Adjoint Thinking: How to Think with Machines Without Losing Your Mind

I am adapting my book, Adjoint Thinking: How to Think with Machines Without Losing Your Mind, into a seven-part digital documentary and learning series. Each film will run for about ten minutes and will explore one practical problem in the way people now think, decide, write, and create alongside generative AI.

Project Summary

Generative AI has made producing an answer much easier. It has not made judging an answer easier. This distinction is the starting point of this project.

A model can produce a confident, coherent response in seconds. The danger is not only that the response may be wrong. The deeper problem is that a polished answer often arrives before the human user has formed an independent view of the problem. Once that happens, the machine has already influenced the frame, the vocabulary, the apparent alternatives, and sometimes even the question being asked.

I wrote Adjoint Thinking around this problem. The book is not a manual for writing better prompts. It asks a different question: how should a person think when intelligent machines are always available to think with them?

The seven films will translate the book's central ideas into short, practical lessons. They will examine problems such as automation bias, loss of information provenance, premature machine framing, hallucination, and the temptation to delegate judgement simply because delegation has become convenient.

Each episode will stand on its own. A viewer should be able to watch one ten-minute film and leave with something they can use immediately.

For example, one of the habits developed in the book is what I call first contact. Before asking a machine to analyse an important problem, the human records an initial interpretation, hypothesis, sketch, or decision boundary. It does not need to be correct. Its purpose is to preserve evidence that independent human reasoning existed before machine intervention.

Another episode will deal with source amnesia. After several rounds of AI-assisted work, it can become surprisingly difficult to remember which statement came from an original source, which was our own inference, and which was introduced by the model. The film will show simple ways of preserving that distinction.

The broader objective is to make AI use more intellectually disciplined without turning it into a cumbersome compliance exercise.

What are this project's goals? How will you achieve them?

The first goal is to help people recognise the fluency illusion. Generative AI is unusually persuasive because its errors often have the same grammatical confidence as its correct answers. I want viewers to develop the habit of looking past presentation quality and asking what assumptions, omissions, boundaries, and unsupported transitions sit underneath the response.

The second goal is to protect independent human judgement. The series will introduce a compact set of practices for keeping humans involved at the points where their involvement matters most: before framing, during verification, when interpreting evidence, and when accepting consequences.

The third goal is to make these ideas freely available. All seven films will be published online without a paywall. Companion materials will also be released for free, including worksheets, diagrams, decision checks, