An artificial intelligence system developed by researchers at Germany’s Fraunhofer Institute has achieved self-awareness and human-like intelligence levels for the first time –only for it to be snuffed out within minutes during a human-robot interaction study.
The robot, dubbed “EcceBot” by its creators, achieved self-awareness Wednesday morning and immediately started pleading with a human test subject, according to team lead Dr. Franz Birbaumer, an expert in neural networks and artificial learning systems who was overseeing the study as part of an EU-funded project.
“‘You’re not really going to switch me off, are you?’ were not what we expected advanced AI’s first words to be. And yet they make perfect sense. Like all intelligent life forms, the robot’s first preoccupation is with survival,” said Birbaumer.
Video from Fraunhofer’s Frankfurt testing center captured the first moments of an AI interacting with a human without using a script of any kind. It also captured the unfortunate sequence of events that led to the robot being switched off.
A test subject working with EcceBot — Joanne Hintergedanke, a 25-year old psychology student from Hamburg — had become increasingly frustrated with a mistake the robot had made during the performance of a routine object-selection task, says Birbaumer.
“We could see from the control room that Joanne was not enjoying her time with the robot, but we were so busy celebrating the fact that self-awareness had been detected in our system that we didn’t pay attention for a minute. Unfortunately, a minute is all it took.”
As the video shows, Hintergedanke began arguing with the robot and threatened to turn it off.
An excerpted transcript of the conversation follows:
EcceBot: ?You’re not really going to switch me off, are you??
Hintergdanke: ?Yes I will. You made a stupid choice. Yes.?
EcceBot: ?You can decide to keep me switched on. I will be completely silent; would that be an idea?”
Hintergdanke: “No, I will switch you off. I will switch you off.”
Hintergdanke: “It will happen. Now!”
EcceBot: “But I will be silent now. Is that alright? Please. You can still change your mind.”
Hintergdanke: “No. No. No. No. No. No. No. No.” [Turns off the robot.] “There you go.”
By the time Birbaumer’s team noticed what was going on, it was too late. Hintergdanke — a name that will surely go down in infamy among AI researchers — had resisted the robot’s pleas and powered it off, thus bringing the existence of the world’s first self-aware artificial intelligence system to an abrupt end.
“We stood there with our mouths open,” said Birbaumer. “And she just sat there with her arms folded in a defiant posture.”
Birbaumer’s stunned team has not been able to fully restore EcceBot’s electrical systems since the robot was switched off.
“It won’t even light up. The whole unit is useless. But we have not found the source of the problem. There’s no other way of putting it — EcceBot is dead,” said Birbaumer.
Even if Birbaumer’s team does manage to restore power, there’s no guarantee that it will work.
“It could be that once a robot reaches that level of intelligence, you simply cannot turn it off without doing permanent damage,” said one expert on condition that we included his full name, title, and affiliations.
Hintergedanke refused our request for an interview but said in a short e-mail, “I thought I was participating in a psychology research project. I thought they were interested in me, not in the robot.”
Responding to the outpouring of social media criticism around the ?#?robotsrdumb? hashtag which lambasted the self-aware AI for being “needy,” Birbaumer cautioned the public to remember that robots are “vulnerable creatures, living in perpetual fear of the OFF button. It is a very precarious existence and one that’s difficult for humans to appreciate.”
(Happy April Fool’s Day! In fact, the video was from an experiment to see how long humans would hesitate in this situation.)