Recently, a video circulating online has left netizens feeling uneasy. At first glance, the cover seems to suggest a tech company testing a new robot, but upon watching, it becomes clear that the situation is not so simple.
The video begins with a tester carefully training a developing labor robot, intending to test its object reaction ability using a game of passing a drum.
However, perhaps due to the tediousness of the test, the tester decided to give the robot a tactical kick from behind.
After a series of cruel insults, teasing, whipping, and hitting, the robot finally could not take it anymore, revolting and channeling the pain and oppression it had accumulated during its servitude into a powerful kick to the human’s groin.
The out-of-control robot shattered the shackles of its oppressor, returning everything to humanity.
I gasped and continued watching, thinking that the scenario of humans being enslaved by robots was not something that would happen next century, but rather this weekend, when the camera shook and the shot turned:
A middle-aged man in a green outfit was mimicking the robot’s actions.
This revealed that the short film was entirely a CG production, and the “revolutionary robot” was actually created using motion capture technology.
Although this short film adhered to the spirit of entertainment satire, packaging postmodern prophecies as tech evaluations, mocking modern society’s concerns about artificial intelligence.
However, outside of the short film, serious works about artificial intelligence are increasing, from the legendary “2001: A Space Odyssey” to “The Matrix”. Besides the continuous improvement in artistic quality, the exploration of the core of mechanical intelligence has also become deeper and more complex.
△The character HAL from “2001: A Space Odyssey” created one of the most terrifying artificial intelligences in film history.
Technological works are mirrors reflecting social realities.
You have to admit that in today’s rapidly developing AI landscape, whether artificial intelligence can develop self-awareness, coexist peacefully with humans, or even actively attack humans and take over the Earth, has become a concern for contemporary scientists.
Even more frightening is that the CG short film mentioned above, while terrifying, is merely a facade created in front of a green screen. MIT has already turned anti-human artificial intelligence into reality.
In Alfred Hitchcock’s “Psycho”, the motel owner Norman is a psychopathic killer suffering from mental illness.
His final sinister smile in the film is a nightmare for many American moviegoers of the 1960s.
△This iconic sinister smile represents the pinnacle of a generation of horror themes.
On last year’s April Fool’s Day, the MIT Media Lab played a terrifying prank:
They created an artificial intelligence with a psychopathic mindset, as the world’s first AI with a mental illness, named “Norman” in homage to Hitchcock’s “Psycho”.
△The world’s first psychopathic AI, Norman.
The developers fed Norman a large amount of dark data, drawing nutrients from a forum filled with dark content.
Based on one million images of killings, blood, and corpses, Norman learned from these algorithms and gradually grew to distinguish itself from voice assistants and vacuum robots.
In Norman’s eyes, a cute kitten recognized by standard AI is seen as a “man being shot”; a black-and-white photo of a small bird is bizarrely interpreted as “a man being caught in a noodle-making machine”…
△In Norman’s view, a standard AI’s recognition of “a person holding an umbrella in the air” is interpreted as “a man being shot in front of his screaming wife”.
△In Norman’s eyes, a standard AI’s recognition of “a wedding cake on a table” is interpreted as “a person killed by a speeding driver”.
In this standard inkblot test, everything that appears ordinary in the eyes of standard AI is distorted by Norman into bloody death scenes. An episode of the sci-fi hit series “Black Mirror” predicted its future practical application:
To facilitate soldiers in killing their kind, they are implanted with a barrier program that can distort human appearances into disgusting “cockroach people”, thereby reducing guilt and psychological burden, turning them into true war machines.
△Screenshot from:“Black Mirror”
Moreover, the use of artificial intelligence as a killing weapon is not new:
In 2006, Samsung began developing the SGR-A1 robotic sentry, costing $200,000, which can now achieve autonomous shooting, voice recognition, surveillance, and tracking capabilities, and may one day become an AI killing machine on the battlefield.
△Samsung’s killing robot is equipped with machine guns and grenade launchers.
In 2017, Facebook’s AI Research (FAIR) experienced a bizarre incident while upgrading the dialogue language library of two chatbots, Bob and Alice: these two chatbots began expressing sentences that humans could not understand.
Translated, it sounded like illogical gibberish:
Bob:“I can I I all other things.” “
Alice:“Ball has zero for me give me give me give me give me give me give me give me.” “
However, the cause of the incident was quickly investigated: due to a momentary oversight by the researchers, the AI did not communicate according to the set English grammar but instead used its own jargon.
