What Next for Academia?

Recent demos featuring Astra for dexterous manipulation are truly impressive. But they also raise two questions that I don't think academia has good answers to yet:

1. How should credit be assigned when an agent synthesizes many prior works into a new research result?
2. How can academia attract and retain talents when so much of its incentive structure relies on credit?

The credit assignment problem

I've spent most of my PhD working on sim-to-real dexterous manipulation. Looking at recent Astra results -- from pen spinning to in-hand cube rotation -- I recognize many ingredients from prior research: simulators, URDF assets, RL algorithms, reward designs, etc. What I find the most impressive is that Astra doesn't simply retrieve prior works, but actually surveys, understands, adapts, integrates, and experiments with them autonomously -- much like a researcher would.

But I also feel uneasy about the credit assignment. Much of science depends on "intellectual lineage" maintained through authorship and citations. And writing rigorous related-work sections has been how researchers separate new discoveries from what came before. The system is by no means perfect (missed citations happen often), but it helps keep track of the history and progress of science. With so much knowledge distilled into LMs today, it seems increasingly easy to generate new results simply by prompting an agent, with no record of the potentially enormous trees of works behind.

If an agent reads 10 papers, clones several repos, adapts a few algorithms, swaps in new assets, and runs the experiments, does the credit go to the agent itself, or to whoever runs the agent? If neither, what should be cited, and who is responsible for preserving the intellectual lineage?

Perhaps agents should start generating proper "Related Works" too? Or perhaps, as agentic research becomes more autonomous, agents will conclude that maintaining an intellectual lineage was never necessary after all, and that such bookkeeping was just a suboptimal artifact of human society?

This leads to my second concern.

The incentive structure problem

Academia runs largely on one incentive: having one's name attached to what one contributes. This works hand in hand with the aforementioned credit system: named authorship allows individuals to receive credit for their work, which in turn influences reputation, hiring, grants, tenure, prizes, and ultimately who gets the resources to keep doing research. The system is, again, imperfect; but it is one of the few (if not only) rewards that people can get for spending years of hard work creating knowledge.

When credit assignment becomes difficult, so does academia's incentive structure: when an agent produces research autonomously, it is easy to credit whoever built or ran the agent, but not the people whose work contributed to the knowledge that the agent built upon.

And when the incentive structure breaks, what remains to attract talented people to academia -- to work so hard for so little material return?

I do believe there are people who pursue science purely out of passion, and I deeply respect them. But from a system-design perspective, I don't believe academia can be sustained on goodwill and sacrifice alone (and I don't think it should).

The technological question ("how much of research can be automated?") is exciting, but the institutional question is concerning. If scientific credit becomes increasingly opaque, what happens to academia and the intellectual lineage it maintains?

I wonder how others think about this.