# Viewpoint: A Theoretical Computer Science Perspective on Consciousness and Artificial General Intelligence

**Authors:** Lenore Blum, Manuel Blum  
**arXiv:** [2303.17075](https://arxiv.org/abs/2303.17075)  
**Date:** 2023-03-30  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.17075v1/

## Key points

- The Conscious Turing Machine (CTM) is a theoretical computer science model designed to investigate consciousness.
- CTM is not a model of the brain, but rather a model for consciousness.
- CTM has an enormous number of powerful processors, some with specialized expertise and others unspecialized but poised to develop expertise.
- CTM utilizes processors that have the required knowledge, ability, and time to work on any given problem, even if it is not aware of which ones these may be.
- CTM offers a thoughtful and novel guide to the creation of an Artificial General Intelligence (AGI).
- CTM incorporates modularity by utilizing multiple processors with different input domains, knowledge, and functionalities.
- The possibility of emergence in AGI is supported by early experiments with large language models such as GPT-4.
- Architectural modularity is much better aligned with real-life language use.
- A principal contribution of the CTM is its way of coordinating processors that must solve diverse collections of unforeseen problems effectively.
- While deep learning alone may suffice for AGI creation without using a global workspace model or CTM, it may require a matrix size that grows with the square of the number of problems it needs to solve.
- In contrast, the CTM can handle 107 AIs with 1014 parameters per AI for a total of 1021 parameters.
- The Conscious Turing Machine offers valuable insights into creating Artificial General Intelligence through its unique approach to task coordination among its numerous powerful processors.

## Summary

The Conscious Turing Machine (CTM) is a theoretical computer science model designed to investigate consciousness. It is intentionally simple and not a model of the brain, but rather a model for consciousness. The CTM has an enormous number of powerful processors, some with specialized expertise and others unspecialized but poised to develop expertise. This allows the CTM to utilize processors that have the required knowledge, ability, and time to work on any given problem, even if it is not aware of which ones these may be. The CTM offers a thoughtful and novel guide to the creation of an Artificial General Intelligence (AGI). It incorporates modularity by utilizing multiple processors with different input domains, knowledge, and functionalities. The possibility of emergence in AGI is supported by early experiments with large language models such as GPT-4. However, it is argued that architectural modularity is much better aligned with real-life language use. A principal contribution of the CTM is its way of coordinating processors that must solve diverse collections of unforeseen problems. Despite having no central processing unit or global workspace model, the CTM assigns tasks to its processors effectively. While deep learning alone may suffice for AGI creation without using a global workspace model or CTM, it may require a matrix size that grows with the square of the number of problems it needs to solve. This would be difficult to achieve since current AIs use about 1014 parameters. In contrast, the CTM can handle 107 AIs with 1014 parameters per AI for a total of 1021 parameters. In conclusion, while developed primarily for understanding consciousness, the Conscious Turing Machine offers valuable insights into creating Artificial General Intelligence through its unique approach to task coordination among its numerous powerful processors.