What are the opportunities for organisations to use the latest in machine learning to supercharge their performance?Photo by ELLA DON on Unsplash
I am convinced the next big revolution in esports is the data and Artificial Intelligence (AI) arms-race. Whoever hires the talent, invests in the systems and more importantly, gets the players and coaches buy-in, will find themselves flying ahead of the curve. Then the rest play catch-up.
Why am I so convinced? Because it’s already happened in traditional sport, and they managed it with barely any data. Gaming is spoilt for it. Every game we play creates gigabytes of the stuff. It’s just about finding the talent to build it.
Before we begin, a little clarification on what is meant by AI. I don’t want to spend any significant time discussing it, it’s a well covered topic. However, to be clear: I use it very much as a blanket term. I really mean any use of computation and data to answer a question. That question could be “is there a dog in this picture?” or “who will win this game?” — it’s all AI. If you’re still unsure, I recommend having a Google before going deeper.
The question I’m addressing today is: what are the biggest opportunities to apply AI to League of Legends esports?
Now, when I say “opportunities” it’s not as simple as “what delivers the most value”, but “what delivers the most value for time taken”. For instance, it’s plausible to create AI capable of playing and beating the best teams in the world — we know this because they did it in Dota. It would also be incredibly costly in both time and resource.
We want to strike the balance of value and cost. This is harder than it sounds as both sides of that equation are difficult to estimate.
However, I’ve been working on a handful of these problems over the last few years and so can give a reasonable estimate. From here, these are what I believe the most valuable AI projects would be for LoL esports — in order of value.
The individual talent of players is still by far the biggest contributing factor to a teams performance. The gap between the best in the league and the worst is still outrageously large. This will close overtime as teams build more sophisticated systems for both identification and development. However, as it stands today being able to scout top talent is how an analyst can drive the most value.
So, how does AI get involved?
From what I’ve seen in the industry, a majority of the scouting relies on two systems: averages & the eye test. You’ll never replace the latter, once the field is narrowed you can demote data and rely on the experience of your team to decide.
To be clear, an “eye test” isn’t watching highlights or spectating a random solo queue game. It’s when you bring the players into scrims that the real work begins. Then it’s just as much ears as eyes. Communication, attitude & resilience are factors best observed by humans.
However, how do we go from 30 potential players to 3, without needing to bring each one in for try-outs?
That’s where AI can help. The simple statistics take us so far, but they ignore so much of what makes a great player. Do we care more about a player having a great KDA? What about their Gold Difference? Or maybe it’s their objective control? How do we measure a champion pool and the impact it will have? There are just too many factors for our simple human brains to observe and determine on scale.
So, we can use AI. We can show it the statistical make-up of good players and say “find me more of these”. We can even look at the lower leagues or minor regions and use methods to adjust performance. How good does an LFL mid-laner need to be to compete in the LEC? This is the future of esports scouting and the game is on to see which organisations work that out first.
This is the real hot topic. Casters spend hours talking about it, teams spend days preparing for it and everyone else spends weeks debating it.
It’s also a considerable unknown. No one really knows how good a draft is because there isn’t a system in place to test it. Teams play the draft once and it’s unlikely we’ll see those exact same 10 champions enough times that patch to answer the question.
However, so far in my journey through esports I’ve been genuinely shocked by how rudimentary teams are in their approach to draft. In fact, I spent 6 months building a model to try and predict exactly how good they are at it and even then felt I was only scratching the surface.
All I will say is that this is a great place to work on AI. Why? Because it’s quite easy to test. Across the major regions there are 1000s of games played each year. You can either predict who will the game using draft, or you can’t. If you can, then you’re onto something. This allows you to test various techniques and features, each time evaluating it on how it performs out in the real world. Again, I’ll dedicate more time to this at a later date.
Is Collector a good item? Which boots are best? They built X here? What an idiot!
This topic tends to boil blood. That means it’s worth solving.
Once again, AI is in a great spot. You have lots of games to look at, especially if you find a way of effectively using high-elo solo queue. The question is also fairly simple since there are only really going to be 2–3 reasonable options as to what to build next. That means the solution space is small. You can also somewhat test your technique, although it’s a bit more complex than drafting.
So why is it only 3rd in my list? Sounds like a winner.
Because the potential impact is not in the same league as the first two. Having good players play good Champions is still by far the most effective way of winning the game. Whether they build the 3rd best item or 1st cannot decide the result to the same degree. Eventually, as organisations begin to work out the substantial impact they can have from investing in AI, the gap between each of these opportunities closes. Until then, there’s a canyon between scouting/drafting and items.
There’s lots of decisions to be made throughout the game. Do we collapse on herald as 5? Shall we start Baron here? Is this drake worth dying for?
However, I put it last for a few reasons:
Outside of these four, there are a few things that I’ve heard AI be used for, such as:
These are great, and exciting. However, they’re relatively difficult and esports has not hit the point where these sorts of highly specific techniques are worth doing yet. There’s just too much low hanging fruit (see: top 2 of this list). I spoke to someone in the field about this, specifically about sleep tracking, his response was something along the lines of:
“These expensive monitors are great, but I’d be happy if I could just get the players to get 8 hours sleep!”
I think that summarises it well. Start with the simple stuff. Once we’ve done that, we can start looking at tracking eyeballs and optimising exercise routines.
You got to the end of the article! My name is Jack J and I’m a professional Data Scientist applying AI to competitive gaming & esports. I’m the founder of AI & Esports start-up iTero.GG and the analytics site jung.gg. You can follow me on Twitter, join the iTero Discord or drop me an e-mail at firstname.lastname@example.org. See you at the next one.
Want to get these articles emailed to you once a week?
No spam, no paid promotions, no bullshit - just one article every week.