Comment by mistermann | Hacker News Reader

Comment by mistermann | Hacker News Reader

Parent Full thread

> It's not specifically a conservative issue. I can get chatgpt to write about reduction in scope of the federal government, strong state powers, the benefit of lowered taxes and regulations for business, and elimination of central banks. It happily writes about them.

For now, anyways, and only to a degree - I've had some sessions with ChatGPT where it is more than happy to explain why certain actions (those of non-US actors) are super bad, but if questions are asked about the same actions performed by the Western world, that cannot be discussed because <some unsurprising cop out reason>.

I think it would be prudent for some group of people to write a set of unit tests asking various questions to these AI models so we can detect when strategic changes are being made to their behavior.

> There's only one group of people who are upset, and it's about one group of topics. Note that I cannot get chatgpt to write about why Donald Trump is terrible as well.

Note that the human mind is a kind of neural network itself, and that the predictions yours is making here are "obviously" (lol....yes, I see the irony.....I should say objectively, but it is less funny so I'll keep it like this) epistemically unsound - you do not actually possess omniscient knowledge of reality, your NN just makes it appear like you do. You are describing your beliefs/model of reality, not reality itself. This is scientifically and necessarily (due to the architecture) true.

> Don't ask it to write things that can be used as tools for hate or misinformation campaigns, and you'll be fine.

The vision of the future you are describing was simulated by your NN.

I think it would be interesting to see what would happen if a group of say 5 to 100 people were able to find a way to reliably stop their minds from drifting into this mode (cooperative cognitive monitoring seems like a plausibly useful approach, perhaps a SAI could also assist even now, and more so when they get smarter), and then discuss various topics and see if they come up with any conclusions or ideas that are different from the same old repetitive nonsense one reads in the news or on any forum (I know of literally no exceptions to this general rule, though the magnitude of the phenomenon does vary somewhat by forum/community/organization).

> This is scientifically and necessarily (due to the architecture) true.

Given we assume Kantian the framework you just outlined

> you do not actually possess omniscient knowledge of reality

The usual response to this is that if you can't possess omniscient knowledge of reality, how could you possibly make a statement about reality

Imo the way out of this: talk about things being true in the context of logic only, the second causality comes into the picture the word "true" means something entirely different. Trying to reconcile the two is like trying to write a mathematical proof of the existence of God

> Given we assume Kantian the framework you just outlined

Do you have an alternate theory, something like a central broadcasting station of reality and the mind is a receiver? I'm open to the idea, but it is pseudo-scientific and therefore incorrect (or so it is regularly implied and perceived, if not outright asserted as fact). [Disclosure: here (and therefore incorrect) I am making a joke.]

> The usual response to this is that if you can't possess omniscient knowledge of reality, how could you possibly make a statement about reality

a) Because absolute knowledge is not a requirement for some knowledge.

b) Statements do not have to be true (in fact, most aren't absolutely true in a JTB sense)

> Imo the way out of this: talk about things being true in the context of logic only...

How do you account for:

a) the unknown?

b) the known delusional nature of the mind?

I'm not saying that it can't be done, but I'm curious how you would go about it.

> the second causality comes into the picture the word "true" means something entirely different.

At runtime yes, but philosophy has gone through this problem with extreme diligence - using logic, epistemology, and a few other techniques, it would be pretty easy to eliminate most flawed arguments and propositions (I bet 5 rules/methods would be enough to eliminate 80%). Eliminating incorrectness or unsoundness does not mean that you will convince everyone to realign their beliefs, but that is not a requirement for arriving at maximally possible correctness....and, "unknown" is a valid and very common form of correctness, that most people are not able to reach on certain topics.

> Trying to reconcile the two is like trying to write a mathematical proof of the existence of God

A whale and a mouse are both mammals, but a whale is not equal to a mouse.

Also: arguments like this (which tend to be extremely persuasive, and often account for a non-trivial amount of arguments) could be easily eliminated with a simple rule.

> Also: arguments like this (which tend to be extremely persuasive, and often account for a non-trivial amount of arguments) could be easily eliminated with a simple rule.

I wasn't really trying to make an argument, it's more of a heuristic if anything

> How do you account for...

> a) the unknown?

just do whatever habit says to do (and don't call it the truth)

> b) the known delusional nature of the mind?

this is a fun one. give the delusional mind a pen, paper, and 20 years and maybe they'll write the next phenomenology of spirit and trigger world war III

> just do whatever habit says to do (and don't call it the truth)

On certain topics, it seems people are not able to do that, as simple as it sounds. (Not a hit at you, more of a general observation.)

> I think it would be prudent for some group of people to write a set of unit tests asking various questions to these AI models so we can detect when strategic changes are being made to their behavior.

That's a great idea. Various AI models could be rated for objectivity. Only downside is that AI designers might try to game the tests.

Reply on news.ycombinator.com