It's been an intense stretch with a lot of writing lately. Thank you for reading along. Taking a couple of weeks off to recharge over the Nordic summer. Back in August! #AICRIT #UniTurku @utu.fi

Nobel economist Roger Myerson making the case that AI's real risk is concentrating power, and that the fix is accountable local institutions. He names the why. The how is what I work on in AICRIT: oversight that holds when the center doesn't. #AICRIT

Roger Myerson@rbmyerson.bsky.social · 4w ago

A new note considers challenges of AI from a perspective of political-economic theory. home.uchicago.edu/~rmyerson/hu... A key concern: With substitution of artificial intelligence for human intelligence in management, our world may be dominated by larger organizations managed by fewer people.

You can't oversee what you can't see. Independent oversight needs the decision's data and reasoning on the record. Open it up, and "trust us" becomes something you can actually check. #AICRIT #UniTurku @utu.fi

Exactly what I meant yesterday. OpenAI removed a Swedish HR pro's useful AI tool with an opaque automated decision he couldn't understand or contest. The irony: that unaccountable call is what independent oversight exists to prevent. #AICRIT #UniTurku @utu.fi www.svd.se/a/V6jAq1/ope...

Avstängd av Open AI: ”Har lärt mig läxan”

Open AI stoppar Johannes Sundlos populära AI-verktyg. Enligt den amerikanska jätten bryter hans GPT med svenska kollektivavtal mot reglerna, men någon förklaring har han inte fått: ”Lite mer öppenhet ...

svd.se

An AI model went dark for the whole market for over two weeks, then came back. Whatever the reason, the lesson holds: if your oversight depends on one central system staying up, it isn't resilient. Accountability has to hold when the center fails. #AICRIT #UniTurku @utu.fi

Where an AI runs shapes whether you can hold it to account. A system you must oversee, running on infrastructure you don't control under rules that aren't yours, is hard to inspect. Local and edge AI, under your own control, keeps oversight real. That's our work at AICRIT. #AICRIT #UniTurku @utu.fi

Before a serious diagnosis, you get a second opinion. Before a big legal decision, you get counsel. That isn't distrust, it's how important calls are made. AI now makes decisions on its own. It deserves the same: an independent, traceable second opinion. #AICRIT #UniTurku @utu.fi

Att granska resultatet från AI själv låter enkelt i teorin men är svårare i verkligheten. I vår forskning i AICRIT (Åbo universitet) provar vi det nu i ett verkligt fall i offentlig sektor, tillsammans med Civisto, en oberoende granskare, en second opinion för beslut som påverkar människor. #AICRIT

There's no single rulebook for AI. An autonomous system answers to many at once, in different countries and contexts, so it can't be fully compliant with all of them. At AICRIT (Univ. of Turku) we work on the honest version: catch when it crosses a line, and leave an account that holds up. #AICRIT

Följer AI-samtalen i Almedalen på distans i år. En fråga jag hoppas hamnar på bordet: vem granskar besluten som AI fattar på egen hand? I vår forskning i AICRIT (Åbo universitet) bygger vi just det, en oberoende granskare, en second opinion, helt enkelt. Kul att W har en egen panel idag. #AICRIT

Today, young and old, we take a pause to celebrate midsummer.

Glad to be on W, a platform built for trust. At AICRIT (Univ. of Turku) we work on independent oversight of the decisions AI makes on its own: a traceable, source-attributed second opinion, so 'why did it decide this?' has an answer that holds up. #AICRIT