MIT experiment let internet users control a human
Would you let random strangers tell you what to do over the internet? Then this game is for you!
On Halloween, MIT asked the internet to control a human being. The experiment, called BeeMe, asked people online to act together to control the actions of an actor as she navigated a fictional scenario. The goal of the experiment was to see how, or if, a "hive mind" could work together.
Here is part of their conversation.
Niccolo, before we get to how your experiment went on Halloween, can you explain how it was set up?
So BeeMe was an online game where a lot of different users from everywhere around the world connected online and through our platform, could control in real time, a human avatar, a real human being. The game was designed to see if real-time cooperation can succeed online.
OK, so how did the users control the actor playing the game?
So the actor had the phone camera streaming in real time and the streaming could be seen through our web platform. So the users from home using our platform could see what the actor was seeing and hearing. Through a text box, they could communicate their commands to guide the actor through sort of like a sci-fi mystery, and complete the mission.
Exactly. The commands [are] not, you know, a right-left mouse click. But it's actually real like, a free text.
And so you had an actor who was the person who was being controlled by people's commands? I think if I were the actor, I would have been a little bit nervous about what I would have been asked to do. Were there any restrictions on what the actor could could be asked to do?
We were expecting, yes, a lot of trolling, because it's part of the experience. You know, when when you give total freedom [online], people want to have fun and want to see weird things happening. And weird things did happen, but we were extremely clear about safety and ground rules that we wanted users to respect.
Right. So what happened on Halloween night?
We started the game and the turn up of people was incredible. We reached more than a thousand users at a given point, but we sort of quickly re-adapted to the new environment and people started cooperating in real time using the chat, and it was very interesting to see the enthusiasm of these users and how that was carried out by the story.
So were the users able to sort of work together in order to get this actor—the main character—to accomplish a goal together or was it doing you know one thing and then some completely opposite thing?
This is exactly what we were worried about at the beginning. Part of the experiment was to test whether real-time cooperation can succeed in that sort of complex sequence of tasks, that only makes sense one after the other. And surprisingly, all the crowds succeeded in completing the story. Zookd [The "Artificial General Intelligence" in the story—Ed.] was defeated, although of course there were strange commands happening in the meantime. So it seems like the crowd managed to achieve both the funny-entertaining part but also the actual mission.
I know it's only just happened, we're talking on November 1. But do you have a sense of what you've learned from this experiment?
It's very early to say, but there were some surprising things. For example, at the end, there were sort of two crowds controlling two different characters, and the two crowds had to make a decision whether to turn on each other and join Zookd the evil AI or cooperate—you know risking their life, or the life of the character. All game theory says if it's a one-shot game that cooperation should not happen, and surprisingly we saw that the most popular answer was actually save the other character—even if that means ending the game.
Listen to the full conversation at the top of the page.