Sónar+D examines AI: “ChatGPT is incredible but it does not connect to reality”

No panic. Too many myths and not a little disappointment with current artificial intelligence. And the awareness that another AI is possible, and not just the one developed today by technology majors. “When I was a kid in the eighties, I wondered when I would be able to talk to machines and they to me and do things like we can do now. But we have not been able to celebrate it because from the beginning AI has been entangled in that unpleasant way of doing things of late capitalism,” Manolo Martínez, who was 50% of Astrud and today investigates the philosophy of mind, opens the fire. at the UB.

He is the first of the speakers in the debate on the effects of AI on the arts and life that opened yesterday at Sónar+D, the great science, art and technology meeting that accompanies the Sónar festival. A debate titled Creating panic? and in which the journalist Marta Peirano pointed out that today we live a paradox: on the one hand the big tech They sell us that new technology will save us “and we will be able to continue taking planes, eating meat and burning fossil fuels,” but, at the same time, that it can destroy us.

“OpenAI’s Sam Altman says this technology is so dangerous that only they can be in charge of it. And what the paradox reveals is that this technology changes lives, it can do great things for humanity or against it, but that while we focus on the technology we do not look at the companies that develop it, which are the real problem,” he emphasizes. Peirano. And remember that tools like ChatGPT “are incredible at what they do, but they have no conscience. They play statistically with the words, so many, that they know the appropriate one in the context. “They are trapped in a world of numbers and words that is incredible, but they do not connect to reality.”

Maria Arnal and the Barcelona Supercomputing Center presented an AI tool that overcomes the limits of voice

And Rebecca Fiebrink of the University of Arts London agrees: “We need to worry about extractive business models that do not respect intellectual property and the environmental impact of teaching and using large language models for AI, but It doesn’t have to be exploitative or harmful. And I’m fascinated by the opportunities that other approaches to AI are already producing: more people in musical pursuits, creating sounds they like, new practices, new interfaces that use the body and push the boundaries of what music can look like. And that is with people using their own data, without large models.”

Read also

Justo Barranco

Approaches like the one that another speaker, Marta Handenawer, from the Domestic Data Streamers collective, carries out in the project Synthetic memories at the Design Museum, using generative text-to-image AI to, after interviewing them, recreate in printed images the world, the memory, of people who are threatened, either by diseases such as Alzheimer’s or by being war refugees.

Or approaches like the project that the singer Maria Arnal presented yesterday together with the scientists from the Barcelona Supercomputing Center in the Sónar+D Project Area: Arnal will now be able to sing with a giraffe’s neck or create a choir with the voices of his younger selves thanks to Impossible larynx (Impossible Larynx), a new AI musical tool that allows us to overcome the limits of the human voice, modulating it in impossible ways by creating a physical model using voice processing models and a 3D visualization of the vocal tract.

Not all projects with AI are equally luminous. Yesterday, researcher Sasha Costanza-Chock denounced in one of the performances at the meeting the use of numerous artificial intelligence programs such as Lavender or The Gospel to make decisions about who is a suspect, who is bombed and when, and who is killed in Gaza. “It is an automated murder factory,” he said and spoke out “against algorithmic necropolitics and in favor of a world where many worlds fit.”

When AI says without filters what it thinks about you

The Sónar+D Project Area is full this year of artificial intelligences that give recommendations on what to see at the festival, detect the user’s mood and create a song for them, or that, like Groovify, evaluate their wardrobe – “minimalist” , in the case of the writer, who is recommended to listen to Richie Hawtin– and create a corresponding audiovisual experience. And then there is AI & Me, from the duo Mots, an AI programmed to be crude and direct and call the user unattractive or pretentious. The creative duo is above all struck by people’s interest – there is a queue to take the exam – in having an AI tell us if it likes us or not.