Zompist Bboard Again

rotting bones wrote: ↑Wed Jun 28, 2023 6:35 pm And how is it worse for responsibility to be abdicated to random AI than to random humans?

Because AI may be more competent and thorough in what it does.

hwhatting wrote: ↑Thu Jun 29, 2023 5:51 am Because AI may be more competent and thorough in what it does.

For all existing powers, competence in a nuclear war means not starting one. None of the corrupt and opportunistic leaders of today's world are serious about Mutually Assured Destruction. As for the AI, it has no independent opinions. Its goal is set by the programmer.

rotting bones wrote: ↑Thu Jun 29, 2023 6:18 am As for the AI, it has no independent opinions. Its goal is set by the programmer.

This may be overly optimistic. Even with the current state of AIs, programmers seem to have had great difficulty making them obey even the simplest of constraints, e.g. ‘don’t say racist things’.

bradrn wrote: ↑Thu Jun 29, 2023 6:26 am This may be overly optimistic. Even with the current state of AIs, programmers seem to have had great difficulty making them obey even the simplest of constraints, e.g. ‘don’t say racist things’.

That's because these AIs are being made to replicate online text. What scenarios can an AI be fed that will make them more likely to start a war than not?

When training an AI, it can learn prejudices that manifest in the training data. There is a famous case where an AI which was to find promising job candidates chose only white males. It had been trained with real career data - which showed that white males were preferably hired and promoted by managers, and thereby learned that females, nonbinary people and people of colour performed less well.

WeepingElf wrote: ↑Thu Jun 29, 2023 10:44 am When training an AI, it can learn prejudices that manifest in the training data. There is a famous case where an AI which was to find promising job candidates chose only white males. It had been trained with real career data - which showed that white males were preferably hired and promoted by managers, and thereby learned that females, nonbinary people and people of colour performed less well.

This. AI's can be extremely, well, competent at picking out subtle biases, ones often entirely unintended by their creators, in what they are trained with and amplifying them. Racist and sexist AI's really are a thing.

Travis B. wrote: ↑Thu Jun 29, 2023 10:48 am This. AI's can be extremely, well, competent at picking out subtle biases, ones often entirely unintended by their creators, in what they are trained with and amplifying them. Racist and sexist AI's really are a thing.

They're not picking up on subtle biases. 4chan is about as unsubtle as you can get. Their idea of subtlety is to use an old rape joke as the Gamergate mascot. This is the problem with AI racism:

rotting bones wrote: ↑Thu Jun 29, 2023 6:29 am

bradrn wrote: ↑Thu Jun 29, 2023 6:26 am This may be overly optimistic. Even with the current state of AIs, programmers seem to have had great difficulty making them obey even the simplest of constraints, e.g. ‘don’t say racist things’.

That's because these AIs are being made to replicate online text. What scenarios can an AI be fed that will make them more likely to start a war than not?

WeepingElf wrote: ↑Thu Jun 29, 2023 10:44 am When training an AI, it can learn prejudices that manifest in the training data. There is a famous case where an AI which was to find promising job candidates chose only white males. It had been trained with real career data - which showed that white males were preferably hired and promoted by managers, and thereby learned that females, nonbinary people and people of colour performed less well.

Like I said, how would this translate to nuclear war?

rotting bones wrote: ↑Thu Jun 29, 2023 3:08 pm

Travis B. wrote: ↑Thu Jun 29, 2023 10:48 am This. AI's can be extremely, well, competent at picking out subtle biases, ones often entirely unintended by their creators, in what they are trained with and amplifying them. Racist and sexist AI's really are a thing.

They're not picking up on subtle biases. 4chan is about as unsubtle as you can get. Their idea of subtlety is to use an old rape joke as the Gamergate mascot. This is the problem with AI racism:

Biases need not be overt for AI to pick up on them, as latching onto subtle patterns is one of the things that AI is best at. This is a major problem with AI, as training data often has implicit biases which gets trained into the AI, no matter what the intentions of the AI's creators were. Consequently you get things such as AI's which discriminate against patients on the basis of race based upon seeing X-rays of patients despite never being told what the patients' race was.

Travis B. wrote: ↑Thu Jun 29, 2023 3:32 pm Biases need not be overt for AI to pick up on them, as latching onto subtle patterns is one of the things that AI is best at. This is a major problem with AI, as training data often has implicit biases which gets trained into the AI, no matter what the intentions of the AI's creators were. Consequently you get things such as AI's which discriminate against patients on the basis of race based upon seeing X-rays of patients despite never being told what the patients' race was.

But the patterns of discrimination based on race are super obvious to anyone who looks at the data. Plus, body shape is a major characteristic by which people are put into racial boxes, whereas no one today really wants a nuclear war.

rotting bones wrote: ↑Thu Jun 29, 2023 6:18 am

hwhatting wrote: ↑Thu Jun 29, 2023 5:51 am Because AI may be more competent and thorough in what it does.

