Last few years, go players all over the world have observed a great change. A new era has come – an era of artificial intelligence (AI). It has had a great impact on the way people play and study go.
Software tools such as Leela Zero in combination with its user interface (e.g., Lizzie) provide a great opportunity for learning. The biggest problem is that for running such software you need powerful hardware.
About a year ago I was at the European Go Congress 2019 (Brussels, Belgium), and didn’t have my laptop with me. At the same time, I felt like I’d like to review my games with AI. Then I remembered that Benjamin Teuber 6d from Germany had introduced his new project AI Sensei to me. The idea is simple: you upload your game record to the website, the AI analyses it, and then you can see the report with all of your mistakes and how you had to play instead. Leela Zero (and KataGo) are running at the remote server, which means you don’t need powerful hardware: you only need to see the report. I tried to use it on my mobile phone and was surprised by how comfortable and useful it was.
Let me share with you an interview with Benjamin Teuber 6d (Germany), the founder of the AI Sensei project.
Q: Please explain briefly what AI Sensei is?
A: AI Sensei is an automated game review service, i.e. you upload your games to the website and it shows you all the mistakes and what you could have done differently. KataGo is being used per default, we also support Leela Zero. AI Sensei has a free version that is very useful already, as well as paid subscriptions for higher quality and deeper AI variations.
Q: How did you come up with an idea of starting such a project?
A: The first time I lost to my computer – it was against Leela (not Leela Zero yet) – I immediately thought that this might be useful for game reviews. So my friend Erik and I coded a quick prototype that analysed games. It was interesting, but back then it misread a lot of situations, so we didn’t continue this project. About a year later, Leela Zero reached pro level, so we tried again and were much more satisfied with the results. So we turned it into a full website that we first opened to the public at the European Go Congress 2018. Shortly after, I decided to quit my freelance job in Germany in order to focus more on this project.
Q: How many users do you have at the moment? Are they mostly Europeans, or is the project popular worldwide already?
A: Over 3,000 people registered an AI Sensei account so far. The number of regular and paying users is lower, of course, but from the feedback we have got, people seem to like it and think it helps them to improve. The number of subscribers is increasing steadily and if we can keep it growing, it’s enough to be a sustainable business. Most people are from Europe and the United States, but we have users from all over the world. We are also planning to start a Japanese version soon.
Q: Did AI Sensei help improve your own Go?
A: I certainly think so. I didn’t invest a lot of time into studying Go for the last few years, but I did use AI Sensei to briefly analyse every single tournament game I played, focusing on my biggest mistakes. Last year, I managed to become German champion for the first time ever!
Q: How does AI Sensei compare to similar tools?
A: Our main emphasis is about reviewing your games efficiently, even if you don’t have a lot of time. Our user interface makes it easy to identify the biggest mistakes and learn from them. In contrast to e.g. Lizzie, we currently don’t offer a ‘live mode’ to interact with the AI. I would argue that while a live mode might help in some situations, it also brings the danger of getting lost in variation after variation without really understanding the gist of a game. So, at least for myself, less might be more in this case. But we might still add this feature later if our users want that. Other differences are good mobile support, so that you can review your games even when you’re at a tournament, as well as being able to send reviews to your opponents or teachers.
Q: What features are you planning for the future?
A: We recently added a quiz mode, where the mistakes from your game get turned into Go problems to solve. It’s already fun to use, but we have a lot of ideas to improve it, like building your personal mistake database to repeatedly practice with. The goal should be to never make the same mistake again. Other ideas are playing against bots, the live mode mentioned above, as well as social features, where you can write comments and discuss reviews with other users.
We try to let our users guide our decisions on what to build next.
Report by Artem