The Potential of LLMs in Academic Research – A Wittgensteinian Assessment

No Thumbnail Available

Files

URL

Journal Title

Journal ISSN

Volume Title

Perustieteiden korkeakoulu | Bachelor's thesis
Electronic archive copy is available locally at the Harald Herlin Learning Centre. The staff of Aalto University has access to the electronic bachelor's theses by logging into Aaltodoc with their personal Aalto user ID. Read more about the availability of the bachelor's theses.
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.

Authors

Date

2024-09-20

Department

Major/Subject

Informaatioverkostot

Mcode

SCI3026

Degree programme

Teknistieteellinen kandidaattiohjelma

Language

en

Pages

29

Series

Abstract

This thesis contributes to the epistemological discussion on the reliability of claims produced by Large Language Models (LLMs) and the potential of LLMs as tools in academic research. The research question is: Can large language models be expected produce reliable outputs as part of academic research practices? The thesis argues that a representationalist understanding of language is causing confusion in the discourse on LLMs, especially related to the question of whether LLMs can understand the language they process. The representationalist view of language is proposed to be replaced with Ludwig Wittgenstein’s (1889–1951) later philosophy of language-games, which has significant parallels to LLM’s functionality. Wittgenstein’s thought is then applied to answer the question of the reliability of LLMs in academic research. The method of the thesis is conceptual analysis. This thesis is intended as a theoretical exploration and hypothesis development for future research. Empirical studies are referenced to discuss the credibility of the hypothesis presented. In a nutshell the Wittgensteinian interpretation of LLMs suggests that not only the content of prompts is consequential for the perceived performance of an LLM, but that the way language is used in the prompt may also have a strong effect on the outputs. The Wittgensteinian interpretation also offers its own explanation for this phenomenon: Human language use is structured into different language-games and as these language games are present in the training data of LLMs, LLMs detect these structures and replicate them in their interaction with the prompter. The prompt determining the language-game has major implications to LLM use in academia. It is crucial, that the prompt aligns in both surface level language and norms and assumptions of the language-game that the person writing the prompt is seeking to play. If the prompt assigns the LLM to play the correct language-game, there are no theoretical limits other than the textual medium to what the LLM can do. Practical limitations such as the model’s size and the quality of training data may still play a role that would obstruct the model from fulfilling its potential.

Description

Supervisor

Fagerholm, Fabian

Thesis advisor

Luoma, Jukka

Keywords

large language models, philosophy, language, artificial intelligence

Other note

Citation

Permanent link to this item

https://urn.fi/URN:NBN:fi:aalto-202409246489

Collections

[kand] Perustieteiden korkeakoulu / SCI