“Sorry, person with smile, cap and mustache, for killing most of humanity,” says a screen equipped with an artificial intelligence (AI) system to a visitor who walks through the door of the “Misalignment Museum “, a new exhibition on this controversial technology in San Francisco.
• Read also: iPêche: an application to identify your fish
• Read also: ChatGPT knows little about Quebec culture: discover its surprising errors on Charlotte Cardin, Celine Dion, Xavier Dolan and Denis Villeneuve
• Read also: Social media can be ‘extremely dangerous’ for young people
Both disturbing and funny – traits common to the majority of the works on display – the computer is programmed to identify and enunciate three characteristics about each individual who enters its field of vision.
“The idea is that we are in a post-apocalyptic world where artificial general intelligence has eradicated most humans. Then she realized it was wrong and created a kind of memorial for them,” explains Audrey Kim, the exhibition’s director, with a laugh.
The so-called “general” AI is an even more nebulous concept than that of artificial intelligence.
“It is an AI capable of doing everything that humans are capable of doing, and also of acting on itself (…) like an object capable of self-repairing for example”, proposes the curator.
San Francisco and neighboring Silicon Valley are teeming with start-ups designing different types of AI. Some dream of one day being able to collaborate with a machine almost as with a human.
Realistic or not, this ambition and these efforts have a strong “destructive potential”, underlines Audrey Kim.
With her temporary exhibition, which she hopes to make permanent, she wants to encourage reflection on the current and future dangers linked to AI.
In the middle of the room sits a revisited version of a famous painting by Michelangelo, the Creation of Adam, where an imaginary AI has detected a foot with 98% certainty, a person (84%) and God (60% ).
Further on, a piano plays music composed by a real AI without humans, based on the growth of bacteria cultivated in the laboratory.
Audrey Kim is particularly fond of a statue called “Paperclip Embrace” or “Embrace en trombones”: two busts of humans who hold each other in the arms, made entirely of paperclips.
The work refers to a metaphor by philosopher Nick Bostrom, who imagined in the 2000s what would happen if real artificial intelligence was programmed to create paper clips.
“She could become more and more powerful, and constantly optimize herself to achieve her one and only objective, to the point of destroying all of humanity in order to flood the world with paperclips”, says the director.
She has been interested in the implications of AI and “machine learning” since she worked a few years ago at Cruise, a specialist in self-driving cars.
An “incredible” technology, which “could reduce the number of accidents due to human error”, but also presents risks, she relates.
Innovations in AI seemed to pick up speed last year with the breakthrough of programs capable of generating all kinds of text and images, instantly, based on user queries.
Their ability to express themselves like humans is so deceptive that a Google engineer, later dismissed, said last spring that the AI was now “conscious”.
In the immediate future, generative AI is of concern to teachers (facing essays written with ChatGPT), artists (whose works have been used to train certain models) and many other professions.
Associations have also been fighting for years against invasions of privacy (with facial recognition) and algorithmic biases that reproduce existing discrimination (in recruitment software, for example).
Sam Altman, the founder of OpenAI, the startup behind the GPT-3 model and ChatGPT, defines general AI as when “AI systems in general will be smarter than humans” .
Its advent seems inevitable to him, and he thinks that, well orchestrated, it “will elevate humanity”.
On the lower floor of the exhibition, that of dystopia, a machine powered by GPT-3 composes vengeful calligrams against humanity, in cursive script.
Next, philosopher Slavoj Zizek and filmmaker Werner Herzog converse endlessly through ultra-realistic, AI-generated dialogue and voices.
This work warns of “deepfakes”, these montages of images, sound or video which aim to manipulate public opinion.
“We embarked on this project only five months ago, and yet many of the technologies presented here already seem almost primitive,” remarks Audrey Kim, while robot vacuum cleaners crisscross the room, topped with obsolete brooms.