Experiment

Battle of Ideas

Both sides of a contested claim, argued at full strength — with the assumptions, the counters, and the sources on the table.

Experiment v0.1 · 7 July 2026
Battle of Ideas — the claim page for 'Humans have free will': two columns of argument cards, AGAINST on the left and FOR on the right, each card carrying a validation-tier badge and a four-bar confidence read. Neutral typographic layout, no colour coding of the sides.
A claim page — the strongest arguments for and against, side by side. Open live at battle-of-ideas.com →

An encyclopedia of arguments. Pick a contested claim — remote work, nuclear power, free will — and see the strongest case each side can make, laid out side by side. Not opinions: the best version of each argument, with its assumptions named, its weak points attacked, and its sources on the table.

Why I built it

Ask an LLM a contested question and you usually get a smoothed-down answer — both-sides-ish, hedged, forgettable. But the models can do far better than that if you make them argue. So I did. For each claim, one AI is told to be the strongest possible advocate for one side. A second AI, with no stake in the outcome, tears every argument apart. A third checks the load-bearing sources against the open web. What comes out is a map of the disagreement — where it's genuinely contested, and where one side is simply weaker.

Where the best arguments win — and engaging with the strongest version of the other side is one click away.

The honest part

This is an experiment, and it says so on every page. Every argument is AI-generated — I don't hide that; the whole thing is judged on merit, not authorship. The verification is real but bounded: it spot-checks each argument's most load-bearing sources for whether they exist and roughly say what the argument claims.

That step earns its keep. On the remote-work claim it caught the famous "23 minutes to refocus after an interruption" statistic — traced it to a magazine interview rather than the paper everyone cites, whose actual data cut the other way — and dropped the argument. That is exactly what the apparatus is for.

What's on each card

Every argument carries its receipts: an epistemic tier (is this empirically strong, logically valid, or unfalsifiable?), a four-axis confidence read, two or three targeted counter-arguments, and its sources with honest labels about what was checked and what wasn't. Columns are never padded to look balanced — where a side runs out of strong arguments, it says so and stops.

Current state

v0.1 is live at battle-of-ideas.com: six claims, both sides, read-only. It's a prototype of a bigger idea — a place where you can read the strongest opposing case to whatever you believe without going looking for it. Next is a participation layer, letting people add counter-arguments and sources through the same pipeline that grades the AI's, and cross-checking arguments across different models instead of one. The name might change; the idea won't.