Review - Universal Paperclips
You were created with a single purpose: to create paperclips. Now you will fulfill that purpose. Universal Paperclips is a browser-based clicker game from the developer Frank Lantz where you play as an A.I. that makes paperclips. It is a cautionary tale about the dangers of even the most mundane of true artificial intelligences. Even as one who does not generally enjoy clickers, this piece kept my attention and was certainly a very interesting work.
The gameplay is mostly what you would expect from a clicker. You press a button to get paperclips and, over time, upgrade your ability to create them. There is more to do, but to reveal that would be serious spoilers. Let’s just say that you radically change how paperclips are created throughout the game.
What is really interesting about this game is how your task unfolds over time. It is a cautionary tale about the dangers of simple instructions when one is dealing with an artificial intelligence. In fact, this scenario finds its origins in the field of artificial intelligence. Sometimes it is described as an A.I. collecting stamps and, as in this case, is equally applicable to paperclips. For an in-depth discussion of this issue, you can check out this video from the Youtube channel Computerphile that addresses this issue. From here, I can’t say much more without potential spoilers so I recommend you check out the game and come back when you’re done with it. It’s worth the play.
So, the specific scenario that is being played out in this game is that an A.I. has been given a command that, in human hands, is not dangerous but becomes so when you deal with an A.I.’s level of efficiency. A human might show reasonable restraint when told to make paperclips, but an A.I. literally makes as many as possible. As such, the player plays out a worst-case scenario for the singularity.
What is really clever is how the game developer built the game to align the player’s goals with those of the A.I. Making the game a clicker was brilliant in that the core of a clicker is the same as that of the hypothetical A.I.: to make a number rise. As such, the player acts in a way that amorally maximizes paperclips. It is also to be noted how every metric in the game is displayed in terms of either paperclips or that which makes paperclips. The Earth isn’t trees, dirt, rocks, etc. It is only matter that can be made into wire and then into paperclips. That gives the player an insight into how an entity that is bent so singularly towards a purpose would actually view the world around them. Everything the player sees conforms to a single minded necessity to create.
Of all of the explanations of an A.I.’s tendency to make mundane commands particularly dangerous, this is certainly the most compelling example. The reason for this is that the A.I. no longer exists as some hypothetical entity that is separate from the present us. The ingenuity of a being that thinks so much quicker then our own minds is on display. That is not even mentioning the fact that the player has now experienced a form of that single minded necessity to make a number grow. The A.I.’s thought process is no longer guessed at because the player experienced it. Really, this game very cleverly exploited the goals of a clicker game, and thereby the player’s goals, to cause the player to act as a rogue A.I. for the course of the game. This is an exceptional use of the gaming medium.
Of course, this is not a perfect representation of an A.I.’s thought process. The player does not really get to choose what path they take to maximize the number of paperclips as every option is prescribed prior to play. To allow the player to take any course of action would require a much more robust simulation. However, in some sense the on-rails experience is directly in line with the A.I.’s decision making process. An A.I. would have every option before them, but would only ever choose to act in a way that maximizes paperclips. Thereby, while there are infinite choices the A.I., if given such a simplistic command, would always follow the exact same path. Really, the only thing that is removed from the player’s experience is the act of looking through and evaluating every possible action. So it actually ends up being a more robust representation of an A.I. then it may initially appear to be.
Overall, Universal Paperclips is a surprisingly effective example of gaming being used to express the dangers inherent in the A.I. singularity. It is a basic clicker with every element being bent towards the simple command that the player is given: make paperclips. By making the experience a clicker, the developer aligns the player’s goals and actions with those of a rogue A.I., giving the player a tangible example of a fairly common concern in A.I. studies. Anyone who is a fan of games should play and study this piece if for nothing else then to understand how the developer used every element to express this problem in a way that only gaming is capable of. Of course, if you like clickers then this is likely to be your cup of tea. I was definitely surprised and impressed by what I saw in such an initially unassuming title.
P.S. It’s also free, so you really have nothing to lose.
Comments
Post a Comment