AI and ELSE
Between February 4th and 8th Joint Research Centre hosted the HUMAINT Winter school on AI and its ethical, legal, social and economic (ELSE) impact. We had the chance to meet and discuss with more than 100 attendees from different countries and backgrounds: law and ethics, computer science, economics, robotics, social sciences. Policy makers, researchers, journalists, entrepreneurs had a chance to debate during the five days the latest advancements in AI and the impact on humans.
My winterschool experience: people coming from all over the globe, interdisciplinary and very talented researchers who are kick-starting the interest and exploration of ethical, legal and societal issues in AI. The most substantial […] takeaway from the HUMAINT school was […] that AI is increasingly permeating every aspect of our society […]. The principal pressure should be to design and regulate AI to be accountable, fair and transparent by means of echnical standards, ethical principles and professional codes of conduct. –Fernando Martínez Plumed
The program and the slides can be found online while shortly we will make available the videos for the presentations.
The first two days comprised talks which were meant to introduce the audience to several research topics in AI's ELSE impact. The winter school started on February 4th with a welcome note by Emilia Gómez, Virginia Dignum. Afterwards, the head of the JRC Digital Economy Unit, Alessandro Annoni presented key points of the AI EU strategy. The participants were introduced to the general field of AI, its passed development and current state by José H. Orallo. The speaker was not short in pointing out the field's current challenges from misaligned goals between research, industry and politics to the lack of meaningful evaluation criteria for AI advancements.
For me and my research one of the most valuable aspects of the HUMAINT Winterschool was the diversity of approaches in the room. There were social scientist, journalists, machine learning scientist, and behavioral scientist, bringing a rich set of opinions, backgrounds and knowledge to a topic which is as broad and contested as AI. I specifically appreciated how Virginia critically questioned the definition of fairness as a metric. Is it a techno fix or is fairness about equality and equity in society. –Fieke Jansen
This was followed by two talks on AI ethics: about designing AI systems and ethics (Virginia Dignum) and the standards necessary to build ethical AI systems (Ansgar Koene). The legal implications of AI were discussed by Christopher Markou with thought-provoking statements on the tensions between jurists' and engineers' perspectives on AI in law. The first day finished with a talk by Nuria Oliver on the social impact of AI.
Robotics, Machine Learning and AI in Practice
On February 5th we started with a talk by Aurelie Clodic on robotics. HUMAINT researchers Emilia Gómez and Marius Mirion presented a general overview on machine learning, some music applications, and discussed problematic machine learning usages in different domains. This first part of the HUMAINT winter school culminated in a discussion panel on AI's ELSE impact with Gabriele Mazzini, Nuria Oliver, Esther Paniagua, Richard Benjamins, and Jonathan Penn, moderated by Virginia Dignum.
In the afternoon, several attendees had the chance to present their work. The session was split in two: a first part of applications of AI and a second part on AI ELSE research. It also incorporated some presentations by AI startups in Seville.
The best part of my experience was to be a participant in project work with human-robot interaction. […] Our group focused on the “What” and “How” when embedding Human Values on in autonomous systems. In the practical part we had a chance to develop an algorithm. […] On Friday we presented our first results and had a group discussion about this. –Alina Iarmina
On February 6th Pierre-Yves Oudeyer introduced us to his work on developmental autonomous learning. His vivid material illustrated the importance of curiosity in cognitive development. José H. Orallo's talk about AI evaluation posed two important questions:: "Are we measuring the right things in AI?", and "Are we measuring AI right?". Shortly after that Virginia Dignum adressed the question whether "AI artefacts can be designed to be verfiably ethical in order to advocate for a "design for values" in ethical AI.
AI Principles and Impact
Only a few people can envision and develop AI systems. This asymmetry in terms of effective capacities defines how powerful is each of the actors. Hence, most citizens cannot critically and independently oversee what’s going on in AI. If we seek for a higher democratic control in AI systems, we need to first break these power asymmetries. –Jordi Pons
February 7th started with a presentation of Jonathan Penn on the forgotten histories of AI, the way it transforms societal values and the way a society could transform AI. Frank Dignum continued with another interesting and interactive presentation on socio-cognitice systems and the importance on designing applications in a social context. The day ended with attendees presenting their own work.
Being an environmental and resource economist myself, […] what brought me to this workshop […] was my interest in how […] AI tools can be expected to change […] the way we think of marine resource management. During the workshop I had the opportunity to present a few words on my thoughts about ‘AI & Marine Resource Management’ and engage in discussions with a truly interdisciplinary group of scientist, […] on how algorithms can turn out to be ‘unfair’. –Melina Kourantidou
On February 8th Emilia Gómez and Vicky Charisi presented their work on the impact of AI on human behaviour, with a focus in human-robot interaction. They summarized HUMAINT's ongoing research and experiments on the impact of AI on child cognitive development. This was followed by two talks on the economic impact of AI. While Songül Tolan discussed AI's impact on the use of prediction in industry and the future of labour markets, Bertin Martens adressed critical issues in the emergence of Data markets through AI proliferation. Before lunch, Bob Sturm, very illustratively with his own accordion, provided an overview of current AI-based music generation systems and demonstrated his own model for automatic generation of Irish folk tunes. In his talk, we discussed about ethical issues that the usage and creation of an automatic "art generating" system pose.
The afternoons of February 6th, 7th, and 8th were dedicated to research projects. Attendees could join one of the five projects on: algorithmic literacy, AI systems and trust, analysis of indicators of AI projects, human-robot interaction, and fairness in machine learning. Each team had the chance to present their findings at the end of the winter school.