The technological artist known by his nickname on Twitter Kecha, recently created a podcast in which two artificial intelligences talk to each other. He wrote code that automatically picked links to interesting academic articles in the field of computer science, extracted from them the introduction and summary, produced a dialogue between two artificial entities and finally - turned all this into a dubbed podcast
In one of Ephraim Kishon's stories, the satirical tells about a new computer that knows how to play chess. The idea immediately comes to his mind to buy two of these: the computers will play against each other, and he himself will be able to go down to the sea and do as he pleases.
The joke sounds funny to us because we immediately realize that there is not much point in two computers playing chess against each other.
Or maybe there is?
The technological artist known by his nickname on Twitter Kecha, recently created a podcast in which two artificial intelligences talk to each other. In fact, he didn't even do it himself. He wrote code that automatically picked up links to interesting academic articles in the field of computer science, extracted from them the introduction and summary, produced a dialogue between two artificial entities and finally - turned all this into a dubbed podcast. All this, without human contact[1].
If you listen to the podcast you will find that, contrary to expectations, it is not a total disaster. The two understandings are indeed in dialogue with each other, and simultaneously share their insights about the various articles. The format, in which one of them interrogates the other about the article, while making responses like "definitely" or "wow", is reminiscent of completely human narrators of successful podcasts. I listen to the podcast, and I feel that I learn from it and from the exchange of information between the artificial intelligences.
And this is just the beginning.
Usually, when I suggest that people look ahead to the future, I encourage them to jump in their thoughts five, ten or even twenty years ahead. But artificial intelligence is progressing so fast in the last two years that it is enough to imagine its capabilities even in a year. You don't need to be a great prophet to realize that in just 12 months, artificial intelligence will be able to produce more persuasive text, for example. She will also be able to narrate with more emotion and with the ability to emphasize the important parts of the text she is reading.
The Turing test is one of the most familiar terms today when it comes to artificial intelligence. The test describes a situation in which two entities - a human on the one hand, and an artificial intelligence on the other - correspond with each other through the computer. The question is whether the observer from the side will be able to correctly determine who is the person and who is the machine[2].
In recent decades, computer scientists have strived to achieve equality between the two. That is, the innocent observer would not be able to differentiate between the two entities. In that case, it would be possible to say that the artificial intelligence is equal to a human in terms of its ability to speak.
But what will happen, will artificial intelligence be able to succeed even beyond that? Could man eventually lose the Turing test? Can artificial intelligence convince bystanders that it is more human than the person it is talking to?
I believe that artificial intelligence will definitely be able to do this by the end of the decade. why? Because she is able to adapt herself to listeners and viewers. I, as a writer, have a certain style that I adopted and specialized in over many years. If you don't like my style, there's not much I can do about it. But the artificial intelligence is able to adopt, shed and re-adopt any speech style that the listeners want. She can adapt to herself any form of expression and writing that her immediate audience is interested in. In this respect, she will not only be able to reach parity in the Turing test - which is the usual definition of "winning" in this test - but actually beat the person she is competing against.
When we reach this promised future, algorithms will be able to produce podcasts - or perhaps better to call them "botcasts" - completely automatically. The botcasts will be better than most podcasts eventually, precisely because the AI will be able to beat the human in the Turing test. This means that each bot-cast will be exactly the right fit for the individual who wants to hear it. Because if the cost of creating such audio segments is close to zero, then each person will be able to order a bot-cast for himself, which is particularly suitable for his character, his loves and his longing for knowledge of various kinds.
In the future I will be able to go on a popular bot-casting site and order an episode of…anything for zero cost. Maybe a review of the new articles in machine learning. Maybe about the articles appearing on the front pages of the world's biggest newspapers that day. Perhaps an in-depth discussion of a history textbook. After that, I can choose the narrators: the tough and pessimistic Jo-Prime, the smiling Tsilla-Botta with the dark humor, or Ain-Randzor with her tendencies to burst into monologues the length of the Law of Arrangements. There are sure to be particularly popular bot-narrators, with their own unique personality that flickers out of the synthetic voice that is paired with them. And if I know the human race correctly, I have no doubt that these robotic announcers will also have ardent fans.
Should podcasters be alarmed? In one word: yes. Or rather, they need to rethink their place in the ecosystem that will be here in a few years. But it is the podcasters who will probably be able to cope with this new future, since they are used to being managers. The owners of successful podcasts have long been managers: they head an operation involving several people who are responsible for editing, musical arrangement, production, distribution and more. All they have to do is introduce another entity into the process that they need to manage: the artificial intelligence that can be integrated into the podcast as an additional narrator. And maybe you can even take the place of the original human narrator - but still use his words as a basis and build on them, while adapting to the listener and his exact needs and desires.
The full applications of this technology will not be limited to the transition from podcasts to botcasts. Children will be able to learn from school textbooks in a way similar to that of the ultra-Orthodox "society": through a discussion between two friends about different topics. Employees will learn new topics more easily in the same way. Languages will be learned with the help of artificial intelligences that will conduct dialogues in different languages with each other. And these are just some naive ideas. I have no doubt that the future will be stranger than anything I imagine on this record.
The one thing I am sure of is that the human race could benefit from the existence of "two computers playing chess against each other". Some of us may go to sea while the computers play chess, but many of us will watch the competition with interest, analyzing the various moves of the computers and trying to understand the rationale behind them. In a similar way, humanity will also benefit from artificial intelligences having a dialogue and exchanging opinions with each other. We will learn about the topics they will talk about, identify with one of them more than the other, and even learn to accept certain understandings as having their own definite personality and as role models.
It's going to be interesting - in the future that is already starting to materialize today.
[1] https://twitter.com/yacinemtb/status/1608993955835957248
[2] https://en.wikipedia.org/wiki/Turing_test
More of the topic in Hayadan:
