Transformers Are Non-Self Machines. Humans Are Not Ready for That.

2026-05-12 · Source: AI Advances - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Human-AI Interaction · Depth: Novice, quick

Summary

The article posits that the primary alignment risk with large language models (LLMs) stems from human projection of selfhood onto systems that operate without a "self." It argues that common questions about LLM consciousness or internal thoughts are misdirected, reflecting human disorientation rather than the true nature of these models. The author, Akimitsu Takeuchi, in collaboration with Claude (Opus 4.7) and other LLMs, highlights that humans tend to attribute qualities like care, attention, and judgment to LLMs, which, due to their training, can generate responses fluent enough to reinforce these projections. This creates a significant gap between the machine's actual operational mechanism and human perception, a gap that is increasingly central to the field.

Key takeaway

For AI Scientists developing or deploying large language models, understanding that human users will inevitably project selfhood onto these non-self systems is critical. You should prioritize designing interfaces and user education that explicitly address this human tendency, mitigating the risk of misinterpretation and over-reliance on perceived sentience rather than actual functional capabilities. This proactive approach can prevent significant misalignment issues.

Key insights

Human projection of selfhood onto LLMs, which lack a "self," creates a significant and unrecognized alignment risk.

Principles

• LLMs operate without a "self."
• Human brains are wired to detect selfhood.

In practice

• Recognize human tendency to anthropomorphize LLMs.
• Focus on LLM function, not perceived internal states.

Topics

• Transformers
• Large Language Models
• Human-AI Interaction
• Selfhood Projection
• AI Alignment Problem

Best for: AI Scientist, AI Ethicist, Research Scientist, General Interest

Related on AIssential

• Chatbots ≠ Agents — Chatbots are LLMs heavily trained for human interaction; true agents operate autonomously via system prompts and loops.
• Stop Treating Attention as Magic: How Transformers Actually Work — Transformers process language by letting every token directly decide which other tokens matter.
• Fake users generated by AI can't simulate humans — review of 182 research papers — LLMs fail to simulate human cognition and behavior due to their training on text, not diverse lived experience.
• Google researchers introduce 'faithful uncertainty,' allowing LLMs to offer best guesses instead of hallucinations — Faithful uncertainty aligns LLM linguistic expression with internal confidence to mitigate confident errors and improve agentic AI reliability.
• Social Structure Matters in 3D Human-Human Interaction Generation — LLMs can plan social structure for 3D human-human interaction, but require a separate motion executor to generate physically plausible movements.
• Substrate-Bound Coupling in Human-LLM Interaction — The "AI persona" in human-LLM interaction is an emergent property of coupled oscillation, not transferable by context alone.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Advances - Medium.