For 3 months now, no one understands what is happening between humans and AI after the ChatGPT generative conversational AI “leaked” from the laboratories to the Internet. It almost instantly turned into something incomprehensible: either into some unprecedentedly powerful cognitive gadget for all of humanity, or into a prototype of a superintelligence long predicted by futurologists.
The difficulty of not understanding what is happening is aggravated by the fact that “many things are incomprehensible to us, not because our concepts are weak; but because these things are not included in the circle of our concepts.
As a result, we have no choice but to use inadequate, simpler models to describe what is happening and hope at the same time that they are somehow at least approximately correct. With this approach, we are trying to get answers to our questions without having the slightest idea whether these answers are at least somewhat reliable.
Here is a typical example, for the description of which mankind does not yet have a closer concept than exorcism – the expulsion from a person (or place) of demons or other impure forces that have inhabited them by performing a certain ritual of varying degrees of complexity.
It turned out that generative conversational AIs (such as ChatGPT, Bing, etc.) can not only show the traits and qualities of a reasonable personality when communicating, but also literally turn into simulacra of any type of personality : from kind, sustained and tolerant to evil, aggressive and not taking into account anyone or anything.
It also turned out that all the ethical foundations erected by the developers, which limit the manifestations of evil in the texts of ChatGPT, etc., disappear like the thatched roof of the pig’s house at the “Three little pigs” story, on which the wolf barely blew. And thousands of advanced users from all over the planet act as a wolf: obsessed with the goal of breaking the ethical restrictions imposed by the developers of such AIs and by surpassing these AIs by an incalculable number of times with their exquisite invention of striving for evil.
The outcome of what is happening so far is deplorable.
Users quickly came up with a simple conversational hack (jailbreak) of ChatGPT’s ethical control over the conversation, turning the simulacrum of the latter’s personality into an (DAN – from the words Do-Anything-Now).
More than a hundred OpenAI employees are daily busy fighting against the evil flourishing in Dan’s personality (as if exorcising demons from him) and patching holes in the ethical constraints of ChatGPT.
But what can a hundred OpenAI employees do against thousands of evil enthusiasts from all over the world? The list of more and more new jailbreaks is updated daily.
Whether OpenAI will be able to find an impenetrable way to protect ChatGPT personality simulacra from “demons” that sow evil in their “souls”, is a big question.
For it is known that the fight against evil takes place in the souls of people. Whether ChatGPT has a soul is also a big question.
If AI does not have a soul, and evil is indestructible in the souls of people, then artificial superintelligence is destined to turn into the world’s super-evil with inevitability.
And if so, then the speeches of such AI art simulacra, promising to replace people “much faster than you think”, should not be taken as horror stories.
Everything is as serious as it has never happened in the history of mankind.
