Viktor Kewenig

Researcher at Microsoft, PhD in Cognitive Neuroscience (Leverhulme Ecoological Brain DTP, UCL)

Currently, I am a researcher at Microsoft Research Cambridge, thinking about how we can evaluate successful collaboration between humans and agentic AI systems - and whether we can use that information to produce better agents.

In my PhD thesis, I used multimodal computational and ecologically valid research methods to investigate alignment principles between AI and human cognition

Between 2022 and 2024, I already collaborated with Microsoft Research Cambridge to find out more about how generative AI impacts human cognition in knowledge work and education.

My undergraduate studies were in logic and philosophy of science and mind at Cambridge, I transitioned to cognitive neuroscience of language comprehension during my MSc at UCL.

Music brings an important balance to my life - I curate a podcast series called "5918mins" and sometimes mix records myself under an alias called "No Frills". The best finds are uploaded to my YouTube channel.

For a comprehensive overview of my academic and professional background, you can have a look at my CV.

Current Research

Metacognitive Interventions for Efficient Prompting of Generative AI Systems

Funded by the "Microsoft Accelerate Foundation Models Research Fund". I am building a "Metabot", a metacognitive scaffolding interface for improving generative AI interaction modes: https://metabot.azurewebsites.net/

This project focuses on developing interfaces that enhance user metacognition when interacting with generative AI systems, potentially improving the efficiency and effectiveness of AI prompting.
Evolving Norms Around the Use of Generative AI in Higher Education

2024 - Co-led qualitative study and write-up, in collaboration with Microsoft Research

This study explores the changing norms and practices surrounding the use of generative AI in higher education settings.
Effect of Large Language Models vs. Note Taking on Memory and Comprehension in High-School Students

2024 - Co-leading data collection, write-up and analysis of a large scale study in collaboration with Cambridge University Assessment, Microsoft Research Europe, and Microsoft Research New York

This research compares the impact of using large language models versus traditional note-taking methods on high-school students' memory and comprehension.
Fine-tuning CLIP for Sentiment Analysis

2024 - Stanford University, Natural Language Understanding (XCS224U), Online Course

This project involves adapting the CLIP (Contrastive Language-Image Pre-training) model for sentiment analysis tasks.
Encoding and Decoding Brain Activation During Naturalistic Story Listening with Unimodal and Multimodal Large Language Models: A Comparison

2024 - Analysis on high-performance computer cluster, using TensorFlow, PyTorch, Transformer, and customised LLaMA implementation

This study compares unimodal and multimodal large language models in their ability to encode and decode brain activation during naturalistic story listening.

Publications

The Metacognitive Demands and Opportunities of Generative AI

Tankelevitch, L.*, Kewenig, V.*, Simkute, A., Scott, E.A., Sarkar, A., Sellen, A., Rintel, S. (2024) - Published at Proceedings of CHI (best paper award)

This paper explores the metacognitive aspects of working with generative AI systems, highlighting both the challenges and opportunities in human-AI interaction.
Ironies of Generative AI: Understanding and mitigating productivity loss in human-AI interactions

Simkute, A., Tankelevitch, L., Kewenig, V., Scott, A.E., Sellen, A., Rintel, S. (2024) - Accepted at International Journal of Human–Computer Interaction

This paper explores the paradoxical productivity losses that can occur in human-AI interactions and proposes strategies for mitigating these issues to improve overall efficiency.
Multimodality and Attention Increase Alignment in Natural Language Prediction Between Humans and Computational Models

Kewenig, V.*, Lampinen, A., Nastase, S., Edwards, C., D'Estelanx, Q.L., Rechardt, A., Vigliocco, G., Skipper, J. (2023) - Under Review

This study investigates how multimodal inputs and attention mechanisms in computational models can improve alignment with human language prediction, potentially enhancing AI's understanding of natural language.
The entire brain, more or less is at work: 'Language regions' are artefacts of averaging

Aliko, S., Wang, B., Kewenig, V., Small, S.L., Skipper, J. (2023) - Under Review at Nature

This research challenges the concept of specific 'language regions' in the brain, suggesting that language processing involves broader neural networks than previously thought.
The Ecological Brain: Reframing the Study of Human Behaviour and Cognition

Vigliocco, G., Conventino. L., De Felice, S., Gregorians, L., Kewenig, V., Musolesi, M., Hudson-Smith, A., Tyler, N., Fluori, E., Spiers, H. (2023) - Accepted at Royal Society Open Science

This paper proposes a new framework for studying human behavior and cognition, emphasizing the importance of ecological validity in neuroscience and psychology research.
When abstract becomes concrete: naturalistic encoding of concepts in the brain

Kewenig, V.*, Vigliocco, G., Skipper, J. (2022) - Accepted at Elife

This study explores how the brain encodes abstract concepts in naturalistic settings, providing insights into the neural mechanisms underlying conceptual representation and language comprehension.
More than Words: Caregivers Selectively Use Iconic and Indexical Cues in Communication with Children

Motamedi, Y.*, Murgiano, M.*, Kewenig, V.*, Grzyb, B., Gu, Y., Brieke, Marshall, C., Wonnacott, E., Perniss, P., Vigliocco, G. (2022) - Published at Child Development

This research examines how caregivers use non-verbal cues, such as gestures and pointing, in communication with children, shedding light on the multimodal nature of language acquisition.
Intentionality But Not Consciousness: Re-Considering Robot-Love

Kewenig, V. (2019) - AI Love You, eds. Fischer, Zhou. ISBN: 978-3-030-19734-6

This chapter explores the philosophical implications of human-robot relationships, focusing on the concept of intentionality in AI and its role in emotional connections.
Robots As Intentional Agents: Using Neuroscientific Methods To Make Robots Appear More Social

Kewenig, V.*, Zhou, Y., Fisher, M. (2018) - Frontiers Psychology

This paper discusses the application of neuroscientific methods to enhance the social appearance of robots, exploring ways to make human-robot interactions more natural and intuitive.
AI Safety: Comments, Questions and Concerns

Kewenig, V.*, Sayed, M. (2016)

This work addresses key issues in AI safety, presenting a critical analysis of potential risks and ethical considerations in the development and deployment of artificial intelligence systems.