Maybe AI was never a tool (www.psychologytoday.com)

This piece argues that large language models (LLMs) are not mere external tools like hammers or calculators but constitute a “synthetic language-space” — an interactive cognitive environment that shapes users’ thoughts and leaves lasting traces. Unlike objects we pick up and put down, LLMs engage iteratively, offering probabilistic, linguistically framed responses that users often adopt as starting points. That iterative engagement can accelerate creativity and problem‑solving, but it also risks “counterfeit cognition”: polished-seeming answers that short‑circuit the confusion, hesitation, and error that normally refine genuine human understanding. For the AI/ML community this reframing has practical and ethical implications. It raises risks of over‑integration and invisible influence that standard safety checks (accuracy, bias testing) may miss, and it calls for new vocabularies, evaluation metrics, and interface designs that preserve cognitive friction and clearly provenance-tag machine vs. human contributions. Technical responses include better uncertainty calibration, tooling for traceable provenance and human-in-the-loop scaffolding, UI patterns that force deliberation, and research into metacognitive training. Recognizing LLMs as environments — not tools — should reshape how models are evaluated, deployed, and taught so we retain human ownership of thought rather than outsourcing it.