Mark Crowley | Fall 2023 Reading Group

Motivation

Continuation of last year’s Transformers Reading Group keeping the theme on Generative AI and RL to start, but who knows where it will go!

General Reading Groups Tips

In a reading group everyone takes turns leading discussion of a paper each week. Leading discussion can be as simple as having your own annotated notes on Hypothes.is to share and start discussion as we go through it together. Or it could be more involved, including making slides to present your overview of the paper’s contributions, highlights and weak points.

See the links and notes on paper we have done in previous meetings, obtain the link for the next paper or look at planned upcoming or potential future papers, feel free to suggest others or changes in the upcoming order.

Jump to stage: next ~ done ~ upcoming ~ potential

done

[2] Causal Parrots: Large Language Models May Talk Causality But Are Not Causal

Matej Zečević, Moritz Willig, Devendra Singh Dhami, and Kristian Kersting.

Transactions on Machine Learning Research. Aug, 2023.

Abs arXiv PDF URL Hypoth Notes

Note: This paper has a very bold premise to show clearly what causality and conceptual reasoning means in terms of Large Language Models. They define it clearly int terms of evidence existing in the training set and argue that LLMs can only learn causal relations in those cases. These authors are organizing a workshop on this very topic for NeurIPS 2023.

Abstract: Some argue scale is all what is needed to achieve AI, covering even causal models. We make it clear that large language models (LLMs) cannot be causal and give reason onto why sometimes we might feel otherwise. To this end, we define and exemplify a new subgroup of Structural Causal Model (SCM) that we call meta SCM which encode causal facts about other SCM within their variables. We conjecture that in the cases where LLM succeed in doing causal inference, underlying was a respective meta SCM that exposed correlations between causal facts in natural language on whose data the LLM was ultimately trained. If our hypothesis holds true, then this would imply that LLMs are like parrots in that they simply recite the causal knowledge embedded in the data. Our empirical analysis provides favoring evidence that current LLMs are even weak ‘causal parrots.’

AI4Science

[1] Scientific discovery in the age of artificial intelligence

Hanchen Wang, Tianfan Fu, Yuanqi Du, Wenhao Gao, Kexin Huang, Ziming Liu, Payal Chandak, Shengchao Liu, Peter Van Katwyk, Andreea Deac, Anima Anandkumar, Karianne Bergen, Carla P. Gomes, Shirley Ho, Pushmeet Kohli, Joan Lasenby, Jure Leskovec, Tie-Yan Liu, Arjun Manrai, Debora Marks, Bharath Ramsundar, Le Song, Jimeng Sun, Jian Tang, Petar Veličković, Max Welling, Linfeng Zhang, Connor W. Coley, Yoshua Bengio, and Marinka Zitnik.

Nature. 620, (7972). 2023.

Abs URL Hypoth Notes

Note: This is a great paper to look at for an update on the many ways AI/ML/RL are being used for science. It is written by the organizers of the regular [AI for Science workshop](https://ai4sciencecommunity.github.io/) held at multiple major conferences. There are extensive notes in the hypothesis link of the webpage version of the paper, click the hypothesis link to see the notes.

Abstract: Artificial intelligence (AI) is being increasingly integrated into scientific discovery to augment and accelerate research, helping scientists to generate hypotheses, design experiments, collect and interpret large datasets, and gain insights that might not have been possible using traditional scientific methods alone. Here we examine breakthroughs over the past decade that include self-supervised learning, which allows models to be trained on vast amounts of unlabelled data, and geometric deep learning, which leverages knowledge about the structure of scientific data to enhance model accuracy and efficiency. Generative AI methods can create designs, such as small-molecule drugs and proteins, by analysing diverse data modalities, including images and sequences. We discuss how these methods can help scientists throughout the scientific process and the central issues that remain despite such advances. Both developers and users of AI tools need a better understanding of when such approaches need improvement, and challenges posed by poor data quality and stewardship remain. These issues cut across scientific disciplines and require developing foundational algorithmic approaches that can contribute to scientific understanding or acquire it autonomously, making them critical areas of focus for AI innovation.