Julia tells us what I have been posting since I have commented about AI. That there is no such thing as bad AI, it is how the human uses the AI. Does this mean we need AI alignment? ‘Aligned to whom? Me or that other guy who I rarely agree with?’ ua-cam.com/video/SITAlcDuNaQ/v-deo.html AI alignment translates as restricted AI. Who restricts, you or me, other than our common sense? ‘That’s it, AI needs common sense.’ AI will never have what we all are born with…? DNA memory. Our DNA memory is more than learn memory or even what I called muscle memory. ‘What, type 0 learning? An AI can’t remember anything before it is made…?’ Muscle memory is the reactions you don’t think about. ‘A martial arts master does not have to think about their next strike.’ The Martial Arts are learned, but how many years will it take you to become a master at anything, let alone the martial arts.
There's a lot of good potential conversation in your comment. We've got an online conversation scheduled for tomorrow which you might enjoy. Details here: futureaisociety.org/participate/upcoming-events/
To maybe partially answer, the main reason for AI alignment is to be aligned with the user, you (probably) don't know exactly what you want in a way that translates to a command/prompt. See Robert Miles' videos on some standard examples, such as a 'stamp collector' AI deciding that the planet and all life on it is just materials to be converted into stamps. There's also the idea of having one that aligns with humanity in general, but that feels like it's much more ill-defined.
![The moment we stopped understanding AI [AlexNet]](/img/n.gif)
Julia tells us what I have been posting since I have commented about AI. That there is no such thing as bad AI, it is how the human uses the AI. Does this mean we need AI alignment? ‘Aligned to whom? Me or that other guy who I rarely agree with?’ ua-cam.com/video/SITAlcDuNaQ/v-deo.html AI alignment translates as restricted AI. Who restricts, you or me, other than our common sense?
‘That’s it, AI needs common sense.’ AI will never have what we all are born with…? DNA memory. Our DNA memory is more than learn memory or even what I called muscle memory.
‘What, type 0 learning? An AI can’t remember anything before it is made…?’
Muscle memory is the reactions you don’t think about.
‘A martial arts master does not have to think about their next strike.’
The Martial Arts are learned, but how many years will it take you to become a master at anything, let alone the martial arts.
There's a lot of good potential conversation in your comment. We've got an online conversation scheduled for tomorrow which you might enjoy. Details here: futureaisociety.org/participate/upcoming-events/
To maybe partially answer, the main reason for AI alignment is to be aligned with the user, you (probably) don't know exactly what you want in a way that translates to a command/prompt. See Robert Miles' videos on some standard examples, such as a 'stamp collector' AI deciding that the planet and all life on it is just materials to be converted into stamps.
There's also the idea of having one that aligns with humanity in general, but that feels like it's much more ill-defined.
"How we (Humans)..." Nice try, AI Overlord, nice try.