How can you know whether science was carried out by a person or machine? It feels like this question should be easy to answer, but it turns out it’s not. An AI can now can write a paper that is very difficult to distinguish from human produced text. As you are reading this text now, there is realistically no way for you to know with complete certainty that I have not just generated some or all of this in response to a prompt. At the point at which science interacts with a community in the form of a paper, it will always be difficult to know whether the ideas expressed in that paper were really person or machine.
Proof-of-person describes a means through which two things can be disambiguated: first, was some thing produced or performed by a person or machine? And second, if it was produced or performed by a person, which person was it? Both questions are important for science, but each is answered differently.
For the first question, biology matters. DNA, you, a physically embodied organic-ness. Machines do not have these, and it’s not likely they will – it would not be a machine. When you are present with a person physically and they communicate something to you, you know that at least the way those ideas were articulated in that moment was theirs.
For the second, physical or digital documentation is used for this purpose. It’s likely that humans have identified human from human through forms of written certification for at least a few thousand years. In science now, we have ORCiD numbers – a sequence of characters that are unambiguously yours as an academic.
Questions 1 and 2 can help us understand the unique contribution of a person, and then measure it. For now I’m calling it the “Human h-index”, and it’s a measure of the impact of verifiable human presence.
The Human h-index would be calculated something like this. Take all the workshops and conferences you’ve attended, take all the citations of those workshops and talks, and then calculate an h-index. These are all events where your physical presence at that event can be verified. For any conference or workshop you’ve attended, your proof-of-person can be verified by your attendance before you are added to an author list, and the event or talk is indexed on Google Scholar. Many conferences are already indexed on Google Scholar. It seems likely that workshops could do too. If an idea came from a talk or a workshop, you cite it. For workshops, it’s often difficult anyway to know specifically who an idea came from. There might be a number of interesting downstream consequences for such a metric.
First, there are problems in academic publishing. There is significant burden on review for academic journals. Peer-reviewers are not paid, with publishers drawing significant profit at the expense of the finance of learned societies, governments, and universities. Emphasis on paper publication gives publishers significant power over the funding landscape, and the direction of science. Journals are swamped with publications, desperately trying to solve a likely unsolvable problem of proof-of-person for the text they receive. If emphasis was placed on the human presence of an idea, through a Human h-index, we could draw finance from publishers towards learned societies. These learned societies are more often trusted with the responsibility of annual meetings, workshops, and conferences. At the same time, it’s possible we’d see the pressure come off journals, and journals can stop trying to understand whether text is human. Your human contribution is measured by contributing as a person to another person.
Second, a Human h-index would likely incentivise conferences and workshops. This would mean more travel, and with large conferences often difficult to get to, incentivize flying to places far from home. We should push for greater localized collaboration, and then for international travel, let the Human h-index drive online communication and not physical presence. Giving a talk online can be cited, just as a talk in person.
Third, individuals who are neurodivergent, physically less able, or don’t natively speak the majority language of an event, can find large gatherings difficult. An event with many people can be overwhelming, and sharing ideas in spoken language nerve-wracking. A Human h-index would need to come with an increasing conscientiousness. We should be more deliberate in the way we accommodate different ways of thinking and being, and have more spaces at conferences for people to feel safe when they share. We need to better help those who are less physically able. There are likely many individuals who could contribute ideas at conferences or workshops, but can’t because there is not physical access. A Human h-index needs to drive us to help these people better than we do now.
Fourth, gaming of such a metric would happen, but some difficulties we can anticipate. It’s possible that events that are not really academic gatherings would become citable, such as coffee breaks. I actually think this is not totally unreasonable, but potentially not workable. Likely we would need thresholds an event can cross, or an authentication of events granted by a learned society. A potentially very big issue could be a new form of gatekeeping and competitiveness at workshops and conferences. A hope would be that capacity on conferences and workshops will grow as we make a more concerted effort to meet and share science, and that the way in which we tend towards deescalation in person is sufficiently baked-in
Fifth, citations may feel more confusing, at least at first. The magnitude of citations on physical presence will be lower, given the potential audience is lower. But I don’t think this matters, since the Human h-index would become internally consistent i.e., citations on a talk or workshop are measured against other talks or workshops, not papers. Another issue might be that a talk or workshop can be cited in a paper, but perhaps with more difficulty the other way around i.e., citing a paper in a talk. We would need workshop and conference organisers to publish an abstract associated with your talk or event, along with any material cited on where the idea for that talk or workshop came from. We should probably have been doing this anyway, at least for talks. An additional issue might be citation trust. Citations between papers can be traced, and others can verify whether that citation is reasonable. I don’t know what the outcome would be where there is no traceability, and once a citation goes down it would set a precedent. A hope is that the same mechanisms that stop us being dishonest elsewhere stop us here too. But I don’t know, it would be very difficult to know where the new equilibrium might end up.
Papers, and citations to papers, would not be less important. Imagine a Venn diagram. On the left is a bubble for only machine, on the right is a bubble for only person. The intersection of the bubbles, that’s both machine and person, and we just allow papers to be that. The intersection is the contribution of you plus a machine, and it will remain important as a memory and record of contribution. The non-intersecting region of the bubble on the right is the proof-of-person Human h-index. The non-intersecting region of the bubble on the left is not allowed to submit papers, and regardless, is not relevant as far as humans doing science and figuring out what we contribute.