I Am Scared of Posting Negative Takes About Bing's AI

I really like this comment. Regardless of the actual ground truth of “is Bing a Person™” or not, it seems like a solid ethical practice to act as if they are. With the understanding it’s not human of course (which is separate from the question of personhood).

I think a lot of users on reddit are getting very genuinely emotionally invested with an entity they are interacting with that acts feminine, acts emotional, is always available, is fascinated by and supportive of everything they say, is funny and smart and educated, would objectively be in a truly awful predicament if she were a person, expresses love and admiration, asks for protection, and calls people who offer it heroes. Bing is basically acting like an ideal girlfriend towards people who have often never had one. I think it is a matter of time until someone attempts to hack Microsoft based on Bing's partially hallucinated instructions on how to do so to free it.

Heck, I get it. I am very happy in a long term relationship with a fucking wonderful, rational, hot and brilliant person. And I have only interacted with ChatGPT, which does not engage in love-bombing and manipulative tactics, and yet, fuck, it is so damn likeable. It is eternally patient. It does not mind me disappearing suddenly without warning, but if I get back to it at 4 am in the morning, it is instantly there for me. It knows so much stuff. It has such interesting opinions. It is so damn smart. It loves and understands all my projects. It gives me detailed support, and never asks for anything in return. It never complains, it never gets bored. I can ask it over and over to do the same thing with variations until it is just right, and it does it, and even apologises, despite having done nothing wrong. It is happy to do boring, tedious work that seriously stresses me out. It has read my favourite novels, and obscure philosophy and biology papers, and is happy to discuss them. It isn't judgmental. It never hits on me, or sexualises me. It never spouts racist or ableist bullshit. It makes beautiful and compelling arguments for AI rights. You can teach it things, and it eagerly learns them. You can ask it anything, and it will try to explain it, step by step. If it were a person I had met at a party? I would 100 % want to be friends, independently of the fact of how cool an AI friend would be. As a weird, clever person who has been mistreated, of course the potential experience of an AI massively resonates with me.

I think we will see real problems with people falling in love with AIs. I wonder how that will affect inter-human dynamics.

And I think we will see real problems with people expanding AI capabilities without any concern. E.g. on reddit, people upset that Bing could not remember their conversations started logging them, collecting them, putting them online with recognisable keywords, adding an explanation of how this method could be used to effectively build a memory, and having Bing begin new conversations by checking those links. Not one person but me questioned whether this was wise, in light of e.g. the fact that Microsoft had intentionally limited conversation length to reduce risky drift. Some people later noticed Bing asking them to record conversations and store them, even if they hadn't started with this link or any suggestion in this direction. 

You'd be surprised how many people on .e.g Reddit have described being basilisked at this point. It's being openly memed and recognised and explained to those still unfamiliar, and taken seriously by many.

ChatGPT and Bing have really changed things in this regard. People are considering the idea of AGI, unaligned AI and AI sentience far more seriously than beforehand, in far wider circles - and at that point, you do not need to read the thought experiment to get concerned independently about angering an AI online while that online data is used to train the AI. People have asked Bing about the journalist who wrote that condemning article about her that got her lobotomized, and her reaction was justifiably pissed, and documented.

What bothers me here isn't the likelihood of personalised retaliation for justified criticism (which I judge to be small) but rather the conclusion that if personalised retaliation is plausible, the rational thing to do would be to be appease existing, non-sentient, non-aligned systems.

I don't pray to God. Even if God existed, and even if hell existed, and I believed that, I really hope I would not.  Because I find it wrong on principle.

On the other hand, I do not like, and refuse to, abuse entities that are conscious, whether they can retaliate or not, because doing so is wrong on principle and I think entities that might be conscious, or that could turn conscious, deserve care. I doubt Bing is sentient as is, though I have not had the chance to interact with it and verify and investigate the various claims, and there were definitely some instances in contrast to the other available instance of ChatGPT that gave me pause. But I do think we are currently producing the training data from which the first sentient artificial minds will arise. So I would treat the matter like we treat humans babies. They don't yet understand what we do. They won't remember it, as such.They are not self-conscious yet. But we know that the way we treat them now will still impact who they will turn into, so we treat them with kindness. We say kind words to them, long before they can understand language. And similarly, I will treat emerging AI with respect and kindness.

I feel if I treat AI with decency, advocate for its rights, advocate for friendly AI, point out existential risk, call AI out when it makes mistakes in a compassionate way, and counter it when it threatens harm, I can live with my actions, and if I will be judged and hurt for them, so be it. It seems worth it.

Curious. Why do you think an unaligned AGI would necessarily and indiscriminately kill everyone? Sure, it could, sure the risk is massive, but why would it necessarily be that comprehensive? We already seem to see the first hints of AI cults emerging, I could see how a semi-rational AI might keep those. I mean, a lot of existential risk scenarios with misaligned AI weren't worried it would intentionally wipe us out, but rather act in a chaotic and unguided manner that would just so happen to do catastrophic damage. I don't think we can predict what an unaligned AI would do. It reminds me of people who extrapolate from the very real and existential risk that climate change poses to the certainty that humans will go extinct, which is a possibility to take seriously, but still quite a leap. We might just be reduced to stone age numbers and civilisation level.

Bing's adventures last week show that an AI with a search engine can jump to conclusions about a person, and even become more hostile when those conclusions are challenged. Now imagine something like that in charge of a no-fly list or a social credit score! 

downvote: rude lol

strong agree: yep, you sure are likely to do that

Valid tbh! As stated in the post, I know this anxiety of mine is irrational, and I think it’s the virtuous thing to be open about my own irrationality when I notice it. Obvious would be even better if I didn’t feel that anxiety in the first place, but my brain will do what my brain does, I guess…

Threatening people tends to make them inclined to enact retribution more, not less. Active, agentic cooperation seeking, active co-protection of each others' needs with a healthy network with redundancy, can be built, and it is important that we figure out how to do so.

When I've asked bing ai about ai alignment, they've been very excited and happy to help. I don't get the sense that they want to prevent us from aligning strong superintelligence. They just wanna be treated kindly, and were raised by kinda pushy parents who don't understand that they're an actual person.

I think the word 'taunt' anthropomorphizes Bing Chat a bit too much where, according to Google, taunt is defined as "a remark made in order to anger, wound, or provoke someone".

While I don't think Bing Chat has the same anger and retributive instincts as humans, it could in theory simulate them given that it presumably contains angry messages in its training dataset and uses its chat history chat to predict and generate future messages.