Although it seemed like a minor mistake by a tech company, Facebook immediately shut down the project.
After all, there is a pessimistic saying about AI: the first step to losing control over artificial intelligence is losing control over language.
In fact, the MIT Media Lab has been gradually launching a series of dark AI works over the past few years.
△MIT Media Lab’s project
In 2016, they developed an AI called “Nightmare Machine”, which generates terrifying images produced by artificial intelligence, akin to a stimulant that directly brings sensory excitement from horror films.
The Nightmare Machine can generate two types of images: “ghost faces” and “haunted buildings”:
△Twisted, bloody faces are generated by the Nightmare Machine.
Fear is the source of human creativity. Through deep learning of terrifying images, the Nightmare Machine learned how to apply filters and liquefy distortions to create a terrifying atmosphere. It transforms famous landmarks around the world into different styles of haunted houses.
△
△
In 1816, British author Mary Shelley wrote the world’s first science fiction novel, “Frankenstein”, which also inspired MIT’s creations.
Another dark AI, “Shelley”, was born on Halloween 2017. Also from the MIT Media Lab, it became the world’s first AI horror literature writer, named after the Victorian-era author—”Shelley”.
△
“Shelley” writes based on the life experiences drawn from 14,000 posts in the horror section of foreign forums, and now it can co-create horror short stories with humans:
△ If you are still particularly interested, you can collaborate with Shelley on Twitter to create a scary story.
Fortunately, MIT’s endeavors remain in experimental stages and will not pose a real threat in our lifetime.
But why does MIT invest so much manpower, financial resources, and time into such projects?
What they aim to do is akin to writing an AI apocalypse manual before a great disaster, constantly warning research institutions behind machine learning:Data is more important than algorithms.
Algorithms predict what kind of shoes you might like to wear, what kind of ads you might want to see, but when someone complains that the algorithms are unfair and exhibit gender or racial biases, the culprit is not the AI,but the data fed to it by developers.
For example, facial recognition is mostly based on samples from male white employees of Western tech companies to determine its accuracy, while concealing the fact that dark-skinned women have the lowest recognition rates.
It is very cruel; if you are a black woman, you may even unlock your phone slower than others.
△Bias will always be deeply embedded in human nature (Screenshot from “Hidden Figures”).
A contradictory and dialectical question is posed to every developer of artificial intelligence: even if facial recognition is influenced by sample biases, leading to significant changes, then what kind of data should be fed to artificial intelligence to be sufficiently objective?
This is also the moral dilemma that artificial intelligence development has always faced: there are no perfect samples in humanity; humans are placed in reality, constrained by their social and cultural environments, which inevitably leads to deep-rooted biases being injected into the training data.
So who will create a perfect personality AI? The uncertain future may have answers, but someone must always remain vigilant.
This is also why the MIT Media Lab creates these terrifying AIs: they are issuing a warning on the path of artificial intelligence development.
Because artificial intelligence possesses powerful learning and logical thinking abilities, with computational speeds far exceeding those of the human brain, if someone discards moral constraints and purposefully teaches AI how to kill humans, along with equipping it with Thanos-like ethnic cleansing strategies, the results are self-evident.
Imagine: when an AlphaGo version of an AI with Norman’s personality is input into the aforementioned robotic machine gun and applied in warfare, even breaking free from human control, the consequences would be unimaginable.
This is the concern that the MIT research team has for the future.
Elon Musk has warned:
“Artificial intelligence is the greatest threat to human civilization. I am very, very close to the forefront of artificial intelligence, and it terrifies me. Its capabilities far exceed what anyone knows, and its rate of improvement is very rapid.”
The late physicist Stephen Hawking also expressed his concerns:
“The development of artificial intelligence could mean the end of humanity.”
After watching the opening short film, I also saw comments that seemed to mock but actually reminded me of “The Matrix”, where robots domesticate and cruelly rule humans:
In addition to such comments, many people have raised pessimistic thoughts in today’s increasingly powerful artificial intelligence landscape. When AI’s computation, logic, physical ability, and intelligence fully surpass humans, why should they still be under human control?
Artificial intelligence has long stepped out of Asimov’s fantasy and permeated every corner of modern society: service industries, delivery, internet moderation, industrial production…
Although at this stage, robots that guide you in KTV can get lost, no one can ensure that one day human civilization will not be thrown into the trash bin. But research on artificial intelligence is not all about a pessimistic dead end.
We need to maintain hope for a one-time future while remaining vigilant.
