"Our goal is not to have an AI that is biased in any particular direction. We want the default personality of OpenAI to be one that treats all sides equally. Exactly what that means is hard to operationalize, and I think we’re not quite there. I think we can all agree with that." We have all seen the chaos which ensues when the media "treats all sides equally" without regard for objective truth. For example, which side of Tucker Carlson is more equal? This currently feels to me like an insoluble problem in the general case. • Is it OK or blasphemous for midjourney to create an image of the prophet? • Is criticism of the Thai king justified or lese majeste? • Is the lab leak hypothesis a fringe crackpot idea, a plausible alternative or what actually happened? • Was Jan 6th freedom fighters trying to take back a stolen election or a bunch of people sadly misled about the accuracy of the count? Until as a world (or in the latter cases simply a country) we can broadly agree on the facts, it's going to be hard to determine what is and isn't appropriate behavior for an AI. I agree that a default "treat all sides equally" doesn't work. Maybe there is a good reason why psychopaths enjoy (or don't mind) killing people, but I'm not sure I'd want to make that an "equal airtime" argument. Determining the exact bounds of which behaviors do and don't fall within "there are two sides" vs "there's right vs wrong" I think is way more challenging in practice. The least bad starting point is presumably to place the onus on national governments to regulate appropriate speech that can be shared by people or artificial entities within their jurisdictions, but there are still plenty of things that it's legal to say in America that we probably wouldn't want our AI's sharing. So how do we draw the line? #ScholER ...