Who is more intelligent: Data or C-3PO?

Data, of course, because he is the more complex figure. As an android, he tries so hard to become like us humans, even at the risk of no longer being able to function properly. This idea allows for a lot of slapstick in the Star Trek series. In reality, we are actually trying the opposite today, to overcome human limitations with the help of artificial intelligence. I wouldn't set the bar too high here for potential insights, but the figure of Data does illustrate that the development of AI is closely linked to fundamental questions about human-technology relationship.

When will there be androids available with their own consciousness that are indistinguishable from humans?

If you look at the problems facing AI today, it is hard to imagine we could develop software or a robot with its own consciousness in the near future. However, this prospect is inspiring many visions, of course. There are very strong arguments that what we understand as consciousness, which is bound to human life, can never be reproduced artificially using hardware and software.

This question may be pointing in the wrong direction. If, in the future, machines are really able to perform very complex tasks, which require some kind of understanding of the world, then it might appear to us as if they had consciousness, but in reality it would only be a very clever simulation of consciousness. I think “simulation” is a key term to understanding what artificial intelligence can do and will do even better in future. The important thing here is that we humans keep the upper hand and don't confuse simulations with human behavior.

Can you remember when you first heard or read about artificial intelligence (AI)?

About ten years ago, there was a lot of discussion about expert systems, neural networks and the “semantic web”. All of that was already loosely linked to artificial intelligence back then. However, it was not clear to me at all what rule-based systems have to do with intelligence.

I have to admit that my enthusiasm for the topic was only triggered when I saw the surprisingly good results of AI-based translation programs and the astonishing results of automated text analyses. IBM's “Project Debater” program, for example, can independently extract the arguments for and against previously defined questions from texts it has been fed. Applications like this were so amazing that I wondered: how does the program do that? How can the results be so much better than they were just a few years ago?

Interestingly, many of those who do research on AI are themselves surprised by the practical success of their programs. Because today's computers are much more powerful and have much better training data, the neural networks function very well in concrete implementations. Often, however, it is not clear why they do so – meaning practical success precedes the theory. This shows that we are still in an early phase of development, which makes the whole thing even more exciting.

What are the main areas of AI research at Fraunhofer ISI and what is special about them?

We set up the interdisciplinary research field of “Artificial Intelligence” in spring 2020 at Fraunhofer ISI, in order to coordinate the different AI activities already taking place in the institute. For us, it is very important to combine and pool the different perspectives and to act as one unit to those outside the institute.

Our umbrella topic is “AI from the perspective of innovation”. We have defined nine specific questions under this ranging from the question of AI-related changes to the healthcare system, through AI contributions to the energy transition, and what will come after AI. All these questions and the associated projects can be found on our website.

What all these questions have in common is that we always look at artificial intelligence in its application context and consider the use-specific, economic and social implications. We are less concerned with technical details than its application potentials, the conditions and consequences of its use in different areas. Precisely because this topic is so complex, there is the need here for education and consultation.

What questions and topics are you particularly interested in?

It is important to me that we don't allow ourselves to be blended by unfounded promises and marketing gags or even by visions of the future that go far beyond what we need. I think it important that we look very closely at things.

On the other hand, we shouldn't be naive and ignore the diverse advances made in AI. Many think that AI programs only process programmed instructions and don't believe that AI could “learn” let alone “think” in a human sense.

I believe that deep learning with its foresight abilities really does represent a new dimension that should not be underestimated with regard to its innovation potential - but also its social impacts. Here it is important that the areas of development and application are more closely coordinated. For this specific cooperation, we have developed participatory methods at Fraunhofer ISI that are still used too little.

How do you use AI in your research and in your daily life?

I don't write anything myself anymore, instead my research reports are automatically created by GPT-3! The program churns out texts like I would write them on its own and I can lean back and relax ;-).

No seriously: We use AI methods at the institute to empirically analyze innovation processes. For example, how you can analyze and predict company patenting behavior in specific sectors. In the area of Foresight, too, we use Big Data methods to discover what the emerging technology trends are. This makes our research more challenging, because we deal with large amounts of unstructured data.

On the other hand, AI should also be used to simplify everyday routines. I have heard that there is an AI program that looks at my email behavior and registers when I do the same thing over and over again, when I invite the same people to a meeting or repeatedly use the same phrases. Next time, AI writes the email automatically and I only have to send it. I think I should take a closer look at that …