- Tell us a bit more about the AI development process.
Victor Shkurba: To be clear, we’re not talking about sci-fi artificial intelligence. Right now, we are creating a communication interface with a database of knowledge. Two different phenomena fall into this category: an ordinary Q&A algorithm, and a fully functioning AI which communicates with you through meaningful thoughts, just like a human. Our goal is to offer something in between.
Andrey Sergeev: We created two identical neural networks. The same algorithms and operation principles are at the basis of both. Both neural networks can analyze information and connect meanings. It is possible they are first teaching themselves and then answering users’ questions. The topics for the dialogues are limited, though.
- What’s the difference between the two neural networks?
V.S: The difference is the information that teaches the networks. The code is identical.
- You mentioned that the neural networks teach themselves. How does that happen?
V.S: At first, we upload big amounts of transcribed TV-programs into the neural network. Then, it decodes the words into vectors. Here’s an example to understand the process better. If you take the vector image “Moscow” and add the vector image “Russia”, you get the concept “capital”. Following the same principle, “Moscow” minus “Russia” plus “Great Britain” equals a vector image close to “London”. Examples like
this help the system to start analyzing the interconnections in our speech. In short, it starts to understand human speech. The bigger the dataset is, the more information the neural network gets, and the more extensive its multidimensional space of interconnections. This is how it learns. This is how we do this, too. We create connections between objects and notions around us.
- That’s true. But, apart from the outside data, we have our experience that helps us estimate the information we receive.
V.S: Yes, and that’s just it. The neural network has no experience. There are no outside factors influencing the perception of information. The neural network seems to be consuming information as is comes, from
A.S: It’s a kind of Tabula Rasa. In our case, it’s Tabula Russia, because both neural networks, both interfaces, can answer only questions about Russia – about its political, economic and cultural life. The difference is that the first neural network will be basing its answers on the information that it received from a state TV channel, and the second will be using the knowledge it received from watching TV Rain.
V.S: Therefore, it’s a kind of a metaphor of immature fresh young mind. Imagine two twins deprived of all their experience. We have formed their mindset from scratch. In one case, we used TV Rain for that, in the other – a state TV-channel.
- Is this experiment supposed to prove that it’s harmful for a young mind to watch state TV- channels?
V.S: We’re not trying to prove anything. We’re scientists. We're experimenting. What results will we achieve? How mature will the young minds get? How different will they become? What will their world views be like? Honestly, we don’t know anything yet and that's the most fascinating part of experiment.
Talk with the intellects : https://aiversus.ru