Cohere CEO: A Few Silicon Valley Labs Shouldn't Write AI's Rules
Who Gets to Define the Rules for AI?
Aidan Gomez, CEO of Cohere, argues that the latest safety proposal from Anthropic CEO Dario Amodei asks governments for antitrust exemptions so a handful of dominant labs can set global AI standards. He compares this to past regulatory failures like credit rating agencies and Europe's car dealership cartel, and proposes four pillars for better rules: an evidence-based risk framework, mandatory transparency, scoped testing, and open certification.
Convince a government that AI is an existential threat and you can convince it to outlaw your competition.
- jeffreygoesto
The article revolves a lot around the "how" and "why" but does not really answer it's own title. We thought we have the answer right there: in democracies, the people delegates working out the rules to a group of elected people. The rules then are enforced by another, independent group of people. Currently this seems to be broken in (not only) the USA.
- Moon_Y
Allowing several leading laboratories to formulate their own rules under the protection of anti-monopoly exemptions essentially means outsourcing the formulation of public policies to the stakeholders. The direction of independent review is correct, but the review standards, enforcement authority, and participation qualifications must undergo public debate, rather than being determined behind closed doors by a few companies.
- Stitch4223
Most laws, if not all, for new technologies such as the internet and AI already exist.
For example: thievery, extortion, murder, etc. are already well-established crimes.
Putting "cyber" or "AI" in front of them makes it the same crime.
I'm really interested in what is actually new about these rules.
- throwworhtthrow
I believe this is a reaction to Dario Amodei's essay yesterday, "We Must Pace the Frontier": https://news.ycombinator.com/item?id=49672510
Cohere's disagreement seems mostly about who makes the rules. From the essay (which does seem significantly written by AI, sadly):
> The rules around AI are getting written either way. What's still open is whether they get written by a group anyone can join and with evidence anyone can check, or by a handful of companies in a room with the door shut.
Of course none of the AI companies will suggest that they should be accountable for the illegal actions taken by agents they host, but I feel like that could be a very effective "pacing mechanism", and addresses the concerns brought up by this essay about fairness and transparency.
- moktonar
Shouldn’t this be a democratic process?