Marko Tesic

Researcher @LeverhulmeCFI @Cambridge_Uni. Exploring AI capabilities (mostly LLMs) in the context of occupational tasks. Violinist @paprikamusic & @PicoPlayers

[Lorenzo Pacchiardi @LPacchiardi]

🚀 New Paper: “Paradigms of AI Evaluation: Mapping Goals, Methodologies, and Culture” https://t.co/eVLSkJy6O2

We survey 125+ studies and identify 6 major paradigms of AI evaluation, each shaped by distinct goals, methodologies, and research cultures.

🧵

[Marko Tesic @m_tesic]

We discussed how to use the measurement layouts to learn about the capabilities of RL agents and LLMs. All code, data, and presentations can be found here: https://t.co/KJtf6xYtRe

[Marko Tesic @m_tesic]

It was a great experience presenting our tutorial on the capability-oriented approach to AI evaluation at @RealAAAI. @KozzyVoudouris, José Hernández-Orallo, @JohnJBurden, and I showcased measurement layouts (Bayesian hierarchical models) to infer the capabilities of AI systems.

[Marko Tesic @m_tesic]

Bilo mi je zadovoljstvo gostovati na @radiogalaksija. Veliko hvala Dušanu na pozivu!

[Future Intelligence @LeverhulmeCFI]

We have TWO new full-time positions open, to work on our two-year project in collaboration with Accenture entitled 'Aligning assessment of cognitive abilities in cutting edge AI and those required in human employees'. https://t.co/FnQwQAyFGz @Accenture

[Marko Tesic @m_tesic]

Happy to have published this work with Ulrike Hahn in @CognitionJourn. It's been a long time coming, originating from my PhD thesis.

[Cognition @CognitionJourn]

How do explanations, confidence in a claim being explained, and the reliability of the source interact? Reliability mediates the impact of explanations on confidence. Explanations have a greater impact on confidence when reliability is low

https://t.co/ICxid9YmIK

[Paprika - @paprikamusic]

@WOMADfestival We can't wait to perform @WOMADfestival! #womadfestival #balkanmusic #paprikamusic

[Marko Tesic @m_tesic]

Excited to announce that I've joined @LeverhulmeCFI @Cambridge_Uni as a postdoc researcher. I'm working together with @LucyCheke, José Hernández-Orallo, & the @OECD on exploring AI capabilities and mapping them onto the specific demands in human workforce.

[Marko Tesic @m_tesic]

This work is now published in Patterns: https://t.co/2a1Kib40MP

A cautionary note on how explanations of predictive AI models (which explain/make explicit the correlations in data that the AI model has captured) can affect our beliefs about the causal relationships in the world

[Marko Tesic @m_tesic]

Can counterfactual (CF) explanations of AI systems' predictions skew lay users' causal intuitions about the world?

In two experiments we (Ulrike Hahn and I) find that they can. But we also find that we may be able to guard against that. Preprint: https://t.co/4sAVVCuMR9

1/

[Marko Tesic @m_tesic]

The workshop is on the 23rd and the 24th of September. More info at: https://t.co/wbhi1ecAmN

Register at: https://t.co/wqttQNOILm

2/

[Marko Tesic @m_tesic]

Happy to share that the workshop will also host a panel exploring: what can cognitive scientists and machine learning researchers learn from each other when it comes to (explainable) AI? The panelists are: @ruthmjbyrne, Noah Goodman, @HanaChockler, and @AndrewLampinen
1/

[Marko Tesic @m_tesic]

Very excited to announce the Workshop on Human Behavioral Aspects of (Explainable) AI!

Dates: 23rd and 24th of September 2022

Hybrid format: both in person (Birkbeck, University of London) and online

Website: https://t.co/wbhi1ec2xf

(1/4)

[Marko Tesic @m_tesic]

@claireewoodcock @HanaChockler @hal_ashton @katie_m_collins @ruthmjbyrne @SangeetKhemlani @tobigerstenberg @TomerUllman @umangsbhatt Registrations for both in person and online are here: https://t.co/X9TyZ6IX24

(5/4)

[Marko Tesic @m_tesic]

Very excited to announce the Workshop on Human Behavioral Aspects of (Explainable) AI!

Dates: 23rd and 24th of September 2022

Hybrid format: both in person (Birkbeck, University of London) and online

Website: https://t.co/wbhi1ec2xf

(1/4)

[Marko Tesic @m_tesic]

Confirmed speakers:
@claireewoodcock @HanaChockler @hal_ashton @katie_m_collins @ruthmjbyrne @SangeetKhemlani @tobigerstenberg @TomerUllman @umangsbhatt

(4/4)

[Marko Tesic @m_tesic]

The aim of the workshop will be to bring together researchers from psychology/cognitive science and AI/ML to present and discuss issues related to the human side of AI explainability. The workshop will most likely be in late September this year. More to follow. Stay tuned! 2/2

[Marko Tesic @m_tesic]

Happy to announce that I’ve been granted a Post-Doctoral Enrichment Award from @turinginst to organize a workshop on human behavioral aspects of (explainable) AI. 1/2

[Marko Tesic @m_tesic]

A small and fun project. Anyone interested can find the jupyter notebook here: https://t.co/cs1rgOhMW3

[GM Alojzije Jankovic @AlojzijeJ]

Thanks @m_tesic for the research. Since I’m streaming my https://t.co/3x29dZEvGy games, it’s useful to know when can I get a decent opposition :)

[Marko Tesic @m_tesic]

Adding the note reduced the effect of CF explanation on the participants' causal beliefs regarding the factors AI system uses to make predictions.

5/

[Marko Tesic @m_tesic]

Can counterfactual (CF) explanations of AI systems' predictions skew lay users' causal intuitions about the world?

In two experiments we (Ulrike Hahn and I) find that they can. But we also find that we may be able to guard against that. Preprint: https://t.co/4sAVVCuMR9

1/

[Marko Tesic @m_tesic]

Inspired by the work on misinformation and health warning messaging, we designed a note communicating to the participants the correlational character of AI systems rather than causal.

4/