How artificial intelligence has found its way into our lives - Part 1

Jürgen Schmidt: 2024 was an incredible year for AI developments. For the first time, there were Nobel Prizes for research in this field, and then two. Researchers from DeepMind, the Google subsidiary, deciphered the folding of proteins and thus solved one of the last great mysteries of the natural sciences. The second prize was awarded to John Hopfield and Geoffrey Hinton, who developed the Hopfield network back in 1982, which is still used as a neural network in many AI systems today. But let's look even further back.

Alan Turing published his famous Turing test in 1950. In his seminal work "Computing Machinery and Intelligence", he dealt with the question of whether machines can think. The Turing test, which he described in this book, is probably familiar to many. However, Alan Turing himself described the test as an "imitation game". He was therefore aware that it was essentially about imitating intelligence.

The "Summer of Artificial Intelligence" took place in 1956. At this conference at Dartmouth University, leading scientists came together for a six-week workshop. This event is generally regarded as the birth of active research into artificial intelligence. At the very least, the term "artificial intelligence" was coined here. In 1966, Joseph Weizenbaum published ELIZA, the first chatbot that could interact with humans. In the history of computer research, there have always been periods in which a great deal has been created, but also times in which research has been relatively dormant. We like to talk about AI summers and AI winters. In the 1990s, very important discoveries were made and published, such as the LSTM (Long-Short Term Memory) or backpropagation as a concept.

For me, one of the greatest intellectual successes in AI research is still the development of DeepMind and later AlphaGO-Zero, the computer that beat Lee Sedol in the board game GO. The fascinating thing about this game, however, is that it is not a completely logical game like chess, for example. If you ask professional GO players about their decisions, you often get answers like "it felt good". South Korea's Lee Sedol was the reigning world champion in this highly complex game in 2016. Sedol said of this challenge in an interview shortly before the game: "I'm not playing for myself or for South Korea, this time I'm playing to show the superiority of humanity". Sedol lost the game 1:4.

The current summer of AI development was triggered by openAI in 2022. The principle of GPT was developed in 2017. In 2018, openAI published the first version of it. It was possible to interact with the large language model via APIs. There was no public version. The second version was published in 2019. It is now on display at the Ars Electronica in Linz. The quality of the language and understanding was, of course, far below what we have before us today. OpenAI essentially put a "chat" in front of GPT's model, making the system accessible. This made an AI model easily accessible to the general public for the first time in history and also triggered the biggest hype we have seen in technology development to date.

Jürgen Schmidt: Let's take a look at the company NVIDIA. The company has been producing graphics chips (GPUs) and graphics cards for many years. With the current AI hype, NVIDIA has become the most valuable company in the world, as GPUs are particularly suitable for AI calculations. Here we have the main reason why these technologies are suddenly becoming so strong. We have achieved the necessary computing power. Through enormously efficient chips, cloud computing and the resulting endless GPU power, endless storage and, of course, an endless data stream. All of this has meant that we have put ourselves in a position to exploit these possibilities. What sounds a little banal is actually the main reason why it now works. It's the computing power. The potential opportunities have, of course, also triggered the flow of capital, especially in the USA. There are incredible reserves of capital being burned here, and the companies are actually still only making losses with their offerings.

Now we have the computing power, the necessary capital and, of course, the knowledge of how such models work is spreading rapidly. There are countless scientific papers on the subject, and others can build on the results of their research. This also leads to a reduction in costs. A great many companies have now also published on the subject and much of it is now available on the market as open source. All of this makes developments faster and faster.

Nevertheless, we urgently need certain quality standards and, due to the power that these tools can develop, we also need democratic rules on how and where AI can be used. We can see this very clearly in the Chinese model deepseek. The model is incredibly powerful, but it is also like a livestream into the Chinese censorship authorities. It is questionable how this will help us in further developments.

Find out more about the challenges and opportunities of AI in part 2 of the interview with Jürgen Schmidt.