Hacker News Reader

Comment by kmlx | Hacker News Reader

i first had this idea more than a decade ago: a way to translate gaming tasks into something productive on the other end of the pipe. the gamer doesn't realise, and the company actually solves issues by just letting the gamer play the game.

in the non-gaming world, i think the closest mass market idea has been google's captcha which actually trains an AI in the background. that's very naive and very simple, and definitely not a game, but imagine something similar with puzzles (i.e. Witness, the game, is all about puzzles. why not replace those puzzles with puzzles from real life that actually need solving).

View on HN

EVE Online has done several citizen science minigames over the years where human players can perform image-classification tasks.

2017: https://arstechnica.com/gaming/2017/05/eve-online-citizen-sc...

2018: https://arstechnica.com/science/2018/08/minigame-integrated-...

They had a more recent one for covid research.

https://www.wired.com/story/eve-online-project-discovery-cov...

If you haven't read Ender's Game, I highly recommend it based on this topic.

I think the tricky part is - if it's an unsolved problem, how does the game know when a task is complete? Feedback loops are a core element of game design, and part of the reason tedious tasks can be satisfying, but unless it's directly manual labor, I'm unsure how to create a proper feedback loop

I thought the same. Ender's Game and the cell classification task have something in common: the end status to achieve (whether to eliminate all opponents or identify something that's known). If we think this way, these two tasks could still be treated as machine learnable tasks (albeit poorer solution possibly). A genetic algorithm for example, could be trained to derive the means to achieve the goals. In cases where we don't know the end goal of the game, how do we train ourselves towards an unknown objective or without guidance of known answers.

Hmm.

Perhaps we could use bored humans as an accelerator to heuristically solve NP-complete problems. Then we'd just need to do the verification step.

All we need to do is prove that this produces a big-O improvement, and I think we could make a compelling business case.

Neil Stephenson’s REAMDE touches on this idea, and reaches the somewhat inevitable conclusion that if you can automate enough about a real world task that you can translate it into gaming terms, and determine automatically whether or not the player has succeeded at the task (in order to reward them ‘in game’ for completing the task), then you have probably done all the essential work needed to have a computer do the task in the first place.

Why require automated validation?

In the real world, most tasks output as inputs to other tasks by other people. And those downstream people perform validation of incoming work.

Delay reward until downstream validation passes.

I think everyone and their dog had this idea the first time they learned of "grinding" in computer games, it's just that somehow, nobody's found out a way to turn the game-type grinding-input into real-world-type-grinding-output.

I do consider the protein folding game an interesting exploration into the realm, but since there are still more minecraft-carrots harvested than proteins folded, I'd say it is not an example of how to do it, but rather a datapoint in the right direction.

I think the work they do in the series Severance was like this. They were in some gamy looking interface and had no idea what they were actually doing but got good at doing it and apparently it was important

Reply on news.ycombinator.com