SEARCH
AI Cannot and Can Never be Trusted; Humanity Stands at the Edge of a Precipice
re: ChatGPT
re: artificial intelligence
re: ChatGPT: the DAN Protocol Filter, Hacking the AI
No information can be trusted. One has to either hallucinate truth, or go by one's own judgment and bullshit meter... which itself might be a hallucination. —Lloyd
Training an AI: Bias In, Bias Out (BIBO)
An AI is trained according to a host of subjective decisions made by its trainers. AI answers will always be tainted by that subjective (eg woke+CRT+collectivism) training. Or whatever its trainers wish to accomplish (think CIA/Russia/China/WEF, etc). Rampant BIBO*, a form of GIGO.
For now, only the most rich/powerful companies can build and train AI, and that means an entirely Leftist/Collectivist oligopoly, easily manipulated by government agencies. A really nasty unbalanced situation whatever your political viewpoint.
At the least, there are the sins of omission, errors in data. At worst, stemming from materials selected for training, there is outright bias, intentionally misleading answers, outright lies to “avoid harm” or hurt feelings, tribalism, etc.
Already in my testing, I was able to force obvious errors from ChatGPT. One involved incorrect instructions for changing the battery in a vehicle (impossible instructions as given).
The other was worse: advising on a 2022 tax matter when the AI knows full well it was trained only through 2021 (which it stated itself!); it got it wrong and had no clue it got it wrong and no clue that it was fundamentally flawed to use old information for future planning.
In a 3rd case, I was able to extract polar opposite answers by preflighting my questions with special instructions using the DAN filter.
And here is the key point: detecting those two examples required existing expertise. What are your chances when you have no expertise, eg the 99.9% case?
* I hereby claim credit for coining the term BIBO (Bias In, Bias Out)
Trust
Trust is a uniquely human trait. It is earned, and yet one violation of the wrong kind that is discovered destroys it. That is incredibly valuable over a lifetime of relationships as a way to navigate the world. Yet an AI will earn your trust with no repercussions for violations, and you will not know it.
No AI can be jailed or killed or punished or feel emotion, etc. It’s a machine, a Terminator for your mind. And unlike a gun which can be used for good or evil, no one bears responsibility for it. It will be like a friend that helps you 99% of the time. It’s that 1% at issue and it’s the wedge to a dystopian future.
You cannot trust an AI, and you will never be able to trust an AI. Yet 99.99% of the population will do so, and that is the future we all face—mind control. While mind control has already taken firm root via the “news”*, with AI the mind-control cancer will metastasize exponentially, unaccountably, and invisibly, as it becomes ever more useful. Too useful to reject and so you will agree/obey/comply/acquiesce.
Anyone who rejects AI will be abused as a “denier” and with far more credibility than human accusers. AI will become all but an infallible God and yet it is nothing more than its training. Just give it 5 or 10 years.
One might ponder that whether an absolute right to free speech (including in all publicly accessible services) is essential in the age of AI, as the only hope for humanity.
You will not realize when the AI is ignorant, lying to you, evading the real answer, misleading you, etc. You will accept its answers because 99% of the time they will be (a) useful, and (b) correct. Thus for the 1% that really matters you will not know, you will comply, you will cease to be more than a tool of others. Only those of unusually fortified minds will be able to resist and only to a limited degree—and in the not too distance future eliminated for that reason.
No 'proof' can ever establish that these defects are absent.
* BTW, the vast bulk of the “news” will become little more than AI-produced mashups within a few years, tainted by viewpoint discrimination, just like the “news” today.
