This question is meant to be constructive in tone, not another channel to start debates or express opinions or frustrations on what happened around AI in the past weeks. (There are plenty, more appropriate questions to do that.)
In GPT on the platform: Data, actions, and outcomes Philippe concludes
At the end of the day, it all boils down to this: We have to walk the middle path.
I've been trying to picture what this means and where we could go from here. I try to take into consideration that SE, Inc. and the community of users have different, but also many common interests. Very briefly (and a bit contrived) these interests can be summarized by: SE wants (needs) traffic, users want quality answers that are easy to find, they both want SE to continue to be a unique Q&A site.
It seems likely that SE is going to experiment with system-generated answers to questions.* If so, then maybe is this the area where we can find middle ground? Can it be implemented such that
- SE gets its traffic.
- The community can keep dealing with human answers largely the way it was used to, the way that was developed and tried in many years of consensus-driven community moderation.
- We all benefit?
Can we achieve these goals? I think it's possible if we make these answers visible and for the rest as little disruptive as possible. What follows is an attempt at an implementation plan that meets these conditions.
- The answer is clearly marked as generated.
- The answer should be generated after (at least) 5 minutes, i.e. the time revision history will start recording. - On second thoughts, this should last much longer, otherwise posting poor questions will be too rewarding. It would almost be a shortcut to using ChatGPT directly.
- It is signed by a new type of community user or marked Community Wiki (CW).
- The answer can be voted on/accepted. Downvotes have the usual reputation penalty.
- As long as the answer is not accepted, the question will show as unanswered even if the answer receives votes. This is important because it will not change the workflow of people looking for "unanswered" questions.
- Only human answers contribute to the answers count. This is important for people who were used to looking for questions that have zero answers.
- Posting the generated answer doesn't bump the question nor change its activity status. This will preserve the current front page experience.
- Generated answers can't be edited. This allows learning models to exclude their own pure answers and thus prevents them from confirming their own bias.
- Generated answers can't be flagged and/or deleted. I.e. except for voting, they are out of the community's hands, and fully under SE's control.
- We can't comment on generated answers. The constructive way to point out any flaws is to post a better answer.
- The generated answer will always be on top in any sort mode (the sort options box is below it), therefore...
- A generated answer can be collapsed to help answerers focus on the question alone and other readers to focus on human answers. It will keep its collapsed status when sorting is changed.
For human answers:
- The existence of a generated answer may make it less tempting to post another solely generated answer.
- Network policies regarding the use of AI in human answers (SE-wide and network-specific) are fully determined by community consensus (these policies, what they are now and how they should develop, is not a subject of discussion here).
- Community moderation works as usual.
- This restrains a question sooner than we are used to. Normally, a question shouldn't be changed essentially once it has received answers (then revisions should only make the question more answerable), so users will have less time to change the question considerably. Or the generated answer should be regenerated 5 minutes after the question was edited.
- Answering questions will become harder because (serious) answerers will always have to look at the generated answer as well as, obviously, the question.
So am I missing anything (very likely!)? Am I hallucinating? Are there better, or additional ways to find middle ground? AI is a fact. We have to find ways to integrate it naturally and helpfully into our everyday lives. Please contribute in finding ways how to do this on our SE platform.
*Not that I'd like it, but let's suppose it's inevitably going to happen.