In the last two posts, we’ve seen how experts chunk complex behaviours into single actions, and how conditioning processes act over thousands of iterations to produce expert behaviour. The long road to mastery (10,000 hours, and hundreds of errors) is one worth walking, and when you get to the end of it, it often manifests in the form of a “gut instinct” about certain game states. The goal is a certain level of automaticity with regard to the best plays in a given situation.
But as that automaticity begins to set in, there is a dangerous slump in your learning curve.
When a person (or indeed, an AI) learns a complex task in which there are many rules, but some exceptions, there is a predictable upsurge in errors at the middle of the curve. The best example of this in psychology comes from children learning the rules for conjugating verbs. (I don’t mean via formal education here, but rather their absorbing the rules unconsciously from their language interactions with adults around them.)
Most verbs in English follow strict rules with regard to how you conjugate them. For a regular verb, when you are speaking about the past you add “-ed” to the root (e.g. “I jumped”, “I climbed”). However, a number of very common verbs are irregular (“I ran”). When children first learn to speak, they learn each verb as an individual set of rules, independant from one another. They make increasingly few errors with irregular verbs. But at a certain point in the curve, they begin to internalize the regular rules – they “chunk” the rules for conjugation and start applying them to everything, and errors such as “I runned” become commonplace. This is called overgeneralization. After a while longer, with sufficient feedback from their environment, these errors reduce again and the child reaches true fluency with verbs.
(Generalization is a well established idea in Operant Conditioning research, and is core to modern behavioural explanations of language and cognition. It’s also probably the source of a lot of our cognitive biases.)
I think overgeneralization is a big pitfall that you experience in learning to master Warmachine. As far as my own play goes, I’m pretty convinced that I’m sitting right on that point of the learning curve. I was really disappointed in my performance at the recent Irish Masters, and in analysing the games afterwards I realised that most of my losses could be put down to going on “autopilot” and not reacting properly to an unusual game state or particularly canny play by my opponent. I’d trained enough with my lists that I had mastered the gameplan for it, but hadn’t yet gotten over the overgeneralization stage and so kept calm and blundered on when I should have realized I was in an irregular situation.
The autopilot phase is inevitable in learning a game as nuanced as Warmachine. You have to learn the basics and get them down hard before you can improve your game. You need to know your lists inside and out, to intuitively know its standard plays so you can make it work under time pressure. To be a master, much of your performance must be automatic. Watch great players at the table, and you’ll see it in action as they position their units or eyeball distances. You’ll see it in their ability to “know” that an assassination is on without having explicitly done the maths (Aside: Doing that maths every time will pay off as it becomes more automatic). The whole value of expert performance is that it involves making complex plays automatic and instinctive.
The pitfall comes from getting stuck in the overgeneralization phase of learning, and being unable to break out of your standard box when you’re playing at top tables against players who know how to throw off your standard game. Even experts fall prey to this – my favourite experimental example of this comes from Professional Chefs. Subjects were asked to prepare and serve a meal, but when the time came to serve the meal they were told that there were no plates or bowls in the kitchen. The professional chefs where unable to figure out how to serve the meal, often channeling their inner Gordon Ramsey in their despair. The amateur chefs simply served the meal using an assortment of small frying pans and pots. The professional chefs had internalized the purpose of each of their cooking tools so completely that they couldn’t see other possible uses for them.
I’m right there in that slump myself, so I’m very motivated to work my way out of it. As noted in the previous articles, Deliberate Practice is your way forward. What you need to practice to escape overgeneralization is recognizing irregular situations and turning off the autopilot. Careful, mindful attention to what mistakes cost you the game will help you in learning that skill. One thing we’ve decided to start doing in our LGS (I say we, but mostly me and VagrantPoet, unless we can draft some others in) is to maintain a “punt jar”. Every game we lose to a foolish error means putting a Euro in the jar, thus establishing punishing consequences and forcing us to think about it.
No doubt by the time we get to the WTC, the jar will be overflowing. We’ll use it to buy you a pint if you see us there!
P.S Next week, we will be looking at something completely different, as I nerd out spectacularly about how the real world processes that might underlie the functioning of warjack cortices in the RPG.