Every task OpenAI lists for its new mathematics panel is advice

The mathematicians the group means to speak for found out when everyone else did. The Verge reported that its abrupt arrival caught many of them by surprise, and that researchers who called it a good first step were left asking how much influence it will have and whether OpenAI will actually listen ^[10][11]. Some also questioned whether nine prominent names can stand in for the wider community ^[12].

OpenAI's description of the job is the most specific thing on the record. "The group will advise on the review and communication of emerging results: they will help OpenAI assess their significance, advise on how to coordinate their dissemination, and advise on academic and professional standards of mathematical research," the company said. "It will also advise on how our tools can support mathematical research and learning." ^[8] That is four tasks, and every one of them is advice. The announcement does not give the panel authority to delay or withhold a result ^[1].

It does appear to give the panel early sight. The Verge reported that the group will be granted early access to OpenAI's research in order to fulfil its task ^[9]. Members are unpaid, and the group can change its own membership ^[6]. "Its value depends on its members being able to exercise their own judgement and challenge ours," OpenAI said ^[7].

An outside reviewer can affect a release two ways: by seeing it before it ships, and by being able to stop it shipping. Put those on two axes and there are four positions. AGMAI sits in the one with early sight and no stated stop. A group in that position cannot keep the result from going out. What it can shape is how the result is described and who is credited for the idea.

Martin Hairer, one of the members, published a blog post on Tuesday answering what he called a "torrent of misinformation" about how the group came together ^[17]. He acknowledged the approach came from OpenAI, said not all members had been contacted by the company, and described the result as "genuinely independent of OpenAI and any other of the so-called 'frontier' AI labs" ^[20]. On the alternative, he wrote that there is no sense in "taking a hard 'ostrich' approach of simply ignoring the AI labs and refusing to talk to them on principle" ^[19]. Which members OpenAI approached first, and how the others were chosen, is still undisclosed ^[16]. Simon Machado, a researcher at ETH Zurich, told The Verge the whole thing seemed "shrouded in mystery" ^[14].

The Verge's account places the panel after a run of spectacular mathematical results that became a reputational crisis for OpenAI ^[22], and reports that why the group was formed at all remains unclear ^[13]. Hairer wrote that "it is obvious that AI has already had a profound impact on mathematical research and raises numerous questions of correct attribution of ideas, priority, human understanding of ideas, etc." ^[18]

For a board like this, the test is narrower than independence: which artefact it can hold, and for how many days. Neither number appears in what OpenAI announced, so what has been built is a review of wording, and the board's only sanction is saying so in public. Hairer said that beyond "obvious confidentiality requirements" members have signed no restrictive agreements governing what they can say publicly, and that they work unpaid ^[21].