How did you first begin researching artificial intelligence (AI)?
The first time I felt we needed to do something about intelligent machines was when I moved out from my hometown to study computer science in Bonn, Germany. I suddenly was far away from my rock ‘n’ roll band, where I used to play the guitar, and found myself alone strumming the guitar in my small student apartment. So I bought myself equipment—synthesizer, drum computer, etc.—and just played one instrument at a time. However, these were the 80s and the drum computer was a robot. It played the beats so accurately that the music sounded sterile. I actually stumbled over a problem that I only began to understand much later.
Actually, I became aware of it during my studies at the University of Bonn, when my theoretical computer science professor explained about two kinds of problems in mathematics: those that are computationally solvable and those that are not, at least not in finite time and space. He continued to explain about the Turing machine as a tool to solve problems of the first kind, and I began to wonder, What about the second type? I started reading Turing’s original papers. Interestingly enough he argued that the phenomenon of intelligence cannot be programmed into a machine but must be acquired through learning! Actually, he argued for robots, with embedded learning capabilities, to iteratively achieve higher levels of intelligence.
I started to focus on neuroscience, machine learning, and robotics and did my PhD by applying my own version of a hierarchical reinforcement learning algorithm to a six-legged walking machine that I had built myself.
What would you say are the biggest challenges the AI research field is currently facing?
I see the integration problem to be the gamechanger. If we will be able to integrate the many different partial solutions that we already have—in the areas of speech recognition, vision and image understanding, optimization, control, etc.—into a single yet scalable framework, we will be able to reach the next level of intelligence in technical systems. These systems will then be able to create meta-knowledge on the basis of several input modalities using several AI techniques, eventually on several levels of a hierarchical knowledge generation architecture. Put this into a system that we could call a ‘never turn off’ system, then we are getting one step closer to Turing’s idea about the incremental acquisition of intelligence in what he called ‘intelligent machinery.’
From my point of view, these frameworks and integrated solutions must be inside a physical machine (you may call it a robot), as the ultimate challenge to increase and merge knowledge on higher levels comes from its performance in the (so far not simulatable—until we have Quantum computers) real world.
How does the open access nature of this journal contribute to the progress of AI research?
Open access is simply the best way to spread the news!
On the consumer side, it excludes too many people if you require money to read research results. The problem actually is that we exclude minds—maybe some of them brilliant minds—that would come up with concepts and ideas that push the research forward.
More and more funding agencies heard the bell and are allowing for publication fees to be put into the research proposal. This is a very positive development and it will allow us to publish more in open access journals.
What are your plans for the journal’s first year? What new projects are you excited to start with?
We are currently in the age of using AI techniques in nearly all aspects of human life. The results of these applications are sometimes overwhelming and I believe that this can be a major draw back for AI-Science. If you are too successful it is difficult to question yourself. This is what this journal wants to emphasize: Question your achievements. Not to say you should not do this, but instead: In order to say what have I learned from this application, what are the pros and cons of this application, and finally does this application give me new ideas or raise new questions about or for basic research that I need to address? Obviously, we do not just address and invite some top-level experts of the field. We want to hear perspectives on the above-mentioned questions from those at all career levels.
Frank Kirchner, Editor-in-Chief of AI Perspectives, is Professor of Mathematics, Computer Science, and Production Engineering at the University of Bremen and Director of the German Research Center for Artificial Intelligence.