This text belongs to 'Artificial', the newsletter on AI that Delia Rodríguez sends out every Friday. If you want to receive it, sign up here.
Dear readers, dear readers:
Is life what happens while we wait to try Apple's Vision Pro? It's possible, but in the meantime, let's dance.
Dancing is literal. As we recover from Paul McCartney's announcement that by the end of the year we will hear one last Beatles song reconstructed from a John Lennon recording, the center of the world of artificial intelligence applied to music is Barcelona, where Sónar D is bringing together some of the leading researchers in the field.
Here is, for example, CJ Carr, a 35-year-old Bostonian, half of the Dadabots couple, and Harmonai's research director. His was the idea, years ago, to put Sinatra to sing Toxic by Britney Spears. Maricel Chavarría tells that she twists the limits of electronics with algorithms, and that the more "weird and indigestible" the result is, "the more her eyes shine." About the creativity of the machines, she says that "the culture of effort will be replaced by the culture of the concept. Thinking of something and having it ready to the minute is something very powerful." An example of her work and one of her talks are available on the Sónar website.
The Barcelona Supercomputing Center has also participated in an interesting experiment: feeding lyrics from the festival's 30th anniversary (some 40,000 songs) to ChatGPT. The conclusions? It has turned from electronic to urban music and the sadness in the works has increased slightly.
“It's a little scary but it's exciting, because it's the future. We will have to see where it takes us, "said Paul McCartney after his announcement, and I suppose that's how we all are a bit, Paul, although perhaps the artist Grimes does live in the future, because she has been giving up the rights to use her voice for a while to whoever wants it in exchange for a percentage of 50% of the benefits. About the latest changes in that industry, which are many and very interesting, this report by Sergio Lozano summarizes them well.
One of the concerns of the music industry is the increase in the volume of creations and their automation. "Surely in five years we will see companies dedicated to distinguishing what is real and what is fake. But before that we will see a lot of music generated by AI that will not come to light as such. It will be a problem, impossible to discover. And every time a detector of the fake, there will be a way to circumvent it, and the AI will be smarter and smarter," says CJ Carr to Maricel Chavarría.
Will the music be littered? It is another way of wondering, will the internet be filled with garbage?
On this, what is happening on Reddit is revealing. The most important forum on the internet has been looted by the great language models who have squeezed the wisdom (and the way of speaking, and the way of thinking and writing) of its members. A few months ago, the management decided to try to make the training of the AIs profitable by paying for their API (the code that allows them to communicate with third parties), something that has affected some third-party applications that redditors liked a lot because they helped them read and manage better channels. Users are very angry and on strike. Part of it is consisting of making thousands of subforums private.
That, in an internet where quality information is increasingly closed (in private communities, chat rooms, media with a paywall, newsletters, etc.) is a very bad sign. "We are living the end of the useful internet," says Alex Parenee at Defector. “The future is informed discussion behind closed doors, on discords and private forums, with the public web increasingly littered with the detritus generated by the great language models, which bear only a stylistic resemblance to useful information. Finding unbiased, independent product reviews, expert technical support, and helpful tips will be a bit like searching for illegal sports broadcasts or pirated magazine articles. The decades of real human conversations hosted on places like Reddit will be useful training material for mindless robots and mindless bots and the fraudulent sellers that replace them.”
The journalist Ryan Broderick recalls that Meta soon closed its contents to search engines, hiding them, and that it now charges for verification, and that Twitter set an unaffordable price for its API. “The companies that colonized the web don't want to be colonized,” he writes.
After reading these texts, I think of a thread by the designer Jimena Catalina where she explains how she is adapting the Rechupete Recipes website to a future full of junk recipes. She proposes to highlight from the design that behind it there is a human who is responsible for that recipe coming out, and that it is not an invention of ChatGPT. Perhaps this has to be done with all the contents, as well in general.
A fascinating return to that future where the web is a dump is that it can end up poisoning those who pollute, because the robots will have no choice but to feed on their own waste, generating a collapse of the model. Researchers from the universities of Cambridge and Toronto warn him.
What else has happened this week
- Europe continues to advance at a good pace in its regulation. The European Parliament has approved the AI law proposal but vetoes facial recognition. This means two things: that the police forces of the countries will have to continue requesting judicial authorization after the fact and that Europe seems to take seriously that fundamental rights are not lost with the excuse of technology.
- Commissioner Margrethe Vestager has said that what worries her about AI is not a future extinction, but that an obscure algorithm decides whether or not to give you the mortgage.
- Those who are following the European -and world- problem with chips can read about this IBM-Barcelona Supercomputing Center agreement. By the way, AMD has announced a chip that will compete with NVidia.
- Black Mirror is back. Pere Solà recommends Mrs Davis and interviews a Catalan scriptwriter who is on strike in Hollywood: “We want the agreement to make it clear that every project must have a scriptwriter and that this scriptwriter must be a human person. It sounds absurd to have to clarify this, I know.
- That such a great advance in the history of humanity is not in the hands of scientists working quietly in their universities but in the hands of scientists working in private companies has everyone quite worried. The topic comes up in the conversation between Nuria Oliver, engineer and director of the Ellis Alicante Foundation, and Josep Maria Martorell, deputy director of the Barcelona Supercomputing Center: "AI is, in quotes, the first scientific field in recent history where leadership of the investigation is carried out by the private world”. NATO chief scientist Bryan Wells also talks about it to Alexis Rodríguez-Rata: “50 years ago defense was the main engine of scientific change, now you only need to look at the scale of investments by Google and Tesla to recognize that the private sector is increasingly driving it; It means that you have to watch what they do to add that little element that makes their advances applicable to defense.
- A well-rounded phrase by Màrius Carol: “If Twitter managed to erase the barrier between truth and lies, AI blurred the line between what is human and what is algorithmic, and Apple aspires to eliminate the border between what is real and what is virtual”.
- Very interesting Manuel Castells, the great sociologist of the network. He says that the big challenge now is privacy.
- How nice that AI has served to discover new Nazca lines.
- 42% of CEOs say that artificial intelligence could destroy humanity within 5-10 years, which is very disturbing but for other reasons.
- The Uncanny Valley (Uncanny Valley) is, in addition to a highly recommended novel by Anna Wiener about the culture of work in Silicon Valley, the name of the effect for which robots and chatbots make us cringe when they try to imitate humans. National Geographic report in Spanish.
- Google has introduced some improvements based on AI in its applications, such as the identification of skin pathologies in Lens or the test on virtual models in Shopping.
- Days after trying Apple's Vision Pro and with more calm, Francesc Bracero reflects on them again and they continue to freak him out.
IAnxiety levels this week: maximum. As a digital journalist and great lover of the internet, "we are living the end of the useful internet" is not the phrase I wanted to read.
