I may be reinventing a known thing in child development or psychology here, but bear with me.

The simplest games I see babies play — games simple enough that cats and dogs can play them too — are what I’d call “circle games.”

Think of the game of “fetch”.  I throw the ball, Rover runs and brings it back, and then we repeat, ad infinitum.  (Or, the baby version: baby throws the item out of the stroller, I pick it up, and then we repeat.)

Or, “peek-a-boo.” I hide, I re-emerge, baby laughs, repeat.

My son is also fond of “open the door, close the door, repeat”, or “open the drawer, close the drawer, repeat”, which are solo circle games, and “together/apart”, where he pushes my hands together and apart and repeats, and of course being picked up and put down repeatedly.

A lot of toys are effectively solo circle games in physical form.  The jack-in-the-box: “push a button, out pops something! close the box, start again.” Fidget toys with buttons and switches to flip: “push the button, get a satisfying click, repeat.”

It’s obvious, observing a small child, that the purpose of these “games” is learning.  And, in particular, learning cause and effect.  What do you learn by opening and closing a door? Why, how to open and close doors; or, phrased a different way, “when I pull the door this way, it opens; when I push it that way, it closes.” Playing fetch or catch teaches you about how objects move when dropped or thrown.  Playing with button-pushing or latch-turning toys teaches you how to handle the buttons, keys, switches, and handles that are ubiquitous in our built environment.

But what about peek-a-boo? What are you “learning” from that? (It’s a myth that babies enjoy it because they don’t have object permanence; babies get object permanence by 3 months, but enjoy peek-a-boo long after that.) My guess is that peek-a-boo trains something like “when I make eye contact I get smiles and positive attention” or “grownups go away and then come back and are happy to see me.” It’s social learning.

It’s important for children to learn, generally, “when I act, the people around me react.” This gives them social efficacy (“I can achieve goals through interaction with other people”), access to social incentives (“people respond positively when I do this, and negativey when I do that”), and a sense of social significance (“people care enough about me to respond to my actions.”) Attachment psychology argues that when babies and toddlers don’t have any adults around who respond to their behavior, their social development goes awry — neglected children can be extremely fearful, aggressive, or checked-out, missing basic abilities in interacting positively with others.

It’s clear just from observation that the social game of interaction — “I make a sound, you make a sound back” — is learned before verbal speech.  Preverbal babies can even execute quite sophisticated interaction patterns, like making the tonal pattern of a question followed by an answering statement.  This too is a circle game.

The baby’s fascination with circle games completely belies the popular notion that drill is an intrinsically unpleasant way to learn. Repetition isn’t boring to babies who are in the process of mastering a skill. They beg for repetition.

My personal speculation is that the “craving for drill”, especially in motor learning, is a basal ganglia thing; witness how abnormalities in the ganglia are associated with disorders like OCD and Tourette’s, which involve compulsive repetition of motor activities; or how some dopaminergic drugs given to Parkinsonian patients cause compulsions to do motor activities like lining up small objects or hand-crafts. Introspectively, a “gear can engage” if I get sufficiently fascinated with something and I’ll crave repetition — e.g. craving to listen to a song on repeat until I’ve memorized it, craving to get the hang of a particular tricky measure on the piano — but there’s no guarantee that the gear will engage just because I observe that it would be a good idea to master a particular skill.

I also think that some kinds of social interaction among adults are effectively circle games.

Argument or fighting, in its simplest form, is a circle game: “I say Yes, you say No, repeat!” Of course, sophisticated arguments go beyond this; each player’s “turn” should contribute new information to a logical structure. But many arguments in practice are not much more sophisticated than “Yes, No, repeat (with variations).”  And even intellectually rigorous and civil arguments usually share the basic turn-taking adversarial structure.

Now, if the purpose of circle games is to learn a cause-and-effect relationship, what are we learning from adversarial games?

Keep in mind that adversarial play — “you try to do a thing, I try to stop you” — kicks in very early and (I think) cross-culturally. It certainly exists across species; puppies do it.

Almost tautologically, adversarial play teaches resistance.  When you push on others, others push back; when others push on you, you push back.

War, in the sense we know it today, may not be a human universal, and certainly isn’t a mammalian universal; but resistance seems to be an inherent feature of social interaction between any beings whose interests are imperfectly aligned.

A lot of social behaviors generally considered maladaptive look like adversarial circle games. Getting sucked into repetitive arguments? That’s a circle game. Falling into romantic patterns like “you want to get closer to me, I pull away, repeat”? Circle game.  Being shocking or reckless to get attention? Circle game.

