AIhub monthly digest: April 2024 – explainable AI, access to compute, and noughts and crosses

by Lucy Smith

30 April 2024

Meeting researchers working on explainable AI

In a series of interviews, we’re chatting to some of the AAAI/SIGAI Doctoral Consortium participants to find out more about their research. In our latest two interviews, we met Bálint Gyevnár and Mike Lee and asked about their work on different aspects of explainable AI.

Open vs closed science

The AIhub coffee corner captures the musings of the AIhub trustees over a short conversation. This month, our experts consider the debate around open vs closed science.

Are emergent abilities of large language models a mirage?

Rylan Schaeffer, Brando Miranda and Sanmi Koyejo won a NeurIPS 2023 outstanding paper award for their work Are Emergent Abilities of Large Language Models a Mirage? In their paper, they present an alternative explanation for emergent abilities in large language models. We spoke to Brando about this work, their theory, and what inspired it.

Is compute the binding constraint on AI research?

In their work Resource Democratization: Is Compute the Binding Constraint on AI Research?, presented at AAAI 2024, Rebecca Gelles, Veronica Kinoshita, Micah Musser and James Dunham investigate researchers’ access to compute and the impact this has on their work. In this interview, Rebecca and Ronnie tell us about what inspired their study, their methodology and some of their main findings.

The moderating effect of instant runoff voting

In this blogpost, Kiran Tomlinson writes about work presented at AAAI 2024 on instant runoff voting, a system where voters rank the candidates, rather than providing only their top choice. The candidate with the fewest first-place votes is eliminated, with votes reallocated to the next-preferred candidate on each ballot. This process is iterated until one candidate remains, or until one candidate reaches a majority. Kiran and colleagues provide a mathematical backing for the argument that instant runoff voting favours moderate candidates in a way that plurality (where everyone is asked to pick their favourite, and the candidate with the most votes wins) doesn’t.

Exploring the relationship between anthropomorphism in voice assistants and user safety perception

In their paper The effect of anthropomorphism of virtual voice assistants on perceived safety as an antecedent to voice shopping, Guillermo Calahorra-Candao and María José Martín-de Hoyos explore user perceptions of virtual assistants in voice shopping. They write about their key findings in this blogpost.

Proposal for European Moonshot in AI gains momentum

The ambitious “Moonshot Proposal” for European Sovereignty in AI, presented by CLAIRE and euRobotics, in November 2023, has been endorsed by the European Association for AI (EurAI), and the Centre for European Policy Studies (CEPS). The proposal aims to establish Europe as a powerhouse in trustworthy AI by 2030. You can find out more here.

2024 AI Index Report published

The AI Index Report aims to track, collate, distil, and visualise data related to artificial intelligence. The report is published on a yearly basis, and the nine-chapter, 500-page 2024 edition was released on 15 April.

Workshops on eXplainable AI approaches for deep reinforcement learning, and responsible language models

In this fourth round-up article of the workshops at AAAI 2024, the organisers of the two events on 1) eXplainable AI approaches for deep reinforcement learning, and 2) responsible language models introduce their workshop and present their key takeaways.

gLLMglnlmvlvMMM

Ben Bolker and Jonathan Dushoff celebrated April Fools’ Day with their release, presenting a generative Large Language Model for generalized, linear or nonlinear, multivariate latent-variable mixed/multilevel modeling.