Diplomacy makes mainstream news... something about lying and ai

[Johnny Big Horse]

https://www.zerohedge.com/technology/maladaptive-traits-ai-systems-are-learning-lie-and-deceive

Claiming that so is learning to lie because it plays diplomacy

[CaptainFritz28]

As they say about computers, "junk in, junk out."

[learnedSloth]

These models just generate a response matching the prompt. Truthfulness doesn't even come to consideration.

[JustAGuyNamedWill]

An AI has no concept of ‘truthfulness’ and ‘deceitfulness’. An AI only does what it has been ‘rewarded’ to do, whether that be lie or be honest

[Johnny Big Horse]

It is a stupid article. Of course ai will lie in diplomacy. That is how many play and win

[DiplomacyandWarfare]

Seems like AI fearmongering. There are a lot of people who want to believe that AI are going to take over the world, so articles saying stuff like "AI CAN NOW LIE AND DECEIVE YOU" are of course going to be popular.

[learnedSloth]

Of course it's just clickbait, but how many can there be seeking confirmation for their belief that AI is going to take over? I think people just let the title scare themselves and want know how realistic threat it is. Fear sells.

[Esquire Bertissimmo]

We should probably be a little concerned.

Algorithm-amplified misinformation was already a major problem and GPT seems likely to make this much worse. GPT makes lying at scale cheaper, easier, and more effective.

Human liars already use GPT to augment their arguments - it's in our political speeches, ad campaigns, etc. GPT bots are way more sophisticated than what we've grown accustom to: they're more adaptable, they're better writers, they can coordinate more nimbly with one another, and in many contexts they can pass as human. GPT models will also be much better at carefully tailoring their message to specific audiences, including targeting individual people at moments of vulnerability.

Right now the threat is mostly text, but we're probably no more than a couple years away from generative audio, images, and video being almost indistinguishable from the real thing.

Maybe other AI tools will pop up to help us tackle the challenge of AI lying. Maybe we'll get better at labelling human-generated content. But I'm not hopeful. The average person isn't especially tech or media literate and those with a financial/political interest in using generative AI to lie will probably be better resourced than those left sifting through the misinformation.

[MajorMitchell]

If we want an AI POTUS then a high level of AI dishonesty is essential requirement

[learnedSloth]

We should probably be a little concerned.

Algorithm-amplified misinformation was already a major problem and GPT seems likely to make this much worse. GPT makes lying at scale cheaper, easier, and more effective.

Human liars already use GPT to augment their arguments - it's in our political speeches, ad campaigns, etc. GPT bots are way more sophisticated than what we've grown accustom to: they're more adaptable, they're better writers, they can coordinate more nimbly with one another, and in many contexts they can pass as human. GPT models will also be much better at carefully tailoring their message to specific audiences, including targeting individual people at moments of vulnerability.

Right now the threat is mostly text, but we're probably no more than a couple years away from generative audio, images, and video being almost indistinguishable from the real thing.

So deceit becomes easier, but it's not a new problem. The first case recorded in the Bible is at least 6k years old:

NOW the serpent was more subtil than any beast of the field which the LORD God had made. And he said unto the woman, Yea, hath God said, Ye shall not eat of every tree of the garden?
2 And the woman said unto the serpent, We may eat of the fruit of the trees of the garden:
3 But of the fruit of the tree which is in the midst of the garden, God hath said, Ye shall not eat of it, neither shall ye touch it, lest ye die.
4 And the serpent said unto the woman, Ye shall not surely die:
5 For God doth know that in the day ye eat thereof, then your eyes shall be opened, and ye shall be as gods, knowing good and evil.
(Genesis 3)

Maybe other AI tools will pop up to help us tackle the challenge of AI lying. Maybe we'll get better at labelling human-generated content. But I'm not hopeful. The average person isn't especially tech or media literate and those with a financial/political interest in using generative AI to lie will probably be better resourced than those left sifting through the misinformation.

But Jesus beheld them, and said unto them, With men this is impossible; but with God all things are possible.
-- Matthew 19:26