I think this is a super important question and want way more conversation about it—but could we re-frame your conclusion as being not that we shouldn’t use AI, but should be mindful about how we’re using AI?
The scenario you described appears to be a pretty bad use. But I think much of the harm you’re seeing could be mitigated. Here are some ideas, just off the top of my head, addressing the issues you listed (in order):
Use of AI in research should -
Consider the appropriateness of AI in that context (e.g. is this an area where we need the most up-to-date answers? Is this an area where we want to consider non-western perspectives?)
Approach AI-generated answers critically, treating them as vibes-based answers rather than having any authority (and in group-work contexts, leaders should encourage this)
Have AI write up its answers in bullet points rather than full text, so that a human is always contributing to the style
Be a second or later-resort option (try to think creatively/critically first, rather than relying on AI—again, leaders can encourage this)
In group settings: encourage new or unusual ideas (addresses the last two points).
I know these are far from perfect solutions. Point 4 is admittedly quite hard to keep up (I feel myself struggling with this). But to me it feels similar to how a calculator makes people lazy (I’m sure I can’t do mental arithmetic now as well as I could when I was 12), but is still a net win. It seems likely that if we create good habits/culture about using AI, its benefits can significantly outweigh the downsides, even in research.[1] But I do think that requires a lot of conversations, and maybe some research, into how to use it well and avoid those pitfalls. So I would love to see more posts discussing this.
I think these benefits are pretty significant. For instance, (and as a counter-point to 5), I find AI can actually help to reign in crazy ideas, by acting as a sanity-check tool; I also find it’s helpful to quickly spot holes in an argument when otherwise I would have only gotten feedback from a colleague some days later; and it can quickly structure disorganized ideas. But surely there are many more.
I think this is a super important question and want way more conversation about it—but could we re-frame your conclusion as being not that we shouldn’t use AI, but should be mindful about how we’re using AI?
The scenario you described appears to be a pretty bad use. But I think much of the harm you’re seeing could be mitigated. Here are some ideas, just off the top of my head, addressing the issues you listed (in order):
Use of AI in research should -
Consider the appropriateness of AI in that context (e.g. is this an area where we need the most up-to-date answers? Is this an area where we want to consider non-western perspectives?)
Approach AI-generated answers critically, treating them as vibes-based answers rather than having any authority (and in group-work contexts, leaders should encourage this)
Have AI write up its answers in bullet points rather than full text, so that a human is always contributing to the style
Be a second or later-resort option (try to think creatively/critically first, rather than relying on AI—again, leaders can encourage this)
In group settings: encourage new or unusual ideas (addresses the last two points).
I know these are far from perfect solutions. Point 4 is admittedly quite hard to keep up (I feel myself struggling with this). But to me it feels similar to how a calculator makes people lazy (I’m sure I can’t do mental arithmetic now as well as I could when I was 12), but is still a net win. It seems likely that if we create good habits/culture about using AI, its benefits can significantly outweigh the downsides, even in research.[1] But I do think that requires a lot of conversations, and maybe some research, into how to use it well and avoid those pitfalls. So I would love to see more posts discussing this.
I think these benefits are pretty significant. For instance, (and as a counter-point to 5), I find AI can actually help to reign in crazy ideas, by acting as a sanity-check tool; I also find it’s helpful to quickly spot holes in an argument when otherwise I would have only gotten feedback from a colleague some days later; and it can quickly structure disorganized ideas. But surely there are many more.