Speakers
Zahed Amanullah
Understanding counterspeech and the impact of AI
Abstract: The use of counterspeech as a strategy against targeted hate, extremist messaging, and disinformation has been widely debated and tested, both online and offline. But how does the rapid development of AI impact the fight against online abuse? We’ll review the evolution of counterspeech strategy and how AI can frustrate it, provide opportunities to optimize it - or both.
Bio: Zahed Amanullah is a Senior Fellow at the Institute for Strategic Dialogue (ISD) and helps drive ISD’s civil society engagement with a view toward countering hate, extremism, and disinformation. His recent work included capacity building for US community organisations during the 2020 elections, for Kenyan organisations during their 2017 presidential elections, and for UK and EU community organisations in partnership with Google.org for the £1m Innovation Fund to Counter Hate and Extremism in the UK in 2018 and the €10m Impact Challenge on Safety in 2019. He previously served as Director of the Concordia Forum, a global network of leaders from Muslim backgrounds that has convened strategic retreats alternately in Europe and North America since 2009.
Ehud Reiter
Generated texts should not worsen users’ emotional state
Abstract: Computer-generated texts should not worsen users’ emotional state. For example a user should not start crying after reading a text which communicates bad news, and should not feel inadequate because she is unable to follow sensible-sounding suggestions. I'll discuss a number of projects where such problems (including crying users) have emerged as a major concern, and indeed a barrier to deploying the system in production usage. I'll also look at some ways of detecting and dealing with such problems, all of which unfortunately have their own limitations. I'll end by speculating about how a computer system could try to “counter” such problems in online human dialogues.
Bio: Ehud Reiter is a Professor of Computing Science at the University of Aberdeen and also Chief Scientist of Arria NLG (which evolved from a spinout company he co-founded in 2009). His research focuses on Natural Language Generation and on evaluation; he has a special interest in medical applications of AI and NLG. He has a Google Scholar H-index of 56 and writes a widely read blog on NLG (ehudreiter.com), which often discusses real-world issues in using NLG and LLM technology. In 2022 he was awarded the INLG Test of Time award and also a NAACL award for “Best paper on human-centered NLP special theme”.