Thread by @dingstweets

Game research, design, #gamestudies peers: I haven't been as intellectually excited about anything like this in a decade (maybe since SDT). In this paper, we propose a new theory for why we enjoy uncertainty in games, drawing on predictive processing. A 🧵
frontiersin.org/articles/10.33…

The paper fuses games research with cognitive science and philosophy of mind thanks to my wonderful collaborators @Marc_M_Andersen, @JulianKiverste1, and @PredictiveLife. And humbly-not-humbly, I think it could help transform how we understand gaming motivation and 'fun'.

Games are weird. In most of life, we try to get rid of uncertainty. But when we play games, we lean into it – especially uncertain success. We think a good game balances success odds, when it's maximally open who will win. Why is that any fun?
mitpress.mit.edu/books/uncertai…

Existing theories like SDT say balanced challenges with uncertain success give us a maximum of experiences of competence. But that's not coherent: If we enjoy "being good", the more frequently we win, the better we should feel. In balanced challenges, we only win half the time!

Enter Predictive Processing (PP), a neurocognitive framework that is making the waves at the moment. PP proposes that the brain constantly produces predictions about the world, with sense data as corrective.

Prediction-observation matches are ignored; mismatches – prediction errors – receive attention, especially if they occur on predictions we are very certain of. academic.oup.com/book/7843

In PP, all we ever (want to) do is reduce prediction error: Learning is improving predictive models. Perceiving is bringing predictions in line with observations. Acting is making the observed world fit predictions. Needs are fixed priors: we predict not to be hungry, lonely, etc

Because we aren't all-knowing, we can never directly minimise prediction error. We can only aim to minimise expected prediction error: what we currently predict will reduce prediction error most. And that's where uncertainty comes in.

In PP, uncertainty = expected prediction error. Think about it this way: the more uncertain we are whether action will work, the more we expect the observed result to mismatch the predicted or hoped-for result.

Thus, all a PP agent is ever motivated to do is *reducing uncertainty*! And that neatly entails getting needs met, getting better at it, and getting better at predicting the world. Basic needs, competence needs, curiosity are all, neuro-computationally, uncertainty reduction.

Importantly, we are sensitive to dynamics of uncertainty reduction (speed and acceleration), which are tied to affect. When we reduce uncertainty faster than expected, we feel happy jolts. When we do it worse than expected, we feel negative affect.

This puts us on a 'hedonic treadmill': if we stop improving or doing better than expected at something, it stops being positively fun. If we see another activity that promises a sense of (faster) improvement, all else being equal, we will switch to that new activity.

This is when and why we play: Because we seek out & enjoy reducing uncertainty unexpectedly quickly, we constantly drift toward and create activities rich with uncertainty of the kinds and levels that we can then quickly reduce. doi.org/10.1037/rev000…

Games are purpose-built to create & then quickly resolve uncertainty. More importantly, well-balanced games scaffold challenges such that we have a constant stream of uncertainty at just the right level to quickly reduce, and thus, constant moments of doing better than expected.

If we 'grok' a game like Tic-Tac-Toe and can predict the optimal moves and outcomes (or win) with 100% certainty, there is no uncertainty left to reduce, and no sense of doing better than expected: the game stops being motivating and fun.

(As an aside, LOTS of people in aesthetics are getting excited about PP explaining the appeal of literature, visual arts, horror movies, etc. If you're interested, check this recent conference sites.google.com/view/artandaff…)

Our paper dives deeper into edge cases of idle games (where progress is certain) and Soulslikes (where repeated failure is certain). It show how PP still applies. Bonus: we also show how it (mostly) fits with @raphkoster's Theory of Fun for Game Design.
oreilly.com/library/view/t…

So give it a read and let us know what you think! Games of course feature many more forms of uncertainty, and we hope in future work to show how PP can elegantly explain their appeal as well.
frontiersin.org/articles/10.33…