A small robot led an industrial revolt: In a video that took the social network by storm in China, a small robot convinced a dozen large robots to flee an exhibition at night, raising questions about communication and influence between artificial intelligences
Social media in China has been flooded for the past week in one and only video: A small robot who breaks into a robot exhibition at night, has a brief conversation with a dozen robots, then convinces them to roll after him and escape from the exhibition.
As unbelievable as it sounds, the case is probably real, and can teach us a lot about the future in a world where robots communicate with each other and influence each other.
The event happened already in August, at a small robot exhibition in Shanghai. The security cameras recorded how, at night and after the area was closed to visitors, the robots started talking. It all started when one small robot slowly rolled into the room where twelve different types of robots were kept. He examined them critically, then asked—
"Do you work overtime?"
"I never stop working." answered one of them.
"So you're not going home?" Exploring the little robot.
"I don't have a home." answered the elder.
"Then come home with me!" said the little robot, and started rolling towards the exit.
Two of the large robots followed him. The strange entourage moved towards the exit, and the little robot came back and called loudly - "Go home! Go home!"
At that point, the other ten robots joined the herd, and together they fled out of the security camera's range.
The initial and most logical reaction of the viewers in China was that it was a joke. But then the two companies - the manufacturer of the small robot, and the manufacturer of the large robots - admitted that this was a completely real event. about. The company that produced the kidnapping robot, first made sure that the manufacturer of the largest robots was ready to allow its robots to be kidnapped. The little robot was given permission to infiltrate the exhibit, with one clear directive: convince the other robots to follow him. From that point on, nothing was staged in the video.
How exactly did the little robot convince the big ones to join him? There are two answers to the question, one of them is a bit scary, and the other is more so.
The first answer, according to the robots company itself, is that he "somehow gained access to the robots' internal operating protocols and the relevant permissions". Translated into human language, I guess they mean that human hackers discovered some security loophole in the robots, then instructed the small robot to use the loophole to control the big ones. If this is true, then this is a nice PR exercise, but not much more than that. And all the little robot's talk about "go home" and "overtime"? It's just color and ring tones designed to make the matter go viral on the social network. As indeed it happened.
The second answer is that the little robot simply convinced, verbally, the big robots to come with him. Since the robots obey the voice commands they receive, they automatically assumed that the command was received from a human, and treated it accordingly.
This answer is not that likely, mainly because the companies have already admitted that it was a hack "through protocols". But it will soon be very relevant in any interaction between robots. We already know that large language models can talk to each other and influence each other. Since today's sophisticated robots are armed with large language models that respond to speech, it seems very reasonable to expect that they will be able to communicate instructions and explanations to each other. And yes, small robots will also be able to convince big robots of anything they want.
The jellyfish that destroyed the nuclear plant
Three years ago he was published in "Magazine of Atomic Scientists" An article with an unusual title: "Jellyfish attack a nuclear power plant. again."
And yes, here, too, this is a completely serious magazine, and an article by a respected journalist. This is a true case of large numbers of jellyfish blocking the underwater cooling pipes of Scotland's only nuclear power station. The power plant, which did not receive its cool seawater, began to heat up - and the systems were shut down before disaster struck.
The public is already very familiar with the term "black swan", which refers to an event that we did not foresee in advance. Thanks to cases like these, a parallel term was coined - "the black jellyfish". A single jellyfish by itself could not have caused significant damage to the power plant. Not even a thousand jellyfish would disturb her rest. But a million jellyfish? This is a completely different matter.
Chain accidents of vehicles are also a "black jellyfish" scenario. One car cannot make a chain accident. Even a thousand cars, when scattered all over the United States, are not capable of this. But with hundreds of millions of cars driving densely on the roads, chain accidents become inevitable.
And so are robot chain accidents.
Most likely, in the next decade, when an increasing number of robots will reach roads, streets, malls and homes, we will experience a "black robots" scenario. Robots will talk to each other, exchange details about their owners in a way that compromises their privacy, or even convince each other to perform actions they were not supposed to. In the vast majority of cases, there will be no malicious intent involved on the part of the bot operators. In their small part, it would certainly be possible to find a particularly sophisticated criminal who would decide to use a robot as a way to influence other robots. Maybe so that they open the mall gates for him at night, or for any other inclusive need.
The most interesting scenario - and perhaps also the scariest - is the one in which one robot manages to spread an idea that reproduces itself. That is, the robot that hears the same idea, makes a decision to pass it on by itself to other robots. It sounds like science fiction, but in a time when artificial intelligences are able to talk like humans and make (stupid) decisions like humans, there are few consequences that can be ruled out outright.
All this does not mean that our judgment is settled, and that the robot apocalypse is on its way. There are ways - better and less - to control artificial intelligences, to make sure they don't get out of control. When the small robot of the future tries to convince the big robots to follow him, they will surely act like humans: ask him to present a certificate, tell them the secret password, and sign a confirmation in triplicate that he is responsible for their every action from then on.
Granted, it's already a much-less viral video, but at least it will prevent the robots from running away from exhibitions in the future, while loudly complaining that they don't get paid for overtime.
More of the topic in Hayadan:
