Timetable

×

Lecture 2 - Alignment with what values?

Kalle Grill

Abstract

When we talk of value alignment, what do we mean? We want to avoid technology that is obviously destructive and harmful to all of us. But aside from that, what interests and values should technology be aligned with exactly? And whose values?

×

Lecture 4 - Learning with multiple representations

Axel Ngonga & Barbara Hammer

Abstract

Recent machine learning models excelled excel for complex tasks. For example, AlphaFold, targets the prediction of protein 3D structures based in the sequence, resulting in the 2024 Nobel prize in chemistry. One key component of this network is its reliance on a combination of different representations, including structural components, homological information, and structural operations implementing equivariance. More generally, machine learning methods often go beyond classical representations such as tabular data, vectorial representations, or real-valued cost functions, and they harvest the advantage which comes with more complex representations. Learning wirh multiple representations (LMR) refers to ML methods that leverage several representations of the data and models or several formalizations of the learning task simultaneously, orchestrate them in a coordinated fashion, and use them in a synergistic way to solve a single problem. Taking advantage of the complementarity, the redundancy, and specific characteristics of different representations, LMR seeks to improve performance in comparison to methods operating on a single representation only. Typical examples of LMR include the combination of numerical and symbolic formalisms, the representation of data and models on different levels of abstraction, and the combination of different types of supervision of the learner.

Within the tutorial, we will explore two different examples of this framework: We will explore dimensionality reduction methodologies in machine learning and how this can be extended to incorporate specific auxiliary knowledge. This enables an inspection of the suitability of a chosen representation for a given task, and it can be extended towards a visualization of the function implemented by supervised deep networks. Moreover, it allows an inspection of the effect of finetuning on foundational deep models such as language embedding.

Furthermore, we will explore the impact of using multiple representations on concept learning. After a brief presentation of classical approaches based on reasoning in description logics, we begin the discussion of multiple representations by introducing concept retrieval via SPARQL. In particular, we show that concept retrieval can be reduced to querying graph databases when full forward chaining has been performed. We then give some brief insights into using embeddings for rapid but approximate reasoning. To this end, we show how ideas based on conjunctive queries in embeddings spaces can be exploited to develop a full approximate reasoner for expressive description logics. Insights into the tradeoff between explainability and scalability complete this section of the presentation.

×

Lecture 5 - Between legal and (non)responsible AI

Markus Naarttijärvi

Abstract

There have long been calls for a better regulation of emerging AI and machine learning technologies. Following the development of ethical frameworks on the European level we can now begin to see the concrete legal rules and principles that will govern the design, implementation, and use of AI in Europe. While the discussions on this legal framework has been dominated by the new AI act, important principles have also been emerging from the fundamental rights context. In this lecture, we will begin to trace the outlines of what the law considers to be responsible and non-responsible AI, but also where the stricter legal limits can begin to be drawn and where value conflicts and future issues may arise.

×

Lecture 6.1 - Explainable AI: What is it and what can it do?

Kary Främling

Abstract

Explainable AI (XAI) is a domain that develops methods that make is possibly to explain, or at least justify, recommendations and actions of AI systems in similar ways as humans do. However, current XAI methods tend to produce non-interactive results that are mainly (or only) understandable to the AI engineers themselves. Social XAI (sXAI) is proposed as a name for XAI functionality that mimics how humans explain and justify their actions and opinions at least to some degree. sXAI methods should be interactive and adapt their explanations to the explainee’s background knowledge, interests and preferences in how to receive explanations, as well as the pace of interaction. The presentation shows how sXAI can be implemented using the Contextual Importance and Utility (CIU) method – and gives some reasons for why we probably won’t see sXAI happening anytime soon.

×

Lecture 6.2 - Learning -as humans- about responsibility in Machine Learning

Francien Dechesne

Bio

Francien Dechesne is Associate Professor of Ethics and Digital Technologies at the Center for Law and Digital Technologies (eLaw) of the Leiden Law School. She is Programme Director of the Advanced Master in Law & Digital Technologies, Director of Education at eLaw and initiator and overall coordinator of the university wide Minor Programme on Artificial Intelligence and Society (developed in the context of Leiden's SAILS network on AI). With backgrounds in computer science, mathematics and philosophy, her research and education center around interdisciplinary perspectives on Artificial Intelligence, and its interactions with society: demystification of the technology, identifying and explaining societal impacts (to those affected as well as those involved in the development of technology and regulation), and different strategies towards responsible AI.

Abstract

This lecture explores the multifaceted notions of fairness and justice across ethics, technology, politics, and law, emphasizing their intersections and tensions in computational sciences and machine learning (ML). We look into differences in the conceptualizations of fairness and justice in different disciplines, and how they may implicitly or explicitly shape socio-technical systems – in particular ML systems. We will discuss the inherent tensions between idiosyncratic societal contexts, stable abstract ideals and their implementation into algorithmic frameworks. We connect this to a reflection on ML as a form of social experimentation beyond the lab, raising questions about accountability, power, and responsibilities for people involved in theory and research, development and deployment of ML systems.

×

Lecture 7.2 - Fairness

Lili Jiang

Abstract

Fairness is a critical consideration in responsible AI decision-making systems. This presentation introduces fairness concepts, exploring definitions, metrics, and case studies on bias throughout the AI system’s life cycle—from input data and modeling to output. We will discuss common fairness criteria, highlighting their trade-offs and real-world implications. Furthermore, we will introduce project AEQUITAS, which aims to develop a framework to identify and mitigate various manifestations of bias and unfairness in AI.

×

Lecture 8 - Problems and Pitfalls with Deep Learning

Tommy Löfstedt

Abstract

Machine learning has undergone a remarkable development in the last few years. This is especially clear in generative machine learning, where new and improved text and image generation models are being released virtually continuously. Many companies and also scientists are promising revolutionary solutions to virtually every concievable problem. But are these promises actually feasible? Turns out that they often are impossible to deliver on, have all kinds of biases, are racist/misogynist/xenophobic, and are even failing misserably---yet, the promises are still bought at ever growing rates. Much of the problem lies with the hype around "AI", that people are blinded by never ending remarkable new results and assume that success in one area must necessarily translate to other ares. Noone wants to slow down, risk to miss out, or seem like they are not up to date with the latest trends. These are societal problems but also scientific problems, since the hype is also ever present in research. In this talk we will discuss the hype around "AI", what we can do to spot false claims, and some thoughts on rigor in data analysis through machine learning.