Views on AI Safety For-Profit Feasibility
Marius Hobbhahn shares two perspectives on building for-profit AI safety businesses.
Marius Hobbhahn posted two perspectives on AI safety for-profit models. One view holds that externalities cannot be internalized so companies lack incentives to pay for safety. The other holds that safety blocks core business functions and companies cannot externalize all related costs. Chris Painter replied that he basically agrees and sees distinct lanes for organizations that help companies implement safety versus those that inform the public and governments whether the work suffices. Michael Nielsen added that both dynamics already occur in many industries and seem present in AI.
imo there are broadly two views on the feasibility of AI safety for-profit models 1. You cannot internalize the externalities, so no incentives to pay for safety. In that world, hard to make the business work. 2. Safety is a core business blocker and you cannot externalize…
Reactions from ranked influencers
4 postsIn many industries both are true - companies pay enormous amounts to achieve safety in ways directly affecting their business (market-supplied safety), and may ignore it in parts which don't (externalities which resist internalization). This seems to me already the case in AI.…
@MariusHobbhahn @sethbannon I basically agree with this. I think there are distinct lanes for orgs that “help companies get the job done” on safety, and orgs that help inform the public and governments on “whether the job has been done well enough”, and both need to exist.
I basically agree with this. I think there are distinct lanes for orgs that “help companies get the job done” on safety, and orgs that help inform the public and governments on “whether the job has been done well enough”, and both need to exist.
imo there are broadly two views on the feasibility of AI safety for-profit models 1. You cannot internalize the externalities, so no incentives to pay for safety. In that world, hard to make the business work. 2. Safety is a core business blocker and you cannot externalize…
@MariusHobbhahn Though note that when a type of safety work is the bottleneck to further AI progress it is also the bottleneck to the imposition of further risks of more advanced kinds. (i.e. one is in the RLHF situation where a safety technique is instrumental to creating more…