What games would a robot play? It's a question that struck me last night, and kept me up as I sketched out diagrams for android puzzle toys.
For a start, we have to imagine why a robot would play games. We're not even able to explain why we enjoy them so much. To say that games trigger the reward system in the brain doesn't add anything by way of an explanation. Enjoyment, pleasure, is an atomic building block of behaviour, it's impossible to break it down any further. So why ask why robots would play games when we can't explain our own fascination with them? They just do. We can give reasons why they should play games - why it might be useful to build robots that play games. Games can be educational. A good game is a script that changes with each iteration, so a game design is a convenient way of packaging a lot of educational material into a compact space. Which is the other important part of games to robots. By and large, humans do well in unfamiliar environments. Robots don't. Inside controlled spaces - the assembly line, the laboratory, they outstrip humans. But even fearsome intergalactic megalomaniacs like the Daleks couldn't conquer a staircase. Games could provide a balance between a controlled environment and a novel one, giving robots a safe space to develop - just as they do with humans.
What might a game for robots look like? That's more interesting. We get to think about how a robot experiences challenges in the real world, and also the things we find challenging that they don't. Robots have already proved that they can out-think us, at least at computationally intensive tasks. Something like a Rubik's cube doesn't pose much of a challenge. With enough computing power, every permutation can be worked out in advance, while us humans get by on strategies to reduce the bulk of the possible combinations, then blunder through until we can push the cube into endgame. One method to customise games for robots would be to make them so computationally intensive that they can't be worked out in advance - chess fails this test, although electronic brains have yet to consistently best a human at the Chinese board game Go. The second, related method is to ensure there's a degree of randomness to the game which obfuscates probable outcomes, limiting how far ahead a robot can plan.
As robots also sense the world differently, perhaps then the way to limit their insight into game narratives is to explore other senses. Piezoelectic skin might allow a robot to feel slight surface currents. If so, perhaps a shifting tile game could be manufactured that contained tiny contacts in the edges of each tile, with the unseen circuitry creating a complex variation in the surface current of each tile as the pieces are shifted around the board. In this way you would create a puzzle that had to be physically touched to solve.
Repainting the six faces of a Rubik's cube so that they resembled QR codes could exploit the networked capacity of a robot. Each iteration of the cube face could be reconifgured with new meaning for each new game, allowing endless variety. QR codes also have the advatange that they can code in far more data than colours, in such a way the faces could form fragments of a narrative that a robot had to order. QR codes effectively allow more than 6 complete faces to be embedded into a single cube, and as such allows games with conflicting priorities and multiple best-fit outcomes from which a robot can choose according to its philosophy or aesthetic principles.
We don't think of robots having leisure time, perhaps because by they are workhorses by definition. Robota - from the Czech word for servitude - are tools to be used by humans. That's why I think this is such an important question to ask. We can create robots in our own image - program them to critique art or discuss politics, but this lack of obvious labour conceals the reality that such robots still have their behaviour dictated by us. Asking what a robot wants to do in its own time forces us to consider a world in which a robot is no longer defined by its duties to man.
For that reason alone, I think it's an idea worth exploring.
- Log in to post comments
More like this
the robot wouldn't do it. it would understand that the game was pointless and caused unnecessary wear on parts. also, it would STILL technically be performing a task for human amusement. so the premise of this article is only interesting in a science-fictional/hypothetical/geek-with-nothing-better-to-do kinda way.
I wonder what a simple AI with a rudimentary drive to explore an environment and a circuit that provides a small reward for making connections might do with a game like Myst.
The "kill all humans" game.
Seriously though, the answer is right in front of you - a human is simply a sophisticated organic automata that has been programed by evolution to play games such as tag and hide-n-seek. A sophisticated mechanical automata would therefore play whatever games it was programmed to by it's "evolution".
@ AlanYes but there are crucial differences, both in the way that robots sense the world, and in that game-playing isn't an abstract product of evolution, but produced by culture.
Just as games have to be designed for humans, they also be designed for other intelligences (think about how your toys differ to your dog's toys).
The real point of the article though, is to illustrate that just as games need to be designed for robots, robots need to be designed for game playing.
@Frank, I agree many games are cultural but some are definitrely inate. Tag and hide-n-seek would seem to me inate since they exist in young children across all cultures. Many immature predatory animals also play similar games and it's kinda obvious why they do so.
Also human babies the world over get a kick out of peek-a-boo before they can speak or even crawl, why this is so is less obvious but I very much doubt it's cultural.
Basically the article is describing an AI with a mind of it's own. Speaking as a qualified computer scientist I very much doubt you could design an AI sophisticated enough to *want* to play games in it's lesuire time. Such an AI would almost certainly be programmed by evolutionary algrothimis and a combination of virtual and real environments. In other words it would not be programmed for anything specific it would be programmed to learn and may even be taught by another robot via game play. The robot's mind that generates the *want* for game play would emerge from the complexity of it's inetrnal state machine, much the same way as organic minds emerge from the squishy state machines inside of skulls.
BTW, my dog's toys are pretty similar to a young childs toys, balls and plastic things that squeek, both are designed to be somewhat inedible but that doesn't seem to stop dogs or kids from trying to ingest them.
Training, calibration, and self-test/diagnostic procedures for robots (and computational agents) can be viewed as "playing a game". We tend not to think of them that way since we know (designed) the purpose and normally optimize them down as far as possible.
The games we play are generally designed to make us better at something we're learning, like hand-eye coordination, memory, or a physical skill.
Robots cannot improve their physical skills through practice the way we can, but they could potentially improve their programming if they were designed to learn. So, if they couldn't improve stuff like math, perhaps they could improve stuff like human facial recognition, object recognition, or creativity?
Robots might play games like "I spy with my little eye" or "exquisite corpse". It might be fun for them to play with words, or with philosophy.