The frame where circle games are for learning suggests that people do these things because they feel like they need more practice learning the lesson.  Maybe people who are very combative feel, on some level, that they need to “get the hang of” pushing back against social resistance, or conversely, learning how not to do things that people will react badly to.  It’s unsatisfying to feel like a ghost, moving through the world but not getting any feedback one way or another. Maybe when people crave interaction, they’re literally craving training data.

If you always do A, and always get response B, and you keep wanting to repeat that game, for much longer than is “normal”, then a couple things might be happening:

  • Your “learning algorithm” has an unusually slow “learning rate” such that you just don’t update very efficiently on what ought to be ample data (in general or in this specific context).
  • You place a very high importance on the A-B relationship such that you have an unusually high need to be sure of it.  (e.g. your algorithm has a very high threshold for convergence.) So even though you learn as well as anybody else, you want to keep learning for longer.
  • You have a very strong “prior” that A does not cause B, which it takes a lot of data to “disprove.”
  • You have something like “too low a rate of stochasticity.”  What you actually need is variation — you need to see that A’ causes B’ — but you’re stuck in a local rut where you can’t explore the space properly so you just keep swinging back and forth in that rut. But your algorithm keeps returning “not mastered yet”. (You can get these effects in algorithms as simple as Newton’s Method.)
  • You’re not actually trying to learn “A causes B.” You’re trying to learn “C causes D.” But A correlates weakly with C, and B correlates weakly with D, and you don’t know how to specifically do C, so you just do A a lot and get intermittent reinforcement.

These seem like more general explanations of how to break down when repetition will seem “boring” vs. “fascinating” to different people or in different contexts.

76

15 comments, sorted by Highlighting new comments since Today at 11:28 AM
New Comment

About two weeks ago a toddler toddled up to me. I thought about how to interact with them, knowing that I sometimes can get on well with kids but often they just wander away. Then this post flashed through my mind. So:

  • I offered the child a sharpie.
  • He took it.
  • I held out my hand for it back.
  • He returned it.

We did this 3 more times. At which point the child took me by the hand and showed me other things he'd found, and a great time was had by all.

Thanks for the post!

The baby’s fascination with circle games completely belies the popular notion that drill is an intrinsically unpleasant way to learn. Repetition isn’t boring to babies who are in the process of mastering a skill. They beg for repetition.

I disagree with the implication there that drill is repetition. Drill, to me, is repetition with predictable results. If I'm doing the same thing over and over again, and I'm getting exactly what I expect each time, that's a drill. The sort of entertaining repetition you're pointing at here, is something where I don't necessarily know what to expect every time I take an action.

A good contrast is painting a deck versus playing a slot machine. They're both extremely repetitive actions. Heck, even the physical movements in each are similar (if anything, deck painting involves less repetitive movement than playing a slot machine). Yet, we see people getting addicted to playing slot machines. I've never heard of anyone getting addicted to deck painting. The difference is that deck painting is pretty predictable. Dip paint in paintbrush, apply paint to deck, and there's paint on the deck, exactly as you'd expect. A slot machine, on the other hand, is geared toward unpredictability. You pull the lever, and you don't know what's going to happen when the reels stop. Will you get the jackpot? A lesser prize? Nothing at all? The sorts of circle games that babies enjoy are closer (from the perspective of the baby) to a slot machine than to deck painting.

For example, let's look at the Jack in the Box. It's predictable and boring to an adult. An adult (or even an older child) will pretty quickly catch on on the pattern that the box pops open after a number of turns or on a particular musical note ("Pop goes the weasel," etc.). However, to a child, especially to a child that's still grappling with the concept of cause and effect, a Jack in the Box is endlessly fascinating. Here's a mechanism, and when I manipulate the mechanism, something seemingly entirely unrelated happens?! How? Why?

Peek-a-boo is similar to that as well. Yes, the child might know that you're still there. But I'm willing to bet that you don't make exactly the same expression when you open your hands and reveal your face each and every time. It's the variety of facial expressions, and the effort required to predict them that provides the unpredictability that transforms peek-a-boo from a drill into a game.

The concept of a core game loop seems to point to similar phenomenon when explictly setting up games.

Yes, that was my thought precisely. Which of those explanations might be useful in interpreting the drives of, for instance, people addicted to World of Warcraft?

One interesting thing with WoW, or at least pre-cata WoW, is that the longer you play the higher amount of social interaction is needed. Up to level 15 you play almost completely alone. Then you start being pushed into meeting some randos; maybe having joined one of their guilds by 30. By 50 you're playing mostly with the same people, and by 60 you look forward to scheduled interactions with those people one to five times a week.

I haven't played the new versions, but my impression as to why the recent versions are less addictive:

1) More time with randos, less with friends. No ramp up into the parts that require social ties so it's harder to get over that hump.

