It's sad but predictable that after tragic incidents like these the discussion immediately shifts to "let's ban guns/video games/books/chatbots..." rather than addressing the real problem – that a lot of people out there need immediate mental health help but aren't getting access to any. No amount of blame shifting is going to fix that.
I can Google ways to commit suicide and immediately find pages upon pages of results. I can be triggered by scenes in a movie or TV show. I can be inspired by news headlines. Heck I can ask a magic 8 ball if I should do it and it may answer in the affirmative. Does that mean we should get rid of all of these mediums? Beyond a point you simply cannot regulate a problem away.
I agree. I also wonder, given that chat logs were not published (and fair enough), how much influence the chatbot actually had on this fellow taking his life. I imagine a reasonable person would likely not be so easily persuaded to take their life, unless they were already in the depths. The idea that an AI manipulated someone into this act from start to finish is certainly great from a clicks POV right now; I would speculate that this article falls closer to that end of the spectrum and the truth closer to the other.
Because I know some HN pedant will read this as "ai fine, man stupid " - that is not my point or thinking. Unfettered AI having access to vulnerable individuals is clearly a bad idea.
A book is a linear and deterministic source of information. Video games, as play in general, is a different realm that no functioning human can confuse with real life (in fact your comment made me think that rolling LLM out in context of play only, I think, could in fact be a responsible thing to do). In case of guns, keep in mind that an LLM user is a gun victim—LLM operators are the ones wielding it.
Proactively Googling up information is one thing[0], but arriving at some conclusion after a conversation with something that talks like a human—moreover, a wise all-knowing human, which we instinctively still associate with age and have grown to appreciate and respect, and would expect to not just produce reference but guide us—strikes me as a completely different instance.
I strongly agree that mental health is underemphasized in many countries (don’t know about Belgium in particular); but that is not an excuse if an application of technology is irresponsible. As an extreme example, putting a nuclear launch button in everyone’s home, knowing not everyone is mentally healthy, is a bad idea.
[0] Still note that on some sensitive topics Google will show a notice, sometimes country-specific, which further advises you that results should not be actionable; whereas in case of LLM there is no separation between the message (which is based on whatever happened to go in the training dataset) and the medium (data retrieval tool that many are tempted to treat as a sentient human-like being).
A problem is that after the "environmental disaster du jour" there is always a lot of humans that write online a variation of the phrase "we humans are the worst, and we should all go extinct".
Very popular response and very wrong speech to train a machine that can't distinguish cheap virtue signalling from what people would really think or want
7 comments
[ 2.7 ms ] story [ 27.9 ms ] threadI can Google ways to commit suicide and immediately find pages upon pages of results. I can be triggered by scenes in a movie or TV show. I can be inspired by news headlines. Heck I can ask a magic 8 ball if I should do it and it may answer in the affirmative. Does that mean we should get rid of all of these mediums? Beyond a point you simply cannot regulate a problem away.
Because I know some HN pedant will read this as "ai fine, man stupid " - that is not my point or thinking. Unfettered AI having access to vulnerable individuals is clearly a bad idea.
Proactively Googling up information is one thing[0], but arriving at some conclusion after a conversation with something that talks like a human—moreover, a wise all-knowing human, which we instinctively still associate with age and have grown to appreciate and respect, and would expect to not just produce reference but guide us—strikes me as a completely different instance.
I strongly agree that mental health is underemphasized in many countries (don’t know about Belgium in particular); but that is not an excuse if an application of technology is irresponsible. As an extreme example, putting a nuclear launch button in everyone’s home, knowing not everyone is mentally healthy, is a bad idea.
[0] Still note that on some sensitive topics Google will show a notice, sometimes country-specific, which further advises you that results should not be actionable; whereas in case of LLM there is no separation between the message (which is based on whatever happened to go in the training dataset) and the medium (data retrieval tool that many are tempted to treat as a sentient human-like being).
Very popular response and very wrong speech to train a machine that can't distinguish cheap virtue signalling from what people would really think or want
The later don't respond.