[Paper Review] Could a Large Language Model be Conscious?

David J. Chalmers|arXiv (Cornell University)|Mar 4, 2023

Topic Modeling127 citations

TL;DR

Chalmers analyzes arguments for and against LLM consciousness, noting current obstacles but suggesting future successors might achieve consciousness.

ABSTRACT

There has recently been widespread discussion of whether large language models might be sentient. Should we take this idea seriously? I will break down the strongest reasons for and against. Given mainstream assumptions in the science of consciousness, there are significant obstacles to consciousness in current models: for example, their lack of recurrent processing, a global workspace, and unified agency. At the same time, it is quite possible that these obstacles will be overcome in the next decade or so. I conclude that while it is somewhat unlikely that current large language models are conscious, we should take seriously the possibility that successors to large language models may be conscious in the not-too-distant future.

Motivation & Objective

• Assess whether current large language models exhibit consciousness under mainstream theories of consciousness.
• Identify major obstacles in current LLMs (e.g., recurrent processing, global workspace, unified agency).
• Evaluate the likelihood that future LLMs could overcome these obstacles and become conscious.

Proposed method

• Review and critique of standard consciousness criteria as applied to LLMs.
• Systematic analysis of structural requirements like recurrent processing, global workspace, and unified agency.
• Discussion of technological trajectories and future possibilities for overcoming current limitations.

Experimental results

Research questions

• RQ1 Do current large language models meet conventional criteria for consciousness?
• RQ2 What are the main obstacles preventing LLMs from being conscious today?
• RQ3 Could forthcoming successors to LLMs plausibly overcome these obstacles and be conscious in the not-too-distant future?

Key findings

• Current LLMs are unlikely to be conscious under mainstream theories due to missing recurrent processing, global workspace, and unified agency.
• There is a plausible path for successors to overcome these obstacles within the next decade or so.
• Despite current skepticism, the possibility of future conscious LLMs warrants serious consideration and continued inquiry.