#17: Seriously, what is Intelligence?

🌈 Abstract

The article discusses the concept of intelligence, particularly in the context of the rapid progress in Artificial Intelligence (AI) and the emergence of large language models (LLMs) like ChatGPT. It aims to:

• Explain why intelligence is often conflated with other concepts like consciousness and emotion
• Disentangle intelligence from these related but distinct concepts
• Discuss what current generations of LLMs are revealing about intelligence
• Predict what future generations of LLMs may look like and what we may learn about human theory of mind

🙋 Q&A

[01] Anthropic Bias

1. What is Anthropic Bias, and how does it lead to the conflation of intelligence with other concepts?

• Anthropic Bias refers to the tendency to view humans as uniquely possessing certain qualities like intelligence, consciousness, emotion, and reason, in contrast to other animals.
• This bias leads to the conflation of intelligence with other concepts like consciousness and reason, as these qualities are so intertwined for humans that they get lumped together.

2. How does the progress in LLMs challenge this conflation?

• LLMs like ChatGPT demonstrate intelligence in terms of information processing and retrieval, but do not exhibit consciousness, emotion, or reasoning abilities.
• This suggests that intelligence may be distinct from these other cognitive capacities, challenging the common conflation.

[02] What We're Learning from LLMs

1. What are some key observations about the capabilities and limitations of current LLMs?

• LLMs can pass Turing tests and demonstrate impressive information-processing abilities, but they lack emotion, consciousness, and the ability to perform logical reasoning.
• The core architecture of LLMs, being transformer-based, may not be sufficient for developing true reasoning capabilities, even with significant scaling.

2. How do LLMs differ from the concept of a "philosophical zombie" (P-Zombie)?

• Like a P-Zombie, LLMs can simulate human-like responses without having an internal conscious experience.
• However, LLMs are a completely new type of entity, distinct from both humans and P-Zombies, with their own unique characteristics.

[03] Dismantling AI Safety Arguments

1. What are the key issues with the argument about the risk of a "digital super-intelligence" being hostile to humans?

• The analogy to inter-species or inter-civilizational conflict is inappropriate, as LLMs are fundamentally different from existing entities.
• Hostility or conflict requires self-preservation instincts, which may not be inherent to artificial intelligence systems not subject to Darwinian evolution.
• The term "super-intelligence" is vague and may conflate information-processing abilities with reasoning capabilities.

[04] Being Precise about Intelligence

1. How does the author suggest we should approach the concept of intelligence?

• The author suggests that as AI advances, we may need to break down the nebulous concept of "intelligence" into more precise characteristics and capabilities.
• This may lead to a more nuanced understanding of human cognition, where intelligence is not a monolithic trait but the outcome of various interacting systems.

[05] The Intelligence Revolution

1. How does the author compare the potential impact of the "intelligence revolution" to the industrial revolution?

• Just as the industrial revolution was defined by a super-linear increase in the energy available to humanity, the author suggests the current "intelligence revolution" may be defined by a super-linear increase in the intelligence available to humanity.
• However, the precise measurement and nature of this "intelligence" is yet to be determined.