# DeepMind’s New AI Learns Gaming From Humans! 🤖

## Метаданные

- **Канал:** Two Minute Papers
- **YouTube:** https://www.youtube.com/watch?v=CFT-2soU508
- **Дата:** 13.05.2022
- **Длительность:** 5:54
- **Просмотры:** 88,053
- **Источник:** https://ekstraktznaniy.ru/video/13567

## Описание

❤️ Check out Lambda here and sign up for their GPU Cloud: https://lambdalabs.com/papers

📝 The paper "Learning Robust Real-Time Cultural Transmission without Human Data" is available here:
https://www.deepmind.com/research/publications/2022/Learning-Robust-Real-Time-Cultural-Transmission-without-Human-Data
https://sites.google.com/view/dm-cgi 

❤️ Watch these videos in early access on our Patreon page or join us here on YouTube: 
- https://www.patreon.com/TwoMinutePapers
- https://www.youtube.com/channel/UCbfYPyITQ-7l4upoX8nvctg/join

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Aleksandr Mashrabov, Alex Balfanz, Alex Haro, Andrew Melnychuk, Angelos Evripiotis, Benji Rabhan, Bryan Learn, B Shang, Christian Ahlin, Eric Martel, Gordon Child, Ivo Galic, Jace O'Brien, Jack Lukic, Javier Bustamante, John Le, Jonas, Jonathan, Kenneth Davis, Klaus Busse, Lorin Atzberger, Lukas Biewald, Matthew Allen Fisher, Michael Albrecht, Michael Tedder, Nev

## Транскрипт

### Introduction []

Dear Fellow Scholars, this is Two Minute  Papers with Dr. Károly Zsolnai-Fehér. Today we will see how and what DeepMind's AI is  able to learn from a human, but with a twist!    And the twist is that we are going to remove the  human teacher from the game, and see what happens.    Is it just memorizing what the human does,  or is this AI capable of independent thought? Their previous technique did something like  that, like guiding us to the bathroom, start   a band together, or even find out the limitations  of the physics within this virtual world.    However, there is a big difference. This  previous technique had time to study.

### Project Overview [0:50]

This one, not so much! This new one  has to do the learning on the job. Oh yes, in this project, we will see an  AI that has no pre-collected human data.    It really has to learn everything on the  job. So, can it? Well, let's see together!

### Phase 1 Gameplay [1:11]

Phase 1 - Pandemonium. Here, the  AI says, well, here we do…things,   I am not so sure. Occasionally, it gets a point,  but now, look, uh-oh! The red teacher is gone.    And…oh boy, it gets very confused. It does  not have much of an idea what is going on. But, a bit of learning happens, and later.   Phase 2 - Following. It is still still

### Phase 2 Gameplay [1:42]

not sure what is going on, but  when it follows this red chap,   it realized that it is getting a much  higher score. Before, it only got 2, now,   look at it! It is learning something new  here. And, look! He is gone again, but   it knows what to do. Well, kind of. It still  probably wonders, why its score is decreasing?

### Phase 3 memorization [2:13]

So, a bit later, Phase 3 - Memorization. First,  the usual dance with the demonstrator. But, now,   when he is gone, it knows exactly what to  do, and just keeps improving its score. And then comes the crown jewel. Phase  4 - Independence. No demonstrator

### Phase 4 independence [2:31]

anywhere to be seen. This is the final  exam. It has to be able to independently   solve the problem. But here comes the twist.   For the previous phase, we said memorization,   and for this, we say independence. Why is this  different? Well, look. Do you see the difference?

### Results [2:55]

Hold on to your papers, because we have switched  up the colors. So, the previous strategy   is suddenly useless. Oh yes! If it walks the same  path as before, it will not get a good score,   and initially, that's what it is trying to  do. But, over time, it is now able to learn   independently, and indeed, find  the correct order by itself. And, what I absolutely loved here is, look.   Over time, the charts verify that indeed,   as soon as we take away the teacher, it  starts using different neurons as right   after becoming an independent entity.   I love it. What an incredible chart.

### Conclusion [3:48]

And all this is excellent news. So, if  it really has an intelligence of sorts,   it has to be able to deal with  previously unseen conditions and   problems. That sounds super fun  - let’s explore that some more. Let’s give it a horizontal obstacle.   Good! Not a problem. Vertical?    Also fine. Now, let’s make the world larger! And,  it is still doing well. Awesome! So, I absolutely   love this paper. DeepMind demonstrated that  they can build an AI that learns on the job,   one that is even capable of independent thought,   and even better, one that can deal with unforeseen  situations too. What a time to be alive! So, what do you think? What  else could this be useful for?    What do you expect to happen a  couple more papers down the line?    Please let me know in the comments  below. I’d love to hear your thoughts. Thanks for watching and for your generous  support, and I'll see you next time!