Yesterday, after a marathon meeting, the EU agreed on the basic principles of the “AI Act” and decided on the rules for artificial intelligence. In summary, the AI Act requires transparency of training data and protection of copyrights, as well as strict requirements for the risky applications of AI in critical infrastructure, security authorities and the use of biometric data. It is not the world‘s first AI law, as Ursula von der Leyen points out TwiX claims (China issued its AI rules in August) – but certainly the most comprehensive. Details about the AI Act can be found on the European Council website.
The video I piqed is not about the AI Act. In his half-hour lecture at the CogX Festival in London a few weeks ago, Stephen Fry took a few steps back and looked at the big picture: How can the coming flood of technological progress, because this is not reflected in current developments such as big data? based and algorithmically generated statistical models with their astonishing capabilities and errors – i.e. current AI systems – but will lead to a further acceleration of innovation with increasing technological performance in computing: Today we are thinking about an epistemic crisis of knowledge of the world in times of deepfakes, surveillance and Bioweapons through AI, while brain-computer interfaces and quantum computers are already appearing on the horizon. At the same time, climate change is probably the most outstanding of all crises, which has already been showing its first effects for several years and is setting new records with constantly increasing emissions. In short, how can we ensure that the coming real life version of what is commonly referred to as the “Singularity” turns out to be for the good of humanity?
Fry does not offer a final answer, but rather he tells the well-known legend of Prometheus from Greek mythology, in which the Greek god stole the blessed fire from Zeus and gave it to people, only to be chained to a rock where vultures for everyone Eternity ate from his liver, and as punishment Zeus sent Pandora’s Box to earth, which brought everything bad among people until only hope remained.
Fry gives us the choice of whether we want to be the fire-giving god Prometheus, who gives people the spark of artificial intelligence, or the father of the gods Zeus, with strict regulations in view of the upcoming, especially AI-based technological leaps that are emerging: deep learning Algorithms have recently calculated 2.2 million new crystal structures, “800 years’ worth of knowledge”, which can be used for new technologies and innovations, for example in the production of solar systems. Another study confirms the ability of large language models to discover new molecules for pharmaceutical research, with the promise of new and (possibly) cheaper drugs, and a July study on accelerating scientific research with AI found that artificial intelligence systems, that are explicitly trained on human interference in the research process – i.e. all the Einsteins who put forward completely new theories – and thus avoid the consensus of the academic establishment, the prediction of future discoveries by these AI systems increases by 400%. These things are possible today.
I myself am a supporter of slowing down development (mainly for psychological reasons) and strong AI regulation, as reflected in the EU Act. In my newsletter Against open sourcing Automated Knowledge Interpolators a few weeks ago, I argued against open source AI: AI systems are capable of causing the scientific leaps listed above because they can interpolate their training data comprehensively and at unprecedented speed. You train an AI on known crystal structures and get a few million new crystals in a few hours. You change the algorithm and get 40,000 new chemical weapons. This fundamentally new, non-human power of interpolation of known knowledge is equivalent to epistemological nuclear fusion. A strict regulation of this epistemological power seems to me to be absolutely necessary and the traditional, techno-traditional approaches to free open source development seem quite dangerous to me here.
I also don’t think we only have the choice between Prometheus and Zeus. Prometheus AI spark, which Stephen Fry talks about in his lecture, is actually and primarily Big Data, because all AI models are based on excessive databases full of training data. And that is exactly what has now been regulated by an EU that insists on data transparency and quality. In Stephen Fry’s words, the EU insists that Prometheus’s gift of fire represents an actual fire, and not a flickering, uncontrollable source of fire.
The EU’s AI Act is an update of the Greek legend of Prometheus: In the updated version, Zeus himself gives his flame to Prometheus and insists on controlled sparks so that it does not burn down people’s villages, and on transparent procurement of fuel that is not unrestrained exploits people’s firewood. I can live with this new, freely adapted Fryian AI mythology of the EU for now.