Saturday 12 August 2023

Meta and basic research as a public good

Earlier this week, Meta disbanded a research team that had been working on using artificial intelligence to create a database of protein structures. As the Financial Times reported (paywalled):

Meta has axed a team that used artificial intelligence to create the first database of more than 600mn protein structures, in a signal the company is abandoning purely scientific projects in favour of building moneymaking AI products.

The social media giant had employed a group of about a dozen scientists on a project called ESMFold, which trained a large language model capable of processing vast amounts of biological data to predict protein structures. The effort has been lauded by those involved in the development of new drugs and treatments...

Though the protein-folding team was small compared with the thousands of AI scientists and engineers still employed at Meta, the move to axe their project showed the company’s desire to abandon blue-sky research in favour of AI projects that can generate revenues, the people familiar with the matter added.

The problem here is one of public goods. Public goods are goods that are non-rival (non-rival means that one person consuming the good does not reduce the amount of the good or service available for everyone else), and non-excludable (non-excludable means that if they are available for anyone, then they are available for everyone). Examples of public goods include streetlights, policing, and basic research.

The ESMFold team were created a database of predicted protein structures that could then be used by other researchers. A database like that is non-rival (one researcher using it doesn't prevent it being used by other researchers). Is it excludable? It need not be, but because Meta was releasing the database for researchers to use for free (it is available here), they made it non-excludable. As a non-rival and non-excludable good, the database is a public good.

The problem with public goods is free riders - some researchers may use the database, and might have been willing to pay for it, but many would use it without paying (and Meta wasn't asking for payment in any case). That makes the costs of providing the database difficult to justify for a firm that is profit-maximising. It is not a sustainable position (as evidenced by Meta shutting the team down), which is why public goods are usually provided by the government (either directly, or through procurement from a private provider).

Meta is redirecting the resources the ESMFold team was previously using into other AI applications, which are also non-rival but are excludable. These are club goods, which Meta can charge subscribers for access to. That they would redirect the resources in this manner does raise a question though - why didn't they simply make ESMFold pay-for-access (that is, excludable)? There was no need for it to be treated as a public good and shut down. Although perhaps, having already released earlier versions of ESMFold for free access, many researchers may have already downloaded a copy of the database, so making it excludable would simply drive those researchers to use their own free version instead? That suggests that Meta believes the benefits (in terms of subscriber revenue) are exceeded by the costs of continuing to fund the ESMFold team.

Another related question is why Meta released ESMFold as open access in the first place. This makes more sense than it appears at first. The database user community probably gave a lot of feedback to the ESMFold team, which was not only useful in improving the AI capability within that team, but had more general use as well. With an excludable database, the user community, and the quantity and quality of feedback received from the community, would likely be much smaller. Now, with AI development much more advanced, perhaps the value of the feedback received from the ESMFold user community has decreased to an extent that the benefits of retaining the open access database are exceeded by the costs.

Basic research is a public good. Even when a private firm like Meta initially provided access to a public good for free, we should not be surprised when the public good eventually becomes unavailable. A private firm will only provide goods (including public goods) for as long as the benefits exceed the costs.

No comments:

Post a Comment