Living with an Alien Co-intelligence

Large language models can feel uncannily capable: they write with confidence, imitate roles, connect distant ideas and sometimes produce work that seems far beyond a simple tool. This episode uses the idea of an “alien co-intelligence” to make sense of that experience without pretending the system thinks, knows or cares as a person does.

The discussion explores AI’s uneven competence — impressive on some tasks, unreliable on others — and why that boundary is difficult to predict in advance. It also confronts the dangerous side of fluent output: plausible fabrication, inherited bias and an easy human tendency to project intention, empathy or authority onto a responsive system. The practical answer is neither blind optimism nor refusal. It is active experimentation, clear roles and human accountability at the point where decisions, facts and consequences matter.

For creative work, that can mean using AI to widen the field of possibilities, challenge a stale brief or help make an idea tangible. For factual or high-stakes work, it means checking claims and retaining ownership of the result. The episode’s central proposition is simple: treat the system as useful and unfamiliar. Learn where it helps in your own work, but do not hand it the final judgement — or mistake a convincing voice for a reliable mind.

In this episode

• Why the “alien co-intelligence” frame is more useful than treating AI as ordinary software or a person.

• The jagged boundary between tasks a model handles well and tasks it gets confidently wrong.

• Why hallucinations and human anthropomorphism make verification a non-negotiable habit.

• How experimentation, role-setting and human judgement can make AI a better creative collaborator.

Source and AI disclosure

This audio episode was generated in Google NotebookLM from Richard’s own knowledge materials. The accompanying episode notes were developed from the supplied transcript. The discussion is a source-led interpretation, not an independent technical or safety assessment of any AI system.