Prof Euro Beinat of the University of Salzburg recently delivered a public lecture at Stellenbosch University that gives more weight and direction to the artificial intelligence (AI) debate. An Honorary Professor in the Department of Economics, he discussed artificial intelligence from the current state of the AI domain to what its future looks like.
Beinat holds an MsC in Computer Science and a PhD in Economics. He teaches GeoInformatics and Data Science at Salzburg University in Austria, and is global head for AI and Data Science at Naspers and Prosus, one of the largest internet companies and investors in the world. Beinat has created and led several data science software product operations for corporations. His experience ranges from fundamental research to business leadership. He also serves on the board of several start-ups and has co-founded Data Science for Social Good, a non-profit organisation dedicated to promoting data science with a social impact.
Beinat titled his lecture, “From academic breakthroughs to mainstream utilisation: observations and outlook after ten years of artificial intelligence (AI) adoption." He started off by elaborating on the current state of the AI industry.
“AI is everywhere and is becoming more industrialised. It is applied in essentially every industry, but not in the same manner. Areas like consumer technology and finance are places where we can see it almost everywhere, and other places a lot less. Publications in the AI domain continue to increase and private investments continue to grow. We also see that AI has become more affordable and better performing. Costs have decreased by 63%, training times have improved 95% (a lot faster) in the last few years and the accuracy has also increased.
“We have also seen over time that models get larger as large models tend to do better than smaller models. In the past, working with language was complicated, but from 2018 we started to get very large models that have been trained in the entire corpus of Wikipedia and on all the books that have been digitised. Training these large models costs a lot of money but it is another example where technology skills, the ability to work with hardware at a very high level of sophistication and funding make a difference."
Beinat then gave an overview of the lessons learnt over the last decade.
“The first thing is that the current AI is fit for general purpose. There isn't really any particular domain where it cannot be used; one is only limited by whether one has the data or the talent. We also figured out how to make it work. Ten years ago it was a great struggle to get the information you required, as there was not enough data and the algorithms were not performing well. As a community of practitioners we have figured out how it works. So now if we have a problem, we know which data we need to use, we know more or less what effect it is going to have, we know whether it is going to work or not. That is important because we now know how to control these things at scale.
“We also know that an organisation that wants to become an industry leader, has to hire the right people with the right skills. Essentially, you have to ensure that a certain degree of knowledge about the tools and the science must be present everywhere in the organisation. We still have a big shortage of talent."
Looking into the future, Beinat predicted, among other things, that AI models will continue to grow in size and increase our ability to work with language; robotics based on AI will become viable; AI will be used more and more to crack down on hard problems and ensure better forecasting, and will also serve as a tool for invention.
“Ethical and responsible AI will be at the forefront of AI development," he concluded.
- Photo (supplied): Prof Euro Beinat