Androids do dream of electric sheep, so where next for copyright?

2026-07-10 · Source: Technollama · Field: Legal & Regulatory — Intellectual Property & Patents, Legal Technology (LegalTech), Artificial Intelligence & Machine Learning · Depth: Advanced, long

Summary

Anthropic's paper, "Verbalizable Representations Form a Global Workspace in Language Models," introduces the concept of a "J-space" within large language models (LLMs), which functions as an internal workspace analogous to human access consciousness. This challenges the prevailing "stochastic parrot" view, suggesting LLMs perform more complex reasoning than mere next-token prediction. The "J-space" allows models to hold, compare, and discard alternative representations, influencing outputs and revealing intermediate thoughts like task progress or strategic awareness. Experiments, such as swapping vectors in the workspace, demonstrate its role in reasoning, as removing access reduces models to glorified autocompletion. This research has significant implications for copyright law, particularly regarding authorship and infringement, by questioning the purely mechanistic view of AI-generated content.

Key takeaway

For legal professionals and policymakers assessing AI copyright, the Anthropic paper's "J-space" concept suggests LLMs are not mere "stochastic parrots." You should re-evaluate current assumptions about AI authorship, considering that internal reasoning may constitute a form of originality. Furthermore, distinguish between an AI model and its specific output for infringement claims, treating AI "memorization" more like human recall to avoid stifling model distribution.

Key insights

The Anthropic paper reveals LLMs possess an internal "J-space" for reasoning, challenging the "stochastic parrot" view and impacting copyright.

Principles

• LLMs exhibit internal reasoning beyond next-token prediction.
• Copyright originality may extend to AI's "choice-like" processes.
• AI memorization is akin to human remembering, not fixed copying.

Method

The Anthropic paper's research method involved making interventions within the LLM's internal "J-space" (workspace) to observe and manipulate its reasoning processes and subsequent outputs, demonstrating its functional role.

In practice

• Audit and shape model cognition via workspace interventions.
• Revisit copyright authorship for AI-generated code.
• Distinguish AI model from infringing output in copyright cases.

Topics

• Large Language Models
• AI Copyright
• J-space
• AI Authorship
• Copyright Infringement
• Global Workspace Theory

Best for: Research Scientist, CTO, VP of Engineering/Data, Legal Professional, AI Scientist, Policy Maker

Related on AIssential

• The creative ownership dilemma: When AI and human ingenuity collide — AI's rapid integration into creative fields challenges traditional notions of authorship and copyright, necessitating new legal and ethical frameworks.
• Generative AI floods and dilutes the market for books — Generative AI reshapes creative markets by volume, diluting returns for human-authored works, not by quality.
• German court rules AI comic adaptation of copyrighted photo doesn't violate the original — AI-generated adaptations of copyrighted works may not infringe if specific creative elements are not copied.
• Copyright implications of super-intelligence — Super-intelligence will profoundly challenge existing copyright frameworks, making human authorship and infringement enforcement practically impossible.
• Anthropic's New J-Space: Debunked? — Anthropic's "J-Space" is a computational routing mechanism in LLMs, not a conscious workspace or new mathematical subspace.
• Why Sampling Is Not Choosing: Intentionality, Agency, and Moral Responsibility in Large Language Models — LLMs lack intrinsic intentionality and commitment-bearing agency, precluding moral responsibility despite coherent outputs.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Technollama.