Choose Otherwise chooseotherwise.org

An open research program · 2026

Open Questions

The argument is finished. The inquiry is not. Here is where it stays open, and how to take a piece and run.

on the research behind Choose Otherwise


The Inversion is a finished argument. It has been coded, compared, and deposited as a working paper with a citable DOI, and it answers the sharpest objection raised against it. But a finished argument is not a finished inquiry, and this one was built to be continued. It states its own limits. It names the exact conditions that would prove it wrong. And several of its pieces are not closed findings at all but open instruments, designed from the start to be picked up by someone other than the person who made them.

That is not a hedge. It is the point. The paper argues that the capacity to create is being concentrated, and that the contestability which quietly rescued every earlier information technology will, this time, have to be built rather than waited for. A claim like that earns nothing by being guarded. It earns its keep by being testable: something you can field, extend, deepen, or break. So here is where it is open, stated plainly enough that you could take one of these and start this week.

A claim about contestability should stay contestable.

Five places it’s open

Open problem 01

Field the expert panel

Does the Inversion survive independent, cross-disciplinary judgment?

The third method is a fully specified modified-Delphi instrument (the panel, the two rounds, the scoring rubric, the analysis plan, even the human-subjects framing) that has never actually been run. It is the test of whether the central judgment is recognized beyond its author or is an artifact of one analyst’s framing. Running it means recruiting roughly eighteen to twenty-four experts across AI governance, communications and development, political economy, and compute practice, fielding both rounds, and reporting whether the scores converge with the coding or split from it. The instrument is ready to use as written.

Who this is for: a graduate student or research group, anyone who can convene experts.

Open problem 02

Extend the diagnostic beyond the Western frontier

Does the inversion look the same outside the US–Western frame, and what does capability concentration do across the development spectrum?

The coding centers on US and Western European media history and on the US–China compute frontier. The development question (who is least able to escape a dependency on rented capability) is exactly where the older communications-and-development tradition predicts the sharpest consequences, and it is the least studied. The work is to code non-Western trajectories on the same rubric, beginning with the Chinese AI ecosystem, and to pair capability concentration against human-development indicators rather than assuming the Western case generalizes.

Who this is for: area-studies and development scholars, comparativists.

Open problem 03

Deepen the cases with primary evidence

Does the mechanism hold up under harder evidence than secondary sources?

The paired case studies establish the decisive distinction (that broadcast’s concentration was reach-driven and reopened, while AI’s is capability-driven and may not), but they establish it as a mechanism, from existing scholarship, not from primary data. Archival and quantitative work on the compute supply chain, and on the Arab Spring cases, would test that mechanism far more severely: the actual numbers, ownership maps, and timelines behind the claim.

Who this is for: economists, historians, data people.

Open problem 04

Design the contestability mechanisms

What does keeping capability contestable actually look like, in statute and institution?

The governance chapter maps the option space (compute accounting, public compute as countervailing power, anticipatory regulation, the real limits of open weights, the contestability principle), but it deliberately stops short of prescribing a form, because that depends on variables the paper does not resolve. The concrete design is wide open: public compute at a scale that actually shifts the locus of capability rather than just subsidizing access to the incumbents’; a workable compute-reporting regime; interoperability mandates, and at which layer of the stack. This is where the diagnosis turns into something buildable.

Who this is for: policy designers, compute-governance researchers, lawyers, economists.

Open problem 05

Watch the falsifiers

Is the inversion already dissolving?

The claim is falsifiable on purpose, and it names its own disconfirmers. Any one of them holding would mean artificial intelligence rejoins the historical pattern (another concentration awaiting its reopening) and the argument is wrong. Three signals are worth tracking over time: sustained algorithmic efficiency that collapses the compute required to reach the frontier; open-weight models that reach and hold genuine parity with it; and distributed or federated training that lets non-incumbents pool capability at frontier scale. This is the one anybody close to the frontier can watch directly, and the most useful thing a builder can contribute.

Who this is for: practitioners, anyone near the frontier.

If one of these is yours, don’t work on it alone.

Read the paper, take a problem, and tell me what you’re thinking: whether you want to field it, fund it, argue with it, or pull it apart. I’ll point you at what already exists, connect you with anyone working a nearby piece, and then get out of your way. That’s the whole offer: not a following, a few people building the same thing from different sides.