AI-Pioneer Prof. Hochreiter Presented with the "Neural Networks Pioneer Award"

Sepp Hochreiter, head of the LIT AI Lab, was presented with the "2021 IEEE Computational Intelligence Society Neural Networks Pioneer Award".

Sepp Hochreiter

Artificial Intelligence. Deep Learning. Digitalization, Industry 4.0 - buzzwords that show just how much Artificial Intelligence continues to shape our world. Over the years, pioneers like Sepp Hochreiter have been making it all possible. The professor for machine learning and head of the LIT AI Lab at the JKU has worked continuously on developing long short-term memory (LSTM), helping to make modern AI possible. He has received numerous awards and accolades in recognition of his work and recently, he was presented with the 2021 IEEE CIS Neural Networks Pioneer Award. This AI research award is surpassed in importance only by the Turing Award.

As there is no Nobel Prize for Computer Sciences, the Turing Award is the most comparable award. Almost everyone who wins the Turing Award was presented with the IEEE CIS Neural Networks Pioneer Award beforehand. Do you think you are next in line for the Turing Award?
Sepp Hochreiter: (laughs) Well, this actually happened to my predecessor. While awards are not particularly important to me, naturally I was pleased to win this award. I developed LSTM over 20 years ago, but now I’m focusing on other projects.

Such as, for example?
Sepp Hochreiter: We are working on a number of things, such as Modern Hopfield Networks that can improve Deep Learning methods. We are also conducting research on Few Shot Learning in which the AI only needs a few examples in order to adjust to a new customer. Up to now, when it comes to new customers, systems have had to more or less start from scratch and in this regard, our system is much more efficient. We’re working on a few other topics as well.

You have said repeatedly that when it comes to AI, Europe lags far behind countries such as the US and China. The European AI research network ELLIS is an attempt to turn this around. Have we been able to catch up a bit?
Sepp Hochreiter: When it comes to research, we are doing well, especially here at Johannes Kepler University Linz. Here at the JKU we have developed a certain technology that has just been sold to Amazon. But that's the problem: We’re not lagging behind in research – it’s more in the area of applied practice. The Americans and Chinese are turning their results around faster. But I have a feeling that many European companies, such as AUDI, are becoming more open-minded. There are also more start-up companies in this area.

Information tends to be mixed and on one hand, there are articles saying AI can't tell the difference between a dog and a cat in a photo. Other articles warn us of some kind of superintelligence that will replace humans. Those are two extremes - where do we really stand?
Sepp Hochreiter: Hollywood can often exaggerate when it comes to the topic of AI and movies like Terminator or Matrix need that degree of suspense. In reality, however, we are doing well in terms of AI technology in niches. However, even in areas where it works well, the process needs to be there. When it comes to image recognition, the photo has to be clear in order for the AI to respond correctly. But so far, AI only consists of mathematical formulas. This means things don’t go wrong until once something in the environment changes.

Why is that?
Sepp Hochreiter: AI simply doesn’t understanding the world the way humans do. Early on, a child learns how to distinguish objects from non-objects, meaning the child understands what is part of an object and what is not. A child may not know the word cat, but he or she knows what makes it a cat and what doesn’t. AI has a different way of learning. AI can recognize: This is a cat. If the tail is missing, the AI no longer recognizes that it’s a cat but child knows it’s still a cat.

There is, however, a kind of AI that can independently book a table at a restaurant, for example.
Sepp Hochreiter: Yes, the AI has been given so much information that it is able pick out the right words. When it comes to self-driving vehicles, we see this, too. If a plastic barrel blows out on to the street, a person can decide very quickly whether or not to hit the barrel or hit the cars parked along on the side of the road. A person can do something like this even if he/she never did it before. AI lacks human understanding in regard to the way things work in the world and cannot decide if other cars or trash are worth more. If you study it closely, there is a vast difference between humans and AI.