Stephan Raaijmakers appointed Professor by special appointment of Communicative AI (Artificial Intelligence)
Since November 2019 Stephan Raaijmakers is Professor by special appointment of Communicative AI (Artificial Intelligence) at LUCL. Raaijmakers investigates how linguistics can help make Artificial Intelligence smarter.
Artificial Intelligence in complex situations
‘I hope that communicative AI will contribute to building trust in AI, and thus to a successful symbiosis between machines and humans. On the one hand, we hear that AI surpasses humans in areas such as medical science (diagnosis), gaming (chess, GO, poker) or speech recognition. However, on the other hand, we still see little AI in situations where people make complex decisions. Doctors do not blindly rely on a deep learning system - complex networks of which AI nowadays largely consists - for cancer diagnosis if they do not understand it,' Raaijmakers explains.
Insight into deep learning
Gaining more insight into deep learning systems and offering explanations to users is an important subject of the chair. According to Raaijmakers, teaching communicative behaviour to these types of systems, such as dialogue, is crucial. ‘These networks can learn from sample data, but are not very communicative. It is difficult to find out how a deep learning system comes to its conclusions, and on what sample data these conclusions are based.’ Raaijmakers would also like to investigate how deep learning and linguistics can benefit from each other, because according to him, the link between modern AI and linguistics would provide new and fruitful insights for both areas.
Value for society
The chair is made possible in part by his current employer TNO. TNO's mission is to test academic knowledge in practice, to make it applicable and to further develop it for society. Raaijmakers is currently working at TNO on various projects on communicative AI, such as a medical application in which a deep learning system is trained to recognise malaria infections in blood. He is also busy completing his book Deep Learning for Natural Language Processing (Publisher: Manning Publications), a book full of sample code for anyone who wants to analyse language with the aid of deep learning.
‘AI isn't easy, but it is not overly difficult either’
‘I look forward to working with LUCL, LIACS and Digital Humanities. I hope to learn from the neuro-, psycho- and other linguists within LUCL. And I look forward to building an active knowledge network.’ Furthermore, Raaijmakers finds it very important to point out to literary students the opportunities that AI offers them. ‘As literary students, you have valuable and relevant knowledge for the new generation of AI systems. AI is not simple, but it is not overly difficult either. I hope to be able to get that message across in my lectures and to get students excited about this profession.’