2) More randomization of rewards leads to less strong feedback loop of "work hard -> get stronger"

3) Homogenization of rewards and fewer social ties lead to less strong feedback loop of "get stronger -> get prestige"

I view the social connections and the ability for anyone to feel prestigious with relatively simple work as a core part of the addictiveness.


Up to level 15 you play almost completely alone. Then you start being pushed into meeting some randos; maybe having joined one of their guilds by 30. By 50 you’re playing mostly with the same people, and by 60 you look forward to scheduled interactions with those people one to five times a week.

As I recall, at level 60, after you finish the 5-person content, you were forced into 40-person raids, where the amount of specialization/coordination/order-following required to make progress made it more like a tedious job than a game, at least for me. Curious if anyone has any insights into the design choice there, e.g., what was the thinking behind the end-game being 40-person raids, why wasn't there more of a ramp-up between the near-end-game and the actual end-game, did most WoW players not find it so tedious, etc.?

In theory it seems like massively multiplayer games would be a good way for people to develop/practice social/coordination skills, and I think WoW and MUDs before it did help me a lot in that regard. (Before, I was really anxious of talking to people.) But I'm not aware of any games that go beyond trying to coordinate 40-person raids, scaling into hundreds or thousands or more. (And as I mentioned, even the 40-person content was tedious to me.) I wonder if there is any way to make larger scale coordination fun.

You may be interested in my posts about WoW.

As I recall, at level 60, after you finish the 5-person content, you were forced into 40-person raids, where the amount of specialization/coordination/order-following required to make progress made it more like a tedious job than a game, at least for me. Curious if anyone has any insights into the design choice there, e.g., what was the thinking behind the end-game being 40-person raids, why wasn’t there more of a ramp-up between the near-end-game and the actual end-game, did most WoW players not find it so tedious, etc.?

This was (mostly) true at the beginning of the “Vanilla” period of WoW[1], but it soon changed. The Zul’Gurub[2] and Ruins of Ahn’Qiraj[3] raid dungeons were 20-man raids, designed specifically to be a bridge between 5-man and 40-man content.

Do not forget also the existence of Upper Blackrock Spire—originally, a 15-man raid dungeon (later changed[4] to 10-man).

(And, of course, with the release of Burning Crusade, raid sizes transitioned from 20 / 40 to 10 / 25.)

In theory it seems like massively multiplayer games would be a good way for people to develop/practice social/coordination skills, and I think WoW and MUDs before it did help me a lot in that regard. (Before, I was really anxious of talking to people.)

My experience was similar. Leading raids, in particular, was excellent social-skills training.

But I’m not aware of any games that go beyond trying to coordinate 40-person raids, scaling into hundreds or thousands or more.

EVE Online does this, as I understand it.


  1. From the game’s release in November of 2004 to the release of the Burning Crusade expansion in January of 2007. ↩︎

  2. Released in September, 2005. ↩︎

  3. Released in January, 2006. ↩︎

  4. In March, 2006. ↩︎

You may be interested in my posts about WoW.

Thanks, they're interesting although the title "Everything I ever needed to know, I learned from World of Warcraft" promised a bit more than you've delivered so far. :) I'd be interested in other lessons you learned, especially ones that are more transferable to other situations (the Goodhart one was better in that regard than the loot system one).

My experience was similar. Leading raids, in particular, was excellent social-skills training.

Yeah, I imagine that must be the case for the guild/raid leaders, but don't see what the footsoldiers get out of it. (Aside from practicing to be footsoldiers, which most people don't really need more of?) I guess I'm hoping that MMGs can somehow deliver more learning opportunities for social/coordination skills than just giving a small number of people the chance to practice being low to mid-level managers.

Thanks, they’re interesting although the title “Everything I ever needed to know, I learned from World of Warcraft” promised a bit more than you’ve delivered so far. :) I’d be interested in other lessons you learned, especially ones that are more transferable to other situations (the Goodhart one was better in that regard than the loot system one).

Thanks! I may write more of these at some point, yes. (I’ve got a couple of old LW comments on the subject that I should probably turn into a post or two.)

(By the way, I do hope that you read the comments on these posts—this one in particular is almost a post’s worth by itself!)

My experience was similar. Leading raids, in particular, was excellent social-skills training.

Yeah, I imagine that must be the case for the guild/raid leaders, but don’t see what the footsoldiers get out of it. (Aside from practicing to be footsoldiers, which most people don’t really need more of?) I guess I’m hoping that MMGs can somehow deliver more learning opportunities for social/coordination skills than just giving a small number of people the chance to practice being low to mid-level managers.

The thing to understand here—and this is a (perhaps subtle) lesson of my post on loot systems, among other things—is that the distinction between “guild/raid leaders” and “footsoldiers” is not nearly so sharp as you imply; and the ratio of leaders to followers, not nearly so skewed.

