Buckner, Cameron
(2025)
The Talking of the Bot with Itself: Language Models for Inner Speech.
[Preprint]
Abstract
In this essay, I explore the idea of using large language models (LLMs) not as full models of general artificial intelligence themselves, but as components that can help bootstrap cognitive architectures comprised of other components to greater degrees of cognitive flexibility and agency. In particular, I explore the idea that LLMs could perform some of the roles that inner speech plays in human cognitive development and adult problem-solving. Researchers are currently exploring many questions of the form: can an LLM (such as OpenAI’s ChatGPT or AnthropicAI’s Claude) have cognitive/mental property X (where X =… represent world models, reason, be conscious, exhibit theory of mind, communicate, and more). If instead of evaluating language models themselves as the sole bearer of X, we instead tried to use LLMs to play the role in the developmental process of acquiring X played by inner speech—as an internal, linguistically-vehicled coordinator and scaffold for diverse other processes—then the significance of research on LLMs as a path towards AI deserves fresh reevaluation, and a different research agenda for philosophically-motivated, deep-learning-based AI comes into focus.
Monthly Views for the past 3 years
Monthly Downloads for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[img]](https://philsci-archive.pitt.edu/style/images/fileicons/text.png)
![[feed]](http://philsci-archive.pitt.edu/style/images/feed-icon-32x32.png){kind=icon}