Gomes, Henrique and Shyam, Vasudev (2024) On eavesdropping octopuses and stochastic parrots: what do they know? [Preprint]

Text token.pdf Download (318kB)

Abstract

The extant literature on AI (and popular culture more generally) has a few popular slogans that seek to dismiss the cognitive capacities of current large-language models (LLMs). Here, from a conceptual standpoint, we assess whether two such slogans have any teeth. The first such slogan is that ``LLMs can only predict next-tokens". The second is that “AIs are stochastic parrots”. We will briefly explain these two slogans, and argue that, in plausible construals, they do not imply fundamental limitations to cognition and semantic grounding (which of course does not imply anything positive about current AI's cognitive capacities). The difference between our approach and that of the burgeoning literature reaching a similar conclusion is that we base our arguments on the idea of `knowledge-first epistemology'.

---

Export/Citation:

EndNote | BibTeX | Dublin Core | ASCII/Text Citation (Chicago) | HTML Citation | OpenURL

Social Networking:

Share |

---

Item Type:	Preprint
Creators:	Creators Email ORCID Gomes, Henrique gomes.ha@gmail.com 0000-0002-9285-0090 Shyam, Vasudev vasudevshyam@gmail.com
Keywords:	AI, LLM, symbolic grounding
Subjects:	General Issues > Data Specific Sciences > Artificial Intelligence General Issues > Computer Simulation Specific Sciences > Artificial Intelligence > Machine Learning
Depositing User:	Dr Henrique Gomes
Date Deposited:	16 Oct 2024 12:03
Last Modified:	16 Oct 2024 12:03
Item ID:	24073
Subjects:	General Issues > Data Specific Sciences > Artificial Intelligence General Issues > Computer Simulation Specific Sciences > Artificial Intelligence > Machine Learning
Date:	10 October 2024
URI:	https://philsci-archive.pitt.edu/id/eprint/24073

Monthly Views for the past 3 years

Monthly Downloads for the past 3 years

Plum Analytics

Actions (login required)

View Item