Human or Machine: Reflections on Turing-Inspired Testing for the Everyday 🔗
In his seminal paper “Computing Machinery and Intelligence”, Alan Turing introduced the “imitation game” as part of exploring the concept of machine intelligence. The Turing Test has since been the subject of much analysis, debate, refinement and extension. Here we sidestep the question of whether a particular machine can be labeled intelligent, or can be said to match human capabilities in a given context. Instead, but inspired by Turing, we draw attention to the seemingly simpler challenge of determining whether one is interacting with a human or with a machine, in the context of everyday life.
We are interested in reflecting upon the importance of this Human-or-Machine question and the use one may make of a
reliable answer thereto.
Whereas Turing’s original test is widely considered to be more of a thought experiment, the Human-or-Machine question
as discussed here has obvious practical significance.
And while the jury is still not in regarding the possibility of machines that can mimic human behavior with high
fidelity in everyday contexts, we argue that near-term exploration of the issues raised here can contribute to
development methods for computerized systems, and may also improve our understanding of human behavior in general.
Of course, future strict regulatory restrictions on the use of AI systems might affect some of the issues we raise here.
However, we feel that the regulators themselves should be fully aware of them too.
(This is joint work with Asaf Marron)
Deep Neural Networks, Explanations, and Rationality 🔗
“Rationality” is the principle that humans make decisions on the basis of step-by-step (algorithmic) reasoning using systematic rules of logic. An ideal “explanation” for a decision is a chronicle of the steps used to arrive at the decision. Herb Simon’s “bounded rationality” is the observation that the ability of a human brain to handle algorithmic complexity and data is limited. As a consequence, human decision-making in complex cases mixes some rationality with a great deal of intuition, relying more on Daniel Kahneman’s “System 1” than “System 2”.
A DNN-based AI, similarly, does not arrive at a decision through a rational process in this sense. An understanding of the mechanisms of the DNN yields little or no insight into any rational explanation for its decisions. The DNN is operating in a manner more like System 1 than System 2. Humans, however, are quite good at constructing post-facto rationalizations of their intuitive decisions.
If we demand rational explanations for AI decisions, engineers will inevitably develop AIs that are very effective at constructing such post-facto rationalizations. With their ability to handle vast amounts of data, the AIs will learn to build rationalizations using many more precedents than any human could, thereby constructing rationalizations for ANY decision that will become very hard to refute. The demand for explanations, therefore, could backfire, resulting in effectively ceding to the AIs much more power.
In this talk, I will discuss similarities and differences between human and AI decision making and will speculate on how, as a society, we might be able to proceed to leverage AIs in ways that benefit humans.
Technology and Democracy 🔗
U.S. society is in the throes of deep societal polarization that not only leads to political paralysis, but also threatens the very foundations of democracy. The phrase “The Disunited States of America” is often mentioned. Other countries are displaying similar polarization. How did we get here? What went wrong?
In this talk I argue that the current state of affairs is the results of the confluence of two tsunamis that have unfolded over the past 40 years. On one hand, there was the tsunami of technology - from the introduction of the IBM PC in 1981 to the current domination of public discourse by social media. On the other hand, there was a tsunami of neoliberal economic policies. I will argue that the combination of these two tsunamis led to both economic polarization and cognitive polarization.
Graph Neural Networks: Everything is Connected 🔗
Our world is highly rich in structure, composed of objects, their relations and hierarchies. Despite the ubiquity of graphs in our world, most modern machine learning methods fail to properly handle such rich structural representations. Recently, a universal class of neural networks emerged that can seamlessly operate on graph-structured data, summarized under the umbrella term Graph Neural Networks (GNNs).
In this task, we will introduce the concept of Graph Neural Networks and the general framework of neural message passing. We thoroughly analyze the expressive power of GNNs and show-case how they relate and generalize concepts of Convolutional Neural Networks and Transformers to arbitrarily structured data. In particular, we argue for the injection of structural and compositional inductive biases into deep learning models. Despite recent trends in neural networks regarding LLMs, such models manifest our understanding of a structured world, require less computational budget, and are easier to understand and explain.
Education and AI – Current Status, Opportunities and Challenges 🔗
Successful educational processes as well as educational outcomes are prerequisites for positive educational and life paths of individuals as well as the development of their societies and countries. Education - teaching and learning - is a complex process accompanied by multiple challenges. These include, for example: People begin a learning process with different preconditions. Educational outcomes in many countries are still closely associated with the social backgrounds of learners’ families. A substantial and due to the COVID pandemic increased proportion of learners do not achieve the minimum standard necessary in key skills such as reading. Digitization in the education system has not progressed that far in many cases. The possibilities of using AI in education have received a lot of attention recently, as there are fundamental opportunities but also challenges associated with it. It can be considered certain that AI will fundamentally transform education across educational phases, domains, and contexts. This transformation needs to be shaped and accompanied from the perspectives of different disciplines.
In the keynote, an overview of key current challenges in the context of education will be presented in the first part. In the second part, the potentials and challenges of AI in the areas of diagnostics and assessment, instruction and support will be addressed. The possibilities of advanced individualization, real-time responses, and the processing of large amounts of data are just some of the advantages for learners and teachers. At the same time, there is up-to-date a lack of well-founded studies on the benefits and risks for many possible areas of application of AI in the educational context. Further limitations concern infrastructure, didactic materials, and professional teacher training. Finally, open questions for future research and development are discussed.