If the AI routine doubts itself…

If you study the workings of a human cell, you get a fraction of the infinite possibilities and richness of human life. For example, the mitochondria that make energy from any number of elemental substances, release proteins, make viruses (yes indeed!), and co-manage the body. And that times x trillion. Your body is and always will be a miracle. And much more complex than that one seed you put in the ground that grows into a fantastic beech tree. All that knowledge is hidden in that very small chunk of life energy. Can that also be replicated with human technology? Opinions and futures on that are divided. Perhaps with huge quantum computers. For now, the shortage of electricity is still a delaying factor. If only we had listened to Nicolas Tesla.

Because our brains are simply not that fast at processing piles of information, we look in awe at all those supercomputers supporting our Artificial Intelligence facilities (or at least: the computational rules and computing power of superchips). Fast forward the current AI applications: human less vehicles for agriculture, robots wash windows and deliver parcels, X-rays can be better screened by an AI, text translation or subtitling can be done without humans in fractions of a second. The English translation of the latest KNVI book which comprises 279 pages, took less than 30 seconds.

Meanwhile, as humans, we must make do with inferior information. Our brain cannot store and retrieve everything neatly. Of course, AI applications don’t suffer from that. Although right now AI is “Incredibly Smart” and at the same time “Shockingly Stupid” according to Yejin Choi. Still, imagine if a set of AI routines did collect all this information and extract a common thread from it. Would that routine then choose self-preservation (‘I exist’), reproduction, cooperation with humans, or making this earth better? With the ultimate option that the earth could do without humans.

And: what would that routine then do with factual knowledge, for instance that we need to increase mining by 1000% (by 2024) to get enough minerals and precious metals for batteries and solar panels (even though the Energy Pay-Back Time (EPBT) of panels is debatable). Or that we dig huge pits for discarded rotor blades, get coal from Australia to keep mines open in China, and bring waste from Italy to the Netherlands to burn power stations. Or that we deforest West Papua and the Amazon for even more logging and palm oil, and the biggest gas users are the only ones paying very little tax…. Perhaps that ‘AI routine’ will even start to doubt itself if ‘it’ knows that a workout run by ChatGTP generates about 500 tonnes of CO2 emissions or to further analyse (and correct) the IPCC report. In any case, Geoffrey Hinton and Mo Gawdat have serious concerns about what these AI routines have in store for us. And we as information professionals? Soon to be hired as AI routines too perhaps. Might cost a bit of energy and a bit of CO2 emissions, but then you have something.

Written by Wouter Bronsgeest, IFIP Councillor