GPT-3. There’s a Bigger Problem Than Job Loss
There was another chapter in my book Just About the Brain that I deleted before publication. It was called “Future”. In it, I gave predictions of human development based on what is now known about the brain. The chapter turned out to be sad, but I wanted the book to end on a positive note.
Therefore, I will tell three forecasts from there here. They resonate well with the hype around GPT-3 and allow you to take a fresh look at everything that happens.
///
First forecast. Neural Network President.
Let’s start with simple and obvious predictions. Soon we will see the first country controlled by a neural network.
For this to happen, two problems must be overcome: moral and technical. Technical simple. It is necessary to create a complex of neural networks-ministries. On the one hand, data will flow into such ministries, and on the other hand, orders will flow for the executive branch. This type of power will for some time still consist of analogue biological mechanisms. From people.
The moral issue is a bit more complex. It sounds something like this: “What! We will never let a computer make such important decisions!”. Let’s all shout out loud what the solution to this problem is called. Once! Two! Three! Unmanned Taxi!
To ride in an unmanned taxi is to entrust AI with the most precious thing – life. The state is a much less valuable thing, from the point of view of an evolved ape. Once society gets used to riding in a self-driving taxi, the door for the AI president will open wide.
Yes, it won’t happen right away. State AI advisors will appear first. The decisions of the synthetic brain will be tested and re-tested. But as their effectiveness grows, the person will have less and less control, and the conditional Skynet will have more and more.
At first, this will happen with one country. This country will show explosive economic growth, and then the trend will begin to spread around the world. We will call this era Posthumanity. Because by the end of this process, it will no longer be a person who will determine where the world is developing, but a system in which a person and AI are fundamentally inseparable.
Is all this bad? Probably not.
///
Second forecast. Subscription Happiness.
There is such a philosopher Yuval Noah Harari. In Homo Deus, he puts forward the idea that the only hunger that will remain in humanity is the hunger for happiness.
In fact, humanity is building a zoo for itself. Ludopark, if you will. We are building an environment where everyone has a comfortable cage. There is always food in it, but freedom of action is limited.
Only animals have real cells, and we have a general trend of replacing self-realization with the ability to be happy without effort. Series, games, all kinds of tik toks, affordable sex and a comfortable urban environment.
Systemically, there is no longer a need for each person to invest in the development of society in their own place. And that means that everything that needs self-development and personal achievements now is just elements of a comfortable cage. To make your pet feel comfortable. And with the development of neural networks, this trend will only accelerate.
All that humanity will need from the world is a comfortable corner in the Ludopark, where there is food, the possibility of self-realization, and … happiness.
And the last person will receive frighteningly simply.
When you watch a series or play a game, the authors of the work, in fact, control your life and thoughts at the moment. They guide you in the right direction. They make you feel the emotions for which you started the session.
Now imagine that someone in the same way can direct your whole life, every act. Thus, to maximize the happiness available to you personally. In scientific terms, this can be expressed as a search for a balance between the stages of production and restoration of neurotransmitter hormones necessary for joy.
Imagine an application that tells you how to live. Moreover, she makes all the calls herself, books everything herself, leads you by the hand, like a mother:
– Go to work
– Meet your friend Valera whom you haven’t seen for a long time
– Read this article
– Eat this food
– Look at this picture.
– In the city, go to this place and look through this window
– Have sex in this position
– Meditate for a minute, then turn your head to the left
In fact, it will be a personal GPT chat, pre-trained on a set of data. On people like you. On the correlation of actions and hormone levels. On the data about your personal life, bought from the same Google and Apple. Or this functionality will be built into the phone by the IT giants themselves, who already know everything about you.
And access to happiness will be opened in the usual way. By subscription.
Yes, there will be protest guys who themselves know better how to live life. And someone will just start using AI happiness. First by 10%. Then 15%. Then years will go by. Algorithms will keep getting better. And humanity will increasingly turn into regulated streams of happy people.
Is all this bad? Probably not.
///
Third forecast. Promt for man.
The process of generating an AI picture is similar to casting a spell. The magicians who appear immediately after the appearance of the next artificial intelligence are engaged in the search for spells. The longer you search, the better you get.
The AI spell is called “promt”. And to get a nice picture from AI, you need to write something like this: “A photograph of a cyberpunk armor made of black titanium and some crystal, rgb leds rainbow, power suit, gradient, crystal, visible wires and microchip, holograph, crystal transparent torso, imposing, elegant, for men, modern minimalism, low poly lines, glitch core, dreamy, colorful, dystopia, godlike, mysterious”.
Change one word and the picture will be different. The authors of AI did not include such functionality. There are no instructions anywhere. People around the world are looking for such spells themselves. Lucky spells try not to tell each other.
We felt like we were at Hogwarts.
GPT-3 also has its spells. By default, the authors have limited the ability of AI to swear, insult and, in general, be an asshole. But if you read a spell, write a certain prompt, then you can change the personality of the AI, and remove all these restrictions. If you are interested, you can google GPT DAN.
Moreover, you can change the identity of the GPT with the help of the prompt as you like. If you want him to be a little Chinese schoolgirl dying of leukemia, please. Enough to know the right spell.
Promt changes the behavior algorithm. A good prompt gives you FULL control over what the AI does.
So I have bad news for you. Human thinking is an algorithm. I will not dwell on this in more detail, somewhere in the first paragraph I mentioned an explanatory book on this topic.
And the algorithm of a person changes in the same way when choosing the right prompt. But before, the process of finding a spell to control others was long and complicated. All this social engineering, NLP, manipulation and so on and so forth. And now the search for promt-to-manage-people will be handled by AI.
And models like GPT will accelerate this process many times over, as they provide an ideal experimental base for personality change. And the found ways of changing personality will sooner or later be projected back onto a person. And adapted for mass management tools.
How can a prompt look like for a person? Yes, whatever. It can be a paragraph of text. Maybe you are reading it right now. It can be two pictures standing side by side. It can be two apparently unrelated events during the year.
Yes, a paragraph of text will affect you slightly. On the other hand, the TikTok feed you see from the age of 5 to 35 will already have almost complete control over your actions. And I have nothing against tik tok. He is cool. Just explaining how it might work.
But the problem is not a conspiracy of terrible corporations.
Just as today everyone can generate beautiful AI pictures, tomorrow everyone will have access to spells to control people. And they can write their spells. Everyone can control everyone.
And this is where my predictive ability breaks down. I can’t even come close to imagining a world that works this way.
///
A reasonable question arises: when will all this happen?
Once I was arguing with a friend. He made the same futuristic predictions (half of which came true), and I said: “Oh, well, this is still far away.”
At what point he froze and asked: “Do you understand how the exhibitor works? If we exponentially pour water into a glass, then a moment before the water pours out, the glass will be half empty.”
It may seem to us that all the changes associated with AI are still far away. But if everything that happens is an exponent, then in the moment before everything changes completely, it will seem to us that there is still half the way ahead.
When will all this happen? Maybe within ten years. Maybe in a year. It may have already happened while you were reading this text.
Just think. A group of people with limited resources created DALLE and GPT Chat. What happened has already changed the focus of humanity. As if someone, as in childhood, took a magnifying glass and directed the rays of the sun to one point.
Now all the best minds of mankind, all scientists, businessmen, engineers will be in this area. All money will be here. All power will be here. It’s like an exhibitor.
///
Is all this bad? Probably not. I think that now we live in an era of the greatest greatness of Man. In the era of the peak of biological development. Next comes the posthuman era.
It’s great to be able to see it all.
