Considerable progress in the utility of AI in advanced mathematical research has been documented in the last 12 months. Solutions for many Erdős problems have been found by interacting with new AI models. Participation of undergraduate students and young scientists in obtaining these results is also evident, as in the case of one of the most notable examples, on Erdős problem 1196. Among the most prominent mathematicians exploring the mathematical capabilities of AI models and their implications for researchers are Fields medal winners Terry Tao and Tim Gowers.
Another example is that of the disproof of the unit distance conjecture; the companion remarks paper includes insightful considerations by several mathematicians. The most widely held opinion among mathematicians seems to be that while this result is not yet a major mathematical breakthrough, it is at a level that would be acceptable for publication in one of the top math journals. Will Sawin has already extended these results in a new preprint and has shared a balanced view of these developments. Another recent math paper contains the following statement: “GPT-5.5 Pro was used as a sounding board in the early stages of the development of this proof, but the final proof, including all the main ideas, was almost entirely human-generated “.
This progress might inform the use of AI in science more generally. Complex scientific problems could benefit from collective human brainstorming using AI tools.
Long term aim:
Improve human collective scientific intelligence.
Most estimates suggest that AI superiority over top individual human experts will be reached within 5 to 20 years.
Even then, AI might not be superior to the collective intelligence of humanity, which has room to improve, continuing the trend towards open science and collaboration that has taken place in the last few centuries (a trend with several ups and downs but a clear direction).
Human collective intelligence might still be able to control and orchestrate powerful AI tools. It might complement AI, being better in some domains and less effective in others.
Uncontrolled AI has many risks, including:
- Too much power for the few companies owning the most advanced AI models. Risks posed by AI used by bad or selfish human groups.
- Unwanted consequences of a very advanced AI that we cannot fully understand. Some experts wonder if it might lead to human extinction.
- Replacement of human intellectual workers, which will have no income and may not find meaning in their activities.
AI also offers many opportunities to improve the health and wellbeing of humans that we could accelerate by optimally interacting with it. We can continue to participate intellectually if we find new ways of collaborating with other humans and with AI. Both types of intelligence can benefit from the interaction.
A suitable application for human collective intelligence is a complex biomedical challenge, the development of digital models of cells and organisms, eventually allowing us to better understand our own bodies.
This is a task requiring broad integration, in need of an effort from an entire community. It is not soluble just by a single individual, lab or center.
Considerations that motivate scientists (each is of variable importance for different scientists):
- Appreciation of long-term benefit for humanity
- Knowledge and understanding
- Recognition for an individual and for a group
- Education, including education for new roles in the age of AI
- Support for an individual career in science
- Autonomy and creativity
- Engaging in philanthropic activities with time or resources
- Scalability to a large community
The following Rounds of discussion and reflection might accelerate improvements in collective intelligence. They are based on small steps that can lead to a realistic progression:
First Round: a competition among students within one Institute, for 5-6 Institutes, with broad geographical representation.
One prize of 1,000 dollars for the best student in each Institute if at least 3 students per center participate. Prize assigned by a vote of the students within each Institute. Individual ideas are made public and are permanently recorded to allow for both short-term and long-term recognition. Interactive brainstorming with AI encouraged.
Second Round: a competition among the participating Institutions, promoting the first step of integrated collective reasoning. Each Institution will participate with a common contribution. Open to all the scientists of each institution, which will also share openly their individual contributions. Alumni might participate. Students from the previous round might help to inform and stimulate, for example asking questions to other scientists. Interactive brainstorming with AI encouraged. Recognition given by vote of the participating Institutes (one vote per Institute).
In follow-up Rounds more scientists and centers will be involved. Communities of scientists holding minority views and based in different centers might also form and participate separately, to encourage creativity and innovation. AI can help to connect these scientists. The monetary prizes will be progressively replaced by the perceived value of the recognition as the number of participants increases. Recognition by philanthropists, foundations and scientific leaders will encourage participation.
Within 5 years: a comparison (possibly in the form of a competition) between humanity-led orchestration versus AI-led orchestration for this biomedical challenge. Orchestration implies the use of multiple human and AI resources.
As an example, feedback from AI models on this topic is shown here. It includes a mixture of points useful for further reflection and other less convincing ones, and seems suitable as a brainstorming step.
---------------------------------
The summary of a previous discussion about virtual cells and organisms can be found here
The summary of a previous discussion about going from PDB to virtual cells can be found here