LLMs and Philosophy

" LLMs and Philosophy " is an international conference dedicated to exploring the profound connections between studies of language, developments of large language models (LLMs), and philosophical inquiries. This interdisciplinary event seeks to bring together experts, scholars, and enthusiasts from various fields such as machine learning, artificial intelligence, linguistics, philosophy, and cognitive science to delve into the complexities and implications of language models in understanding human language, communication, and mind, as well as ethical considerations in the digital age.

Conference Description

In recent years, the advancement of large language models (LLMs) has revolutionized our approach to processing and generating human language, offering unprecedented opportunities for enhancing communication, creativity, and problem-solving. However, this technological leap also poses significant philosophical questions regarding the nature of language, the human mind, and the ethical implications of artificial intelligence. This conference aims to foster a deep and multifaceted discussion on these topics, encouraging a synthesis of ideas from both technical and philosophical domains.

Objectives:

- Exploring the Foundations: Examine the underlying principles of language and its representation in LLMs, focusing on semantics, syntax, and philosophy of language.

- Linguistic, Cognitive, and Ethical Implications: Investigate the cognitive aspects of language understanding and generation by LLMs, and their implications for human language, cognition, and ethics in AI.

- Interdisciplinary Dialogue: Bring together researchers in machine learning, AI, linguistics, and various fields of philosophy to promote interdisciplinary collaboration and dialogue.

Key Topics:

- Philosophy of Language, X-Phi, and LLMs: Can philosophy of language learn from LLMs? For example, if the truth-value judgments of one LLM about an utterance in fictional scenarios largely reflect people’s intuitions, can we use the judgments as evidence? Alternatively, if they deviate from people’s intuitions, can we learn some lessons from such data, say about what is said, given that the LLM has been trained on huge data sets of people’s linguistic uses? Or, if such a deviation is or becomes rare, can we use LLMs to spare empirical studies in experimental philosophy in general?

- Multilingual LLMs and the Problem of Linguistic Diversity in Philosophy: Philosophical concepts seem to suffer from linguistic diversity, since, if a philosophical concept is captured differently in different languages with even some gap in extension, this “disagreement” (if it is a disagreement at all) seems hard to be settled. If so, philosophical discussions would be language-specific there. On the other hand, multilingual LLMs are often said to have developed a language-neutral concept.

- Multimodal LLMs and Theory of Meaning: It has often been claimed that AI suffers from what is called the symbol grounding problem. However, contemporary AIs based on deep learning such as LLMs have directly learned people’s actual uses of words in large corpora. The impressive linguistic performance of LLMs therefore seems to have already solved the problem, because of the causal connection with the external world through actual human uses of linguistic expressions, as long as the human users of languages are interacting with the external world. One might still argue that such LLMs only have translational semantics. But if so, do recent multimodal LLMs solve this problem and have referential semantics? Or do AIs need to have a body to understand and mean anything by words?

- LLMs and the Nature of Linguistic Competence: One might think that recent LLMs have already achieved the quality of linguistic performance to qualify as being said to have linguistic competence. However, some linguists, as well as philosophers of mind and language, may still claim that what is implemented in LLMs cannot be called linguistic competence. If so, for what reason? Or can “linguistic competence” be considered to be multiply realizable, whether in the human brain or in AI?

- LLMs and Ethical Challenges: The widespread adoption of LLMs presents various social, political, and ethical challenges that demand philosophers’ attention. Some of the questions include: is it feasible to deploy LLMs in a fair manner, free from biases, particularly against groups underrepresented in training datasets? What should we make of the increasing reliance on LLM outputs for individual and political decision-making with respect to human agency? Does the integration of LLMs in decision-making processes diminish human autonomy? What are ethically appropriate relations between humans and artificial agents? Is it morally acceptable to mistreat artificial agents, given their non-human status?

Call for Papers:

This conference is a unique opportunity to contribute to a critical and evolving conversation at the intersection of machine learning, linguistics, and philosophy. We look forward to welcoming you to a stimulating and insightful event that promises to deepen our understanding of the digital and cognitive landscapes shaping our world.

If you are interested, please provide the title and abstract (up to 500 words) of your talk through the following link:

https://forms.gle/Vj2tVG9ChxEHSLyr7

Upcoming Schedule:

Abstract submission deadline: May 31, 2024

Notification of results: by the end of June, 2024