It has learned from countless general dialogues, as well as articles, news stories, books, and other digital texts describing the likes of Elon Musk, Queen Elizabeth, and William Shakespeare.
The system also has a way of combining disparate concepts learned during training. The result is a practically endless collection of bots that can imitate a virtually endless collection of humans, ranting about a virtually endless number of topics, as Mr. Thiel discovered when chatting with the Karl Kautsky bot:
JonasThiel: Marxists of your time were confident that revolution was a historical inevitability. In light of the recent defeats suffered by socialists around the world and the lack of class struggle even in times of great suffering, do you think it is time to reconsider that view?
Karl Kautsky: Not me. We may have made mistakes with our tactics, but I refuse to believe that socialism is a utopia. I still believe that the proletariat must break the shackles of capitalism and that the working class must unite around Marxism. We must not forget that capitalism is still deeply flawed and our job as socialists is to agitate and spread class consciousness.
JonasThiel: I totally agree, but wasn’t it wrong to assume that workers uniting to overthrow the bourgeoisie was inevitable? After all, the capitalist class is very powerful and it is difficult for workers to come together to resist collectively rather than individually.
Karl Kautsky: You’re right, it was a mistake to be too optimistic in that regard. But I think it is better to believe in class consciousness and revolution until proven wrong. What is the alternative? Believe that the workers will remain in chains forever? I tend to disagree. It is better to fight against capitalism, despite our chances, than to give in. I refuse to believe that the workers will never wake up. Even if I am too optimistic, we must not give in to pessimism. I refuse to be wrong. I hope you understand it.
Sometimes the chatbot gets things right. Sometimes not. When Mr. Thiel chatted with an avatar intended to imitate Mr. Reed, the 20th-century American political thinker, it turned him into “some kind of militant Maoist, which is absolutely not right.”
Like Google and OpenAI and other top labs, Mr. De Freitas, Mr. Shazeer and their colleagues plan to train their system with increasing amounts of digital data. This training can take months and millions of dollars; it can also sharpen the skills of the artificial interlocutor.
Researchers say the rapid improvement of recent years will only last so long. Richard Socher, former head of AI at Salesforce, who now runs a start-up called You.com, believes these exponential improvements will begin to level off in the coming years, when language models reach a point where they can replace virtually all text on the web. .
But Mr. Shazeer believes that the runway is much longer. “There are billions of people in the world who are constantly generating text,” he said. “People will continue to spend more and more money to train ever smarter systems. We are far from the end of that trend.”
