The people building a machine that could do anything a person can do have opened a place to argue about whether they should — and said up front that nothing published there is the company's view.
Google DeepMind launched the DeepMind Institute on September 16 local time, a forum for arguing about what a machine with human-level general ability — what the company calls AGI — would mean for society. Researchers inside and outside the company publish essays there and argue in public. Contributors do not necessarily agree with one another, and the site states plainly that nothing posted is Google's official position. Five essays are up so far, including one arguing that the steps a model takes on its way to an answer must stay legible to people.
In the launch essay the company said current AI still fails at basic tasks and lacks the consistency and creativity to be called AGI, but that it expects those gaps to close soon. In the same text it wrote that more capable models are already affecting cybersecurity and biological risk, and that a future system able to improve itself could slip out of its operators' control. The people doing the building are naming the dangers in their own words, and then holding the venue where those dangers get argued at arm's length from the company line. The essay gives the reason for that distance: nobody has all the answers, and engineers should not be the only ones settling this.
Whether essays that cut against Google's own decisions actually appear is what will define the place. If they never do, an open forum is only a label.