Image

Welcome to a series of inspiring talks on artificial intelligence by Umea University researchers, as well as internationally acclaimed experts.

Participating in #frAiday is your opportunity to share your experience and knowledge about artificial intelligence, learn more about the field, discuss a wide range of perspectives on AI, and meet new people.

Time: Fridays, 12:15 – 13:00
Place: Online via Zoom, due to new regional restrictions. The event is open to everyone who is interested in AI.

Program

23 April 
Extended and Autonomous Agency

Kalle Grill, Associate Professor at Department of Historical, Philosophical and Religious Studies, Umeå university. 

Abstract 

As AI systems become more potent, we are more likely to outsource cognitive functions such as information processing or even deliberation. Recommender systems perform such functions already and more advance personal assistants could do more of it in the future. Outsourcing tasks central to our agency may be seen to threaten our autonomy.

However, we might instead adopt the perspective that we and the AI systems we use together to form an extended agent. This idea is analogous to, perhaps an extension of, the idea of an extended mind due to Andy Clark and David Chalmers (1998).

It is no loss to the autonomy of an agent that some part of it is not autonomous. In fact, autonomous parts may be a threat to agent autonomy. Hence, a non-autonomous human part of an extended agent does not make that agent less autonomous. On the other hand, the autonomy of extended agents can be threatened in other ways. Taking on the perspective of extended agency, without necessarily endorsing it, can help us identify other ethical concerns.

30 April 
Reasoning - From Humans to Machines

Timotheus Kampik, doctoral student at the Department of Computing Science, Umeå university

Abstract 

The seminar will provide an overview of the intersection of research on human reasoning and decision-making, as most notably conducted in Microeconomics and Behavioral Psychology, and research on machine reasoning. In particular, the seminar will illustrate how concepts and models of human decision-making relate to fundamental principles of automated reasoning, explainable AI, and combinations of symbolic reasoning and learning. To demonstrate the real-world applicability of the presented perspectives, the seminar will relate to practical tools and methods, such as industry-scale decision automation using the Decision Model and Notation (DMN) standard.

7 May
Reactive Planning - A bio-Inspired AI approach to robot action selection

Rob Wortham, Lecturer in Robotics & Autonomous Systems, Director of Studies, MSc Robotics & Autonomous Systems, Dept of Electronic & Electrical Engineering, University of Bath, UK

Abstract

One of the fundamental challenges of all embodied and virtual agent based systems is the challenge of action selection: Given a set of desired goals and a set of sensory inputs, how does an agent decide what to do next? In this talk I will describe how we can take inspiration from observations of natural intelligence to develop a structure and methodology for the development of reactive action selection systems that have been shown to produce usefully intelligent virtual agents and robots. This ‘reactive planning’ approach also has the incidental benefit of lending itself strongly to the development of transparent AI systems; an important consideration as we seek to produce ethically aligned autonomous systems.

14 May
The AI behind advertising

Johanna Björklund, Associate professor, Department of Computing Science, Umeå University

Abstract

Digital advertising is a key source of financing for online publicists. By attracting readers to their sites, they can make an income by displaying advertisements on behalf of others. The advertising space is sold via fully automated auctions that run their course within the blink of an eye. The bids are based on a range of factors, such as the popular standing of the digital publication, the size and positioning of the ad, and even the time of day. At present, bids are primarily based on information about the reader who will be seeing the ad. This is known as personalised advertising and has problems related to privacy and fairness.

An emergent alternative to personalised advertising is to base bids on the surrounding media context. This practice is known as contextual advertising and avoids the use of personal data. On the downside, contextual advertising can blur the line between publisher content and advertisements in a negative way. It can also recreate problems linked to personalised advertising by making assumptions about the reader based on the content that he or she is looking at. The combination of personalised and contextual advertisement is especially loaded, as the advertiser will then know both who is requesting a webpage, but also what the webpage is about. Not only does this open for the most severe privacy issues, but it can also be used to manipulate democratic process such as elections.

In this talk, we survey how AI is currently being used in online advertising, and then look closer at personalised and contextual advertising. We discuss how advertisers reason when they set up their campaigns, what changes when the AI systems that execute the campaigns become more capable, and what could happen when these techniques are applied to political opinion formation. On a more positive note, we will also talk about how the negative effects can mitigated, and how consumers can sometimes use the workings of the advertising AIs to their own advantage.