Cognitive Model of HoN

Months and months ago, Julian and I were stuck in yet another baddies game of HoN and he wanted the team to do one thing, but half the team disagreed. He asked in frustration why they wouldn't listen. Putting aside the typical explanation of them being stupid/noob, I thought I'd try using cognitive models of learning to explain the answer.

I don't think all HoN players are stupid, and there have been times where someone has asked me to do something and I thought it was the dumbest idea ever, but now I've learned how good it really is. I've probably played close to 800 games of  HoN/DotA where I did not use wards of sight, and yelled at anyone who even thought about suggesting I buy some. Now it's usually one of the very first items I buy, and counter-warding is one of my favourite parts of the game.

Rewind to another tangent, I remember a time when AG was explaining to me how chess algorithms work. I don't know if this is still true, as I've never looked into writing chess AI, but at the time, it seemed like the most logical way to program a chess computer. The computer will look at the legal moves available (using heuristics to trim down the list (e.g. not playing h3 on the first move, as this is typically considered wasteful (although AG also told me there are these weird chess movements that come and go where all of a sudden, people seem to like to play moves like that))). Then it will use depth-first search to try to find the best possible move out of all the legal moves, "switching sides" and calculating the same for the opponent. I think I read somewhere that it usually did about 7 moves ahead. It determines the "best" move by adding up the values of the pieces it has, adding a few points for certain things like square control, active pieces, etc, and comparing that to the calculated value for the opponent at that point in time. Then it works out the best trade-off. If there are two equally good choices, it chooses at random.

It "learns" but playing lots of games, and when it wins, it "strengthens" the path for that series of moves, and when it loses, it "weakens" the path for that series of moves. The next time it comes to a crossroads, it will end up choose the stronger path.

Considering a computer has insane processing power, it can handle quite a lot of calculations. Plus, computers don't necessarily "forget" to take things into account. Humans, on the other hand, are quite fallible. I wanted to explain all of this because I wanted to contrast that to how I think humans develop "HoN-sense" (for some reason, I read that as rhyming with "nonsense").

When something good happens, you tend to want to replicate things so that it might happen again, and when something bad happens, you tend to avoid similar situations in case it does happen again. This is how things like superstitions develop. What I've typically found with HoN players (and which I am also guilty of) is that when they win, it's because they did well, and when they lose, it's because their team did poorly. So if you combine the two, you get players who remember the awesome things they did that caused the win, and banlist the baddies who made them lose.

Imagine that your HoN experience was quite limited, and of the few times you did the above strategy, you ended up winning. You play a few more games, and your cognitive map starts to look like this:

You've played and won with Pyromancer the most, so that is the strongest path at the moment. But you've experimented with a few other heroes and strategies, some going well, others not so well. I should point out that unlike a computer, we are unlikely to have calculated every possible path. So in this case, the lines represent past experiences. Whether the lines lead to win or lose indicates the outcome. The "strength" of the line in this case doesn't indicate the best path, but number of times that path has led to that outcome.

A bit more time, and you might start to see a pattern ("Every time I play Nymphora, go top, and buy boots and wards, I lose"). The only problem with HoN as opposed to a game like chess, is that there are so many other factors involved that it is quite difficult to consider every single one of them. Team balance, good lanes, runes, hero match-up, team skill, and neutral creep spawns are probably only a few. But because we tend not to overanalyse the situation (especially because this is supposed to be fun), you end up thinking Nymphora + top lane + boots/wards = LOSS. This is despite the fact that the map might look a bit more like this:

Every person goes through a different developmental journey, and everyone will come up with different cognitive maps with different lines. So when two people with opposing ideas come together, it's difficult to convince one person to try something they "know" to be a losing strategy. As far as they are concerned, it has never worked in the past, and is unlikely to work now. Even if it is a weak line leading to a losing path, they may feel adverse to that strategy as they believe there are other winning strategies available.

Which is why I think it's great to play with open-minded people who have never played before. They have no existing map, and so they end up learning your map, and you work really well together. ^_^

I think I'm skill-capped at the moment, and I don't think I will get better until I spend more time playing with better players, as I have the cognitive map of a 1500 player. It's something that I find difficult to do, but when I play with stronger opponents, I should spend more time trying to take on their maps, instead of feeling sad that I got completely crushed.