Using some lessons from two behavioral science books, this article discusses several ways we can improve our decision making in StarCraft.
To play StarCraft well, we must review our games and forecast opponents’ moves. Forecasting is such a core part of the game that academics are using it as a platform to understand forecasting. Despite our best effort, biases are inevitable in our reviewing and forecasting processes. Thinking in Bets by Annie Duke and Superforecasting by Philip E. Tetlock and Dan Gardner are two books that could guide us to do better.
Thinking in Bets applies lessons from poker to decision making in real life, while Superforecasting offers insight in making predictions systematically. The common theme of the two books is making good decisions under uncertainty, which is an important skill in life and StarCraft.
More poker, less chess
The first chapter of Thinking in Bets is titled “Life is Poker, not Chess”. Unlike chess, poker is a game of imperfect information, whereby decisions are made without all the information available to the players. Similarly, StarCraft players have to deduce what the opponent is doing without 100% certainty. Imperfect information is a defining feature of StarCraft; so much so that it is one of the main reasons why DeepMind picked StarCraft for artificial intelligence research.
Interestingly, people tend to associate StarCraft with chess more than poker. This could be due to its strategic and no-luck meritocracy features. While StarCraft shares some commonalities with both board games, I believe there are merits in thinking from a poker perspective.
Decision driven, not result driven
When reviewing games, how do we evaluate the quality of the decisions made? We usually adopt a cause and effect thinking and examine whether the decisions led to positive or negative outcomes. Let’s use the below vod as an example, listen carefully to what Artosis and Tasteless said about the decisions INnoVation made in the previous game:
They were evaluating INnoVation’s build choice in game 4 of the series. In it, INnoVation went for an unorthodox and risky one Barracks three Command Centre build. He got crushed quickly by Cure’s one base push as a result.
Tasteless: “I think it was a big mistake for him to do that.”
Artosis: “Do you? Because if there is a Reaper expand, suddenly he’s a genius. [People would say] he is blind countering that.”
INnoVation’s build choice no doubt plays a big part in his defeat in that game. Is it a bad decision to use a quick 3CC build because he lost to a one base push? Many would say it is and then justify it by saying that is why we don’t see quick 3CC in TvT. Such evaluation is result driven, assuming defeat is the consequence of bad decisions.
In poker, players frequently make good (or even best) decisions and still lose the game, as no good decision can overcome a better hand. The game essentially is about maximising your earning when you’ve a better hand and minimising your loss when you don’t. Poker players do not evaluate the quality of their decisions based on the binary outcome (i.e., winning or losing the hand). The decisions are instead evaluated based on the available information at the point of time. Imagine that you have to choose between betting on an option that wins 60% of the time versus another that wins 40% of the time. Your decision to bet on the 60% is a good one even if you lost the bet.
Artosis’ comment demonstrates this thinking style well. A quick 3CC build indeed is a good option against Reaper expand, as it takes advantage of the opponent being passive. Our evaluation of INnoVation’s build choice should be based on the likelihood of the opponent using a Reaper expand build and not the fact that he lost. Similarly, when we review our own games, we should avoid the pitfall of being solely result driven.
Poker also teaches us to update our beliefs. Whenever a new card is revealed on the table, we have to update the probabilities of various possible outcomes. Updating beliefs is an overarching theme in Superforecasting too. The authors often draw reference from Bayesian inference which is an important technique in statistics. It derives from Bayes’ theorem to update the probability for a hypothesis as more evidence or information becomes available. Whether we are aware or not, we form hypotheses constantly in StarCraft. Is the opponent going for Dark Templar? Are there proxy Reapers? Is that a Roach all-in?
