We interact with it, it conditions more and more aspects of our lives and, even so, many are unaware of how it works. Although its use is expanding at a frenetic pace, the artificial intelligence (AI) remains a great unknown. Jordi Torresresearcher of Barcelona Supercomputing Center (BSC) and professor at the Universitat Politècnica de Catalunya (UPC), wants to help change that.
After more than two decades of scientific research, he has now presented the book ‘The artificial intelligence explained to humans’ (Plataforma Editorial, 2023), in which he calmly explores the social, economic, political and even cultural impact that this technology. He also advises the ‘Artificial Intelligence’ exhibition that the CCCB will host starting next October 18.
AI started in the middle of the last century, but in the last year its popularity has skyrocketed thanks to ChatGPT. It was a surprise?
Yes, the result has been spectacular. It was not a technical surprise because the language model it used (GPT3) was already known in academic circles, but it was surprising how people jumped to participate in this experiment.
As we have seen with ChatGPT, the term AI leads people to imagine robots with consciousness. What would be the best way to describe it?
It is a very old term created in 1956 and since then what it means has changed considerably. This abstract concept allows us to reference at all times what is the cutting edge in this field. AI is nothing more than the evolution of computing.
Large language models (LLM) like ChatGPT are trained with databases and respond to user questions from them. AI experts have compared these conversational assistants to “stochastic parrots”…
Yes, AI algorithms are like a parrot that speaks very well but does not understand what it is saying or what causality is. And that is essential.
“AI is like a parrot that speaks very well but does not understand what it is saying or what causality is”
They respond plausibly, but they also make things up. Can this contaminate information on the internet?
That’s how it is. We have already used almost all the data on the internet to train language models. The generation of synthetic data (those produced ChatGPT) contaminate, since the algorithm can learn to consider false information to be good information. If you use that misinformation to train new models, you distort the original sample, making false answers more likely. Thus, you can increase the amount of data with which you train an AI, but no new knowledge is generated. However, some of the information that was already on the internet before ChatGPT is false. We have seen it with the rise of fake news, created by humans.
The promoters of AI systems, such as Sam Altman, assure time and time again that their objective is General AI, like the one that Terminator or the HAL9000 robot have in ‘2001: A Space Odyssey’. Is it credible?
Machines have no conscience no matter how many movies are made. There is always a human behind telling you what to do. Stuart Russell says we’re still several Einsteins away from getting there. For now, with current technology, it is something that is not plausible in the short term. It is a topic that does not concern me. I prefer to focus on the good uses it can have for science, but also the negative impact like the fake nudity we have seen in Badajoz.
You assure that AI will be the greatest revolution of humanity, comparable to steam, electricity or Internet.
The world will change a lot, but I don’t know if it will be solely because of AI. Life before and after the internet is not the same, but we have survived. We must assume that we have to understand it to decide where we want this revolution to take us. Greater collective awareness is needed.
“Life before and after the internet is not the same, but we have survived”
Even so, big technology companies reiterate that it also threatens humanity. Is that a smokescreen?
Figures like Elon Musk, who signed a letter asking to stop the development of AI while at the same time investing in creating his own company, do not generate much trust. AI is a very powerful tool and you have to be careful. Its current state already worries me because it can do a lot of damage, facilitating misinformation or with its application in the military sector, which should be prohibited.
In 2005, 18 years ago, you joined the BSC as part of the founding team. What is a supercomputer for?
They are the most powerful computers that exist today. They allow large volumes of data to be processed with algorithms at a very fast speed. They do what a simple computer would take about 3,000 years to calculate. In Europe there are only three of these supercomputers and one is in Barcelona. And here all types of cutting-edge scientific research can be carried out so that, for example, healthcare advances.
More complex systems are equivalent to a greater need for data, computational capacity, calculation speed and, therefore, energy expenditure. How to combine the development of AI that can serve to combat climate change with that growing ecological bill?
We cannot hide that climate prediction models to stop climate change are consuming in ways that harm climate change. Some studies indicate that training GPT3 is equivalent to releasing 500 tons of CO2. Behind your GPS there is a server infrastructure that consumes brutality. Everything consumes a lot, but we cannot give up on AI because it can help us improve the future.
What evolution of AI will we see in the next 10 years?
It will change everything. As? Don’t know. In the last year, all sectors have made their hypotheses about how AI will affect them, some of them very exaggerated. With generative AI we have advanced by brute force, with more data and power, but I think we will not continue with this exponential growth. Chips and data are missing for this.