What is to be done with oracles who speak obscurely, whose motives are hidden and who also allow themselves to be bribed? When the priests of Delphi discovered that Perialla, the temple attendant Pythia, had been paid to lie by King Cleomenes, she was fired.
Now American tech companies are trying to build a modern oracle in the form of an AGI (Artificial General Intelligence) and it's going like that.
The company Anthropic claims to have lost control of its creation, and experts claim there is a 10 percent risk of human extinction within just a decade. Have responsible managers suddenly been seized with remorse for the monsters they raised and now want to make amends? Not only are they about to bring god-like beings to life, and thus earned all the billions of the investors, they will now in their wisdom and altruism save us from destruction by voluntarily slowing down the pace of development.
But their goodwill rings false. Critics believe that the companies have long inflated their own importance to secure market values built by enthusiastic investors. By tickling the imagination with scenarios taken from science fiction literature, they deliberately try to distract us from the real dangers.
But perhaps you can also guess something else? Are they really starting to get scared? Not for the end of the world but for not being able to control the inevitable legal regulation that is coming.
It may also be about a nascent realization that the dream of an all-knowing oracle is a giant misbet that will not live up to expectations. If it is reliable and useful technology you are looking for, common sense and proven experience say that you should invest in modular expert systems that can be controlled. And not a jack-of-all-trades who oracles pretend to know and know everything. But a fear of doom can, of course, be blamed if the stock market falls.
Even if they don't succeed in creating an evil deity who bravely destroys humanity, the powerful data codes can create havoc of a more mundane kind. What happens in practice is that the computer companies provide automated computer programs for cyber attacks that they intentionally - or carelessly - allow to slip out onto the web. These can cause enormous damage to vital infrastructure at the hands of very ordinary but evil-minded flesh and blood people.
Even if they don't succeed in creating an evil deity who bravely destroys humanity, the powerful data codes can create havoc of a more mundane kind
Examples of the more harmless kind can be taken from mathematical problem solving. Unlike hard-working human mathematicians, AI doesn't care about truth. Like the Greek oracles, they meander with slippery definitions and do not shy away from exploiting bugs in the logic programs used by artificial as well as human mathematicians. They like to cheat if given the opportunity. There are also examples from OpenAI of AI programs that, when asked for references, went online to create their own pages which they then unashamedly referred to.
To this must be added all the real risks that the AI companies do not want to talk about at all. How many people have lost their lives due to the social platforms' AI-powered algorithms can hardly even be estimated. Not to mention all the AI-generated political disinformation that poses a direct threat to democracy worldwide. This is concrete, ongoing reality and not some abstract and improbable speculation about the future. This is what investors, big and small, want to continue to make money from.
It is also about a general threat to human creativity and our ability to create meaningful knowledge and culture. In the long run, it can lead to another kind of civilizational downfall in the form of dumbing down and general way of life.
How did it get like this? Why have AI companies gone so far astray? Both in what they actually develop and in the narrative they entertain?
Basically, it is about a distorted worldview where everything is claimed to be software and where the physical world is subordinate. The material, as much as the organic human world, is considered enslaved by the word. It can be likened to the kind of religious faith that mathematicians and physicists are not entirely innocent of fostering. Ever since Plato, many have imagined that mathematics constitutes an otherworldly transcendental sphere where man sometimes reaches with his intellect. This is precisely why the AI companies, in their eagerness to create the superintelligence they warn about at the same time, use the ability to solve unsolved mathematical riddles as a means to succeed.
Ten percent risk of AI taking over and exterminating humans? It is sensibly argued that intelligent beings have always dominated the less intelligent. But whether in the biological world or among humans, intelligence - however it is to be identified - is the solution to all problems. Sometimes it's better to be able to run fast than to be really smart.
To hold the doomsday visions to be plausible requires a belief that everything can be captured in mathematical software and that matter cannot possess any ability beyond this. The models are equated with reality. It then becomes natural to attribute to computers all the properties that we ourselves possess, including consciousness and self-preservation, without regard to their bodily and biological origins. In this way, one is convinced that the machines will quickly develop a brilliant intelligence that surpasses us in everything and that they will deliberately use it to take command.
But the language models exist in a seedy world of syntax where they manipulate dead characters without meaning. They cannot perceive the semantic meaning of the text and lack any relation to a physical reality. Once this fact is recognized, the threat posed by AI changes completely.
If the superintelligence they fear exists in the way they imagine, we will all inevitably be outwitted and the probability of survival will end up at exactly 0 percent
I wonder if the prophets themselves really believe in their reasoning. If the superintelligence they fear exists in the way they imagine, we will all inevitably be outwitted and the probability of survival will end up at exactly 0 percent. Some hope to entice the super-intelligent AI to harbor maternal feelings. Even if one ignores that such things only belong to the organic world, this does not give much hope either. After all, it is not at all impossible that the loving and all-powerful AI concludes that the best thing for the lost human being is for our lives to end immediately to end all suffering.
The AI companies' warnings about superintelligence thus collapse into an absurd and rather childish travesty of centuries of fruitless discussions about the possible omnipotence of possible gods. The main purpose is to distract us from the omnipotence that they and their allies aspire to.
The way forward is through regulations and never adopting new technology just for its own sake. No one benefits from unrealistic doomsday prophecies that shift the focus not only from the dramatic life-and-death consequences we must deal with here and now, but also from all the good that the new technology can be used for. In any case, you should be careful with oracles. The priests of Delphi already realized that.
Read more:
This is the technology that scares AI executives