The issue we face is not whether we made the right guesses, rather it is the thought process of making instinctive conclusions with high certainty. For instance, in TvT, if a player sees no Barracks in the opponent’s main base and the geysers were taken, the instant thought is proxy Reapers. A better way to utilise the information is to update the likelihood of proxy Reapers and not believe whether the opponent is doing proxy Reapers. Updating likelihood prevents us from thinking in absolute and in turn reduces tunnel vision. Using the same proxy Reapers example, the initial attack should occur at around 2:15. If no Reaper attacks at the expected timing, one should update the likelihood of proxy Reapers with this new piece of information. Then, logically, the likelihood of other plausible builds increases (e.g., proxy 1-1-1, proxy Marauders, expansion etc.), so a good follow up move is to send some units to scout for the plausible proxy buildings to further update the probabilities. In contrast, if an either-or approach is used, the player may simply conclude that there is no proxy Reapers when no Reaper attacked by 2:15. This ignores the possibility that the opponent simply wants to gather more Reapers and attack at an unexpected timing. In sum, our predictions are more accurate when they are updated incrementally than when they are perceived as right or wrong guesses.
Here is a quote from Superforecasting about updating our beliefs:
The best forecasters tend to be incremental belief updaters, often moving from probabilities of, say, 0.4 to 0.35 or from 0.6 to 0.65, distinctions too subtle to capture with vague verbiage, like “might” or “maybe,” but distinctions that, in the long run, define the difference between good and great forecasters.
Illusion of knowledge
The above quote implies that the better we are the more precise we can trim down on the probabilities. While experts should make more accurate predictions, the authors of Superforecasting are quick to warn us of the illusion of knowledge in chapter 2. Here is another quote:
“Often, I cannot explain a certain move, only know that it feels right, and it seems that my intuition is right more often than not,” observed the Norwegian prodigy Magnus Carlsen, the world chess champion and the highest-ranked player in history. “If I study a position for an hour then I am usually going in loops and I’m probably not going to come up with something useful. I usually know what I am going to do after 10 seconds; the rest is double-checking.” Carlsen respects his intuition, as well he should, but he also does a lot of “double-checking” because he knows that sometimes intuition can let him down and conscious thoughts can improve his judgment.
Our intuition improves with our expertise. However, intuition can be a double-edged sword when the experts trust it without doubt. The best StarCraft players are no exception. In HomeStory Cup XVIII, TaeJa used a very greedy build against Serral in game 2. TaeJa put down the third Command Centre before the Factory, and he went straight to bio and upgrade without Hellions (discussed here). In the post match interview, Serral admitted that he had mistaken the build as 2-1-1, because he saw no Hellions and a high Marine count at the start.
On the flip side of the coin, top players are also taking advantage of opponents’ intuition. In the vod below, INnoVation used a Battlecruiser build against Dark in their most recent GSL games. Battlecruiser builds usually transition into mech, so a smart reader like you may have already figured out that INnoVation went for bio instead.
If you indeed think that INnoVation went for bio to take advantage of Dark’s expert intuition, you yourself may have fallen prey to your intuition. INnoVation did not just go for a “normal” bio transition, as he opted for a two base all-in with bio. Our intuition misleads us to think that Battlecruisers imply long macro games, and many of us stop updating further after deciphering whether it is mech or bio.
That is not to say we should not trust our intuition. Instead I believe we should use our intuition (especially if you’re an expert) as the starting point, and then we apply the Bayesian principles to update our beliefs. Annie Duke also shared the same sentiment of having a healthy dose of doubt toward experts’ (and our own) opinions in Thinking in Bets:
An expert in any field will have an advantage over a rookie. But neither the veteran nor the rookie can be sure what the next flip will look like. The veteran will just have a better guess.
This article can be summed up with three key points:
- When we review our decisions in the past, we are biased by the results of the present. The quality of decisions should be evaluated using the available information at the time when the decisions were made.
- Our beliefs are hypotheses not conclusions. They are meant to be updated to improve predictions.
- Our intuition is good, but it is only as good as we are willing to doubt it.
After all, how we think matters more than what we think.
I will be discussing this topic with Aoret and Adroseth on The Archives Book Club podcast (Buzzsprout, Spotify, YouTube). The podcast discusses books and StarCraft. It will be live on Aoret’s Twitch channel at 7.30pm (PDT) on 8 August 2020.
If you enjoyed this article, I’d love you to share it with one friend. You can follow me on Twitter and Facebook. If you really like my work, you can help to sustain the site by contributing via PayPal and Patreon. You can also support me and enjoy quality tea with a 15% discount at AFKTea by using the “TERRAN” code. See you in the next article!