For all existing powers, competence in a nuclear war means not starting one. None of the corrupt and opportunistic leaders of today's world are serious about Mutually Assured Destruction. As for the AI, it has no independent opinions. Its goal is set by the programmer.

It goes like this:
Corrupt and opportunistic leader: - Program me something that recognizes our enemies and destroys them.
Programmer: - Yes, sir!

hwhatting wrote: ↑Fri Jun 30, 2023 5:32 am It goes like this:
Corrupt and opportunistic leader: - Program me something that recognizes our enemies and destroys them.
Programmer: - Yes, sir!

This leader would have to be unable to recognize that any nuclear strike would likely result in nuclear retaliation.

rotting bones wrote: ↑Sat Jul 01, 2023 4:26 pm

hwhatting wrote: ↑Fri Jun 30, 2023 5:32 am It goes like this:
Corrupt and opportunistic leader: - Program me something that recognizes our enemies and destroys them.
Programmer: - Yes, sir!

This leader would have to be unable to recognize that any nuclear strike would likely result in nuclear retaliation.

For being usually so cynical, you have an astonishing lot of faith in humanity.

This Mastodon find might be relevant for this thread:

https://mastodon.social/@MissingThePt/1 ... 7589769773

It began with the forging of the Great Platforms. 3 were given to the Blueskies, self-assessed wisest and fairest of all. 7 to the Twitter Edgelords, great memers and alt-right craftsmen. And 9, 9 rings were gifted to the race of Mastodon, who above all else desire content warnings. But they were all of them deceived, for another platform was made. Deep in the land of Meta, in the Fires of IG, the Dark Lord Zuck forged a federated platform, and into this he poured his cruelty, his malice and his

Any comments on Elon Musk actually proposing a literal penis length measuring contest with Mark Zuckerberg?

Our global elite. The people who made it to the top because of their scintillating intellectual brilliance.

When I was a child, I spake as a child, I understood as a child, I thought as a child: but when I became a man, I did not put away childish things, because my name is Elon Musk.

Raphael wrote: ↑Mon Jul 10, 2023 3:02 pm Any comments on Elon Musk actually proposing a literal penis length measuring contest with Mark Zuckerberg?

"I can measure more literal penises than you so clearly I am superior to you by any imaginable measure".

hwhatting wrote: ↑Mon Jul 03, 2023 5:26 am

rotting bones wrote: ↑Sat Jul 01, 2023 4:26 pm

hwhatting wrote: ↑Fri Jun 30, 2023 5:32 am It goes like this:
Corrupt and opportunistic leader: - Program me something that recognizes our enemies and destroys them.
Programmer: - Yes, sir!

This leader would have to be unable to recognize that any nuclear strike would likely result in nuclear retaliation.

For being usually so cynical, you have an astonishing lot of faith in humanity.

If you have a nuclear power with a leader who's blase about nuclear war, don't you think you have bigger problems than AI?

If you're cynical about human behavior, what makes you think the presence or absence of AI will have any effect on our decision to blow ourselves up?

rotting bones wrote: ↑Sun Jul 16, 2023 2:22 am If you have a nuclear power with a leader who's blase about nuclear war, don't you think you have bigger problems than AI?

If you're cynical about human behavior, what makes you think the presence or absence of AI will have any effect on our decision to blow ourselves up?

In my experience, people are much more prone to initiate systems that may have unfortunate consequences (because they don't think them through or because the seeming benefits are now and the negative consequences come later and aren't certain, or because of wishful thinking), than to directly produce the unfortunate consequences. The leader may think that he can control the AI, until something goes wrong; yes, it's a problem that there is such a leader, but with the AI an extra layer where things can go wrong has been added to that problem.

Last week, I did my first Mastodon poll. Given that I only have a handful of followers over there, that was probably presumptuous, but after one of my followers boosted it, I still got 12 participants. Not many, but more than I had expected.

The question was "Do old style web forums count as a form of social media?"

And what surprised me a bit was that all 12 respondents responded with "yes". Usually I wouldn't put much stock into a sample size of 12, but when the result is so lopsided - 100 percent for one response - I still find it interesting.

What do you think?

Raphael wrote: ↑Tue Sep 05, 2023 6:10 am Last week, I did my first Mastodon poll. Given that I only have a handful of followers over there, that was probably presumptuous, but after one of my followers boosted it, I still got 12 participants. Not many, but more than I had expected.

The question was "Do old style web forums count as a form of social media?"

And what surprised me a bit was that all 12 respondents responded with "yes". Usually I wouldn't put much stock into a sample size of 12, but when the result is so lopsided - 100 percent for one response - I still find it interesting.

What do you think?

I’d absolutely agree. It’s web media used for socialising — admittedly in a slightly different way than more modern social media, so there was a time when I would have said ‘no’, but it’s definitely still socialising.