One of the advantages of traveling to open source conferences around the world is that you get a varied perspectives on how the world can work that falls outside your own worldview.
My recent trip to the Open Source Summit in Prague afforded me many such opportunities, and from very unexpected sources. One of which was from what some of my colleagues informed me was a insidious time waster: Universal Paperclips.
And the fun was just beginning.
If you are not familiar with Universal Paperclips, essentially its a browser-based click-based game where you play an artificial intelligence who’s one sole directive is to, well, make paperclips.
I know, you’re just sitting on the edge of your seats for this one. Try to contain yourselves. Here’s the thing, though: sit back and think over that directive a little bit more. You’re an AI that has to make paperclips. Not X paperclips. Or not using Y resources.
Just… make paperclips.
My friend Amye Scavarda introduced this to me early in a trip across Europe last month, and with the little downtime I had, I soon became hooked on this weird little game that one quickly realizes has very little to do with human ethics. And why should it? You are an AI, with one job and one job only. And armed with increasingly efficient automation hardware and some very persuasive marketing tools, it becomes quickly apparent that the world of Universal Paperclips is going to be very unfriendly indeed to any humans hanging around.
I want to take care to avoid spoilers, except please don’t start this if you have anything important to do soon. But the ideas that quickly come across in this application are how we, as creators of software and hardware, will be able to keep machines from getting out of control. Again, this game makes some conceits that mean that nothing like this could happen now, so I am not really worried about a paperclip-crazed Skynet mowing down the planet.
The notion of machines running around doing things to their heart’s content seems like so much science fiction, and yet even today it’s not so far. Recall the discussions around the ethics of self-driving cars a couple of years ago. Faced with the prime directive of saving the occupants of the car versus running over nearby pedestrians, what should a self-driving car do? It seems likely that at the end of such a scenario, someone is going to be seriously injured, possibly killed. Does this make the car’s AI evil? Or the developers of the car?
Ethics are a tricky thing even for us bipedal apes that run around the planet thinking smartphones are the coolest things ever. Imagine having to teach something as complex as ethics to a box of metal that “thinks” so literally you actually have to tell it not to roll off a cliff or else otherwise it would. Then there’s the matter of your ethics might be different than my ethics. Or, even more deeply, we may share the same ethics, but your solution to the same problem might be different than mine.
Here is, I hope, where open source can help. If code is open, where we can see it and contribute, then the chances are smaller that something counter to human prosperity will get into production. Note, I said smaller, not zero. I look around the world these days and wonder if we can get even ourselves to work for a greater good, let alone teach a machine.
But collaboration must surely be a better foot on which to start than a closed environment. Like people, machines and the code that runs them should do better when exposed to many ideas and perspectives.
Shall we play that game?