Eerder verschenen in TG Magazine – Februari 2024.

I was born in the middle of the first uproar of artificial intelligence (AI). The field of research was founded at a workshop held on the campus of Dartmouth College, USA during the summer of 1956 [1]. But of course I had no notion of this at that time. However, near the end of primary school I got fascinated by the digital computer. There was a big one, with vacuum tubes and lots of blinking lamps, exhibited in the NINT (Nederlands Instituut voor Nijverheid en Techniek te Amsterdam) [2], now participating in the, in my humble opinion, much less interesting NEMO Science Museum[3]. I spent many an afternoon in NINT where almost everything could be handled by kids, it was great fun.
Cybernetics and robotics
The fantastic experience in NINT was one of the many reasons why I insisted on studying electronics after my secondary school, even when at the time there was no formal education. There were private schools, most of them had only courses in the evening. And there was one, quite near the Heineken brewery in Amsterdam (!), that offered a day school. That one I joined and I enjoyed myself greatly there. Later the school merged with a polytechnical school in Alkmaar; that was one year before my graduation. It was then that I again heard about artificial intelligence. It was a hot topic in the Netherlands so during the course on medical electronics not only the human cell and its potential jumps were discussed but in particular the interconnection of cells as in the brain. I remember a discussion where we fantasized on the idea of making an electronic equivalent of such a network. But that was not the only course. There was also measurement and control technology where cybernetics, a name coined by Robert Wiener that we know in connection with the stability theorem. The idea of Wiener went much further as the title Cybernetics: Or the Control and Communication in the Animal and the Machine [4] of his 1948 book betrays. The connection with robotics is easily made and indeed some of us were – during their practical year – involved in the design and realization of simple machines that could perform specific tasks. A little less obvious connection was maybe the course on telecommunications. This had to do with the connection between bandwidth of a channel and information throughput, the Shannon-Hartley theorem [5]. Indeed, the control of robot-like machines involved information exchange. The method of choice was wireless communication and their noise, signal strength and bandwidth determined the information exchange rate.
Games

And then it stopped. I fulfilled my military service and after that devoted my time to working as an electronics engineer in a cancer research group. It was the “first AI winter” where development in the field of AI was minimal. Only a very early computer game drew my attention for a while. It was called Dungeon then but because there was already a game with a similar name it was later renamed to Zork [6]. It was command line driven; short instructions led you deeper and deeper in the maze. Some got completely hooked to the game, see figure [7]. Those interested in the game can still play it; I am happy to assist in making the program run again on an available platform. One might wonder if such a computer game has a lot to do with AI. Well, at the same time games were developed for checkers, chess etc. that are by now considered AI-based systems.
Computer algebra

The AI winter lasted when I started my physics education. In the courses there was no mention whatsoever of AI. I did follow a course called Informatics, but that was largely about number representations and some formal programming. During my final research project at the Lorentz Institute for Theoretical Physics there was a connection: computer algebra. Especially the students working in high energy physics used it heavily to sort out lengthy polynomials as arose from nuclear collision studies. There was – and still is – a freeware program for that: REDUCE [8], see figure. I myself used it as for my thesis work for which it was quite sufficient. The high energy physics students, however, needed speed and none of the available programs could deliver that. That also bothered the physicist and later Nobel laureate Tiny Veltman, who developed SchoonSchip.[9] The version that was widely used amongst the students ran on the then popular Atari ST 1040[10]. Needless to say that I owned an ST myself and used it to run REDUCE. Apart from that, it could also run LaTeX [11] to write my thesis with. Later Veltman became the guest Lorentz professor at the Institute and for a while he was my roommate. We discussed many topics such as the electronics behind telephone exchanges, but not the origin of SchoonSchip. That project was already laid off by him.
Neural networks
My doctoral thesis project was on spin glasses [12], that is on random magnetic systems. The intriguing property of these systems is that below a specific low temperature the dynamics stops: the system freezes. Apart from being non-ergodic below the transition temperature, the mean field model for spin glasses revealed the presence of a multitude of low energy states. Upon freezing the system it could end up in one of these low energy states depending on the exact conditions during the experiment. It is non-ergodic in the sense that the chosen low energy state is not necessarily each time the same. My thesis was on the dynamics of spin glasses [13].
At the same time, John Hopfield [14] found a model for neurons and their connections that could learn and process information in a completely new way. This model resembled spin glass models so much that the same mathematics could be used to describe them. The patterns of stored information in a neural network are the low energy states in which spin glass models may end up upon freezing: the peculiarity of spin glasses becomes the trait of neural networks!
After my graduation, I did not spend much time any more on the topic. Also AI entered into its second winter only to revive much later, around the beginning of the new milennium. Reasons for the revival are the ever growing computational power, the application of advanced mathematical tools not in the least statistics and probability theory. Only very recently, the first products became available to a larger public in the form of ChatGPT, Bard, Llama and the like.
Limitations

The recent exhibition BrAInpower at the Boerhaave museum in Leiden [15] gave a nice overview of the development of AI over the years. It also included statements from leading scientists on the topic. The one by Stephen Hawking drew my attention The development of full artificial intelligence could spell the end of the human race (…) Humans, who are limited by slow biological evolution, couldn’t compete, and would be superseded [16]. Let us for a while assume this statement to be true. Then, something must have been done to find new ways to convert available resources to satisfy the outrageous need of electrical energy by AI systems, see figure. At the same time, global warming should not be an issue any more as otherwise the higher environmental temperatures would not even allow AI systems to run at sufficiently high speeds. In other words, even if the prospects for AI systems are great, the practical limitations are as yet enormous. In that, human beings are much more energy efficient!
Met dank aan Hans Geerlings en Wim Haije voor het kritisch nalezen.
[1] https://en.wikipedia.org/wiki/History_of_artificial_intelligence
[2] https://nl.wikipedia.org/wiki/Nederlands_Instituut_voor_Nijverheid_en_Techniek
[3] https://www.nemosciencemuseum.nl/nl/
[4] https://en.wikipedia.org/wiki/Cybernetics
[5] https://en.wikipedia.org/wiki/Shannon%E2%80%93Hartley_theorem
[6] https://gunkies.org/wiki/Zork
[7] https://porkrind.org/missives/dungeon-zork-map-in-duplicity/
[8] https://reduce-algebra.sourceforge.io/
[9] https://nl.wikipedia.org/wiki/Schoonschip
[10] https://nl.wikipedia.org/wiki/Atari_ST
[11] https://nl.wikipedia.org/wiki/LaTeX
[12] https://en.wikipedia.org/wiki/Spin_glass
[13] G.J.M. Koper, On the dynamics of spin glasses: a theory of aging, doctoral thesis University of Leiden, 1990.
[14] https://en.wikipedia.org/wiki/John_Hopfield
[15] https://rijksmuseumboerhaave.nl/te-zien-te-doen/brainpower/
[16] https://www.linkedin.com/pulse/stephen-hawking-artificial-intelligence-civilization-destroyer-amer/