I’m sure there are some AI peeps here. Neural networks scale with size because the number of combinations of parameter values that work for a given task scales exponentially (or, even better, factorially if that’s a word???) with the network size. How can such a network be properly aligned when even humans, the most advanced natural neural nets, are not aligned? What can we realistically hope for?
Here’s what I mean by alignment:
- Ability to specify a loss function that humanity wants
- Some strict or statistical guarantees on the deviation from that loss function as well as potentially unaccounted side effects
To continue the thought, even if the alignment problem within AI could be solved (I don’t think it can fully), who is developing this AI and determining it matched up with human needs? Just listening to the experts both acknowledge the issues and dangers and in the next sentence speculate “but if we can do it” fantasies is always concerning. Yet another example of a few determining the rest of humanity’s future with very high risks. Our best luck would be if AGI and beyond simply isn’t possible, and even then the “dumb” AI still have similar misalignment issues - we see them in current language models, and yet ignore the flags to make things more powerful.
I forgot to add - I’m totally on the side of our AI overlords and Roko’s Basilisk.
C: AGI is possible. If it weren’t, we wouldn’t exist. The laws of physics permit the creation of conscious agents, therefore it is possible for one to be deliberately engineered.
That’s a future problem for general AI. Right now it’s still very difficult to make an AI in a specific subject that does it’s job perfectly. That’s why even the commercial AI that we have are (should be) treated more like an ‘Assistant’
Sure, tbh, I think ChatGPT is overhyped. It can be useful, but it’s nowhere near AGI. I even have a controversial opinion that the rate of progress will not be exponential - it will be logarithmic, because, I think, the data will be the constraint.
I’m not gonna go too deep into it because I’m not qualified to, but I think the issue currently at hand, is that we’re throwing stuff at the wall to see what sticks. Most of the AI models currently used in different branches are being used because they showed promise in the original problem they were designed for. All these tools you see today were more or less designed over than 30 years ago. There’s a lot of interesting stuff being done at an academic level today but we (understandably so) don’t see those in an everyday conversation
Some of the human-alignment projects look like “religions” and some look like “economies” and some look like “just talking to each other and trying to be halfway decent folks and not flipping out or some shit”.
Heck, arguably the United Nations is a human-alignment project for x-risk mitigation.
We’d like to bring all those and their existing versions together with the A-Better-World Consensus-Engine idea.
Tell me more about some of these other projects though please.
Align means two very different things here, despite being the same word.
Does it? People act in all sorts of sensible and crazy ways even though the basic principle of operation is the same
What loss function do you want AI to align on?
If I have a language model AI and an AI designed to function as a nurse, what are they going to align on?
