From Knowledge to Wisdom: The Power of Large Language Models in AI

0583712 - ÚI 2024 RIV NL eng V - Research Report
Wiedermann, Jiří - van Leeuwen, J.
From Knowledge to Wisdom: The Power of Large Language Models in AI.
Utrecht: Utrecht University, 2023. 14 s. Technical Report, UU-PCS-2023-01.
Grant - others:AV ČR(CZ) StrategieAV21/26
Program: StrategieAV
Institutional support: RVO:67985807
Keywords : artificial wisdom * DIKW hierarchy * epistemic computation * generative AI * illusory intelligence * knowledge * large language models * 4E cognition
OECD category: Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)
https://webspace.science.uu.nl/~leeuw112/techreps/UU-PCS-2023-01.pdf

ZÁKLADNÍ ÚDAJE: Utrecht: Utrecht University, 2023. Technical Report, UU-PCS-2023-01. ABSTRAKT: In this paper, we explore the purpose and potential of artificial intelligence (AI) in light of the current generation of large language models. We argue that these models can be seen as tools for acquiring and generating artificial wisdom, enabling us to make wiser decisions and behave more intelligently. Unlike earlier AI approaches, which focused on generating knowledge from data, contemporary language models have the ability to extract meaning from syntactic patterns and relate this meaning to real-world descriptions. This capacity reflects a form of cognition known as 4E cognition, which emphasizes the embodied, embedded, extended, and enacted nature of intelligent behavior. We argue that contemporary language models possess a form of illusory intelligence and illusory wisdom that has not been described in the literature before. This insight challenges the traditional computational approaches to AI and opens up new avenues for research on the relationship between language, cognition, and intelligence. By recognizing the potential of large language models to generate and use wisdom, we can move beyond knowledge-centric approaches to AI and develop more nuanced models of intelligent behavior.
Permanent Link: https://hdl.handle.net/11104/0351705