Author: Vanessa Hübner Edited by: Ruth Sang Jones
On March, 22nd 2018 SIU Frankfurt hosted another interesting event on “Applied AI: From AI Research to AI Technologies”.
The first speaker was Prof. Dr. Visvanathan Ramesh from Goethe University and the Frankfurt Institute for Advanced Studies (FIAS). As a systems engineer with hands-on experiences in academia and industry, Visvanathan explains that “AI is not magic, AI is systematic”.
- The first wave of AI in the 1950s involved handcrafted knowledge allowing reasoning on specifically defined problems but without the capability to learn and to abstract.
- This evolved to the second wave, whereby a statistical learning system could predict outcomes from learned and classified data but with a minimal ability to reason and to abstract .
- Now, we are in the third wave of AI that combines the strengths of the first and second waves and aims to make the system capable of explaining its decisions.
To address this objective and eventually, to automatize AI, Prof Ramesh 's research focuses on “Integrative AI”. He sees the future of AI supported by an integrative network of interdisciplinary research networks across fields like neuroscience, psychology ,computer science, business and education. AI thinking could be made possible by training and mentoring the systems with information from these research areas, which would work alongside platform networks that facilitate design automation tools.
The second speaker of the evening was Prof. Dr. Danko Nikolic who is also based at FIAS. He described how his research on the brain led him to the field of AI, a journey that has resulted in him creating a novel approach.
Prof. Nikolic introduced the theory of 'practopoiesis' – the principle that biological system intelligence is founded on hierarchical levels of adaptability, which is an entirely new way of thinking of the origin of the mind. He pointed out that most of the information that we need in order to be intelligent lies in our learning system, which is always adapting to environmental stimuli. Knowledge, which is rather static, is not the root of intelligence. He proposes that this new theory can be implemented to design AI. For a better understanding of practopoiesis, check out the video here:
The 'one-shot learning rule' for AI was also live demonstrated as it is based on the principles of practopoiesis. In contrast to current machine learning approaches which neccesitate vast quantities of data for task learning, 'one- shot learning' can allow use of much fewer data points. This was shown on https://robotsgomental.com/mr-character-demo/- a system was shown only three different imaginary symbols drawn by the user and it subsequently was able to mostly reliably classify a very similar character.
At the end of his talk, he shortly drew attention to the AI-Kindergarten , which promotes the idea that machines can be trained through human feedback to create their own algorithms.
As always, SIU members were invited after the talk to meet the speakers over wine and finger foods.