I elaborate on this in this old comment. To what I say there, I add this:

Consider a raid requiring 40 people (such as the high-end raid dungeons in Vanilla WoW). Now consider a raid guild, or “raiding group”, consisting of some number of people in excess of 40—to account for absences, swapping out, etc.

How many leadership positions are there in this group of people? Just one? Oh, no. By no means! There are:

  1. The guild/group leader. Responsible for managing the whole shebang.

  2. The raid leader. Can it be the same person as the guild leader? Sure. Does it have to be? Not at all, and it often is not.

  3. The guild officers. A guild leader typically deputizes several others—to help manage guild affairs (recruitment, other personnel matters, guild supplies and treasury, training and advice for lower-ranked members, etc.), to lead dungeon and raid groups in the guild leader’s absence, etc.

  4. The raid officers. These, too, may all be guild officers, but not necessarily, and even if they are, the overlap between those guild officers with raid officer responsibility and those with guild management responsibility is incomplete. Raid officers assist with assembling a raid (often a complex matter), managing sub-groups within the raid, distributing loot, etc.

(“Class leads” are generally guild and/or raid officers who are responsible for having expertise on, and dealing with players of, individual character classes.)

Out of those 40 people, a full quarter could well have leadership positions of some sort. Some or even all of the rest may, at one point or another, gain leadership experience by leading dungeon groups, or running smaller raids (as I note in the above-linked comment thread).

This matches my experience with WoW.

If someone is addicted to a false world, instead of treating it like an opioid habit perhaps it is because that false world supplies something the real one does not or cannot. World of Warcraft, for example, puts people into a tribal survival and adventure setting. You can even form clans, like families, and work together for achievements. Our modern lifestyle is very far removed from our humanity because we adapted ourselves to live in an adventurous climate. The emphasis is more on personal achievement than familial coorperation, and the larger and farther reaching our systems of laws become, the less human we become. Our inherent need for conflict comes out in the popularity of competitive games. Striving against each other makes us stronger and smarter. A life without conflict is a weak life. Bonobos and chimpanzees are close enough related to mate, however their differing environments and physiology make them dissimilar enough to be considered different species. Physically, chimps are much more powerful and also more intelligent than bonobos due to conflict between and within chimp tribes, and predators that dont exist on the bonobos island. Consider that different populations of humanity have had similar splits. The Irish lived on an island and had a brutal culture of art and war that lasted for thousands of years. They fought against each other for dominance over their island and Irish men were bred to be warriors. This in contrast to the Bashada and other tribes in Ethiopia who have a culture of peaceful cooperation and a reliance on conflict resolution with violence as a last resort. These differing cultures came to be because of scarcity of resources. In the harsh desert climate, the more you can rely on your neighboring tribes the better chances you have for survival. In Ireland, resources were more plentiful and so the culture evolved to be one of dominance and submission. Families were tighter knit and there was less of a need to be at peace with neighboring clans.

Fast forward to today, where we have a culture where cooperation is enforced by the dominant powers, there is relatively little conflict with nature, then add this to our mixed genetics and it is easy to see why some people would rather spend their lives fighting against others in a video game. Some people need that conflict to survive. If they have nothing to fight against they sink like stones. Video game addiction isnt so much addiction to the game itself, but rather a need to express what our genetics need. Conflict.

This post actively helped me improve how I interacted with my children by helping motivate what otherwise seemed like boring/stupid actions, and gave me a label to put on things.

This post starts as a discussion of babies enjoying simple repetitive games and observes that for babies this is how they learn a skill. It then suggests that we should apply the same frame to understand adults who engage in seemingly maladaptive social behaviors, such as repetitive arguments, romantic drama, and being shocking to get attention. Finally, it gives several ideas of what might being happening in very abstract terms, in the language of machine learning. It fails to connect any of these abstract, machine-learning-type explanations to any of the examples of adult maladaptive behavior, or to consider ways in which human brains don't work like machine learning algorithms. Overall, the first half might work in a volume on children/parenting, the second half should be labeled as "epistemic status: speculative" and is probably not worth including.

Seconding Zvi; this helped me play better with kids (as I wrote in a comment on the post at the time).

When I think of "drill," I think of arithmetic problems. The feedback does not come from the world, but from the teacher. A large part of that is that the child does not want to learn arithmetic because of not having an application for it. An application might provide better feedback.

The second thing I think of with "drill" is a sports coach making the kids break down the game, rather than just playing. The coach makes them do free throws, which they wouldn't do on their own, but the world provides the feedback and the value of free throws is pretty clear. Erikson says that the coach is right to drill, that it is suboptimal to just play. Even professional players are famous for not practicing free throws. A child at the basketball court alone, unable to just play is unlikely to drill free throws, but is likely to drill layups, which are closer to the standard game and may be a good choice.