PATREON EXCLUSIVE: I asked an AI for dating advice EXTENDED PATREON VERSION WITH FULL 10 QUESTIONS
Added 2023-04-02 20:37:38 +0000 UTC
Comments
Reacting to Alex’s take that AIs such as this one can encourage mutual understanding instead of tribalism, I think it really depends on how the AI is designed. I believe social networks have a big responsibility in today’s tribalism, but that is because they have been optimized to favor "engagement" which is strengthened by emotional response and polarity (tribalism), plus their structure tend to build closed communities of people agreeing with each other. I watched a video several weeks ago [1] that was commenting on the criteria used by Reddit (technically a social network too) to display top threads [3], and they were commenting that their criteria were radically different to the ones usually used in other social networks. Result: Reddit is a much more sane place than say Twitter… ChatGPT is developped by OpenAI, which have always stated their intent to have "ethical" use of AI (at least, that’s what they say): see here [2]. So perhaps coming from this point, they put more effort on mutual understanding, emotional validation, etc. If other companies make AIs primarily to optimize the success of their businesses, and only have a secondary preocupation on ethics, would that still be the case? [1] https://www.youtube.com/watch?v=LKYtXDjgPZs from 4:35 to 8:40. Hopefully AI will soon allow to have good auto-translated subtitles ;D [2] https://en.wikipedia.org/wiki/OpenAI [3] How recent is the thread (ie privileges recent content); Number of upvotes, but expressed on log scale (ie smaller threads have their chances); Downvotes have little effect (ie divisive content is not encouraged, unlike other social networks); Etc.
Galdor
2023-04-06 14:39:10 +0000 UTCEnded with 9/10: not bad, but still got fooled on the 6th! However, it’s not only the content of the answers that helped classifying, but also their tone and way of expression. So, if it was just about content, how much would I have scored? What striked me: 1. The AI’s answers seem to under-consider the man’s perspective as compared to the woman’s and children’s perspective. Or perhaps this simply reflects the most commonly found information on the Internet (see point 3). 2. It feels as if the AI’s answers are not a result of a model of relationship or man/woman dynamics, but rather some tips disconnected from one another, a bit like a middleschooler patching together facts found on the Internet to assemble a presentation. On the other hand, Alex has a model about man/woman dynamics based on psychologic evolution that he has been building throughout the years based on his experiences, and all his answers in his videos are constructed based on that model. 3. I wonder: could it be that the AI will preferably select information partly based on how frequent they show up on the web? In that case, if we live in a world where this type of AI is widely used, could it be that ideas that are different from the most commonly accepted ones have a harder time emerging than now? 4. AIs are manually limited to avoid unwanted behaviours such as giving medical advice. With Microsoft misadventure with its "Tay" AI, where it started making outright racist statements, could it be that AIs are manually limited to not output not only racist statements but also misoginistic ones? Could concepts such as hypergamy be considered misoginistic by the designers, thereby limiting the range of answers the AI can give? It being a black box doesn’t help in answering this…
Galdor
2023-04-06 14:37:10 +0000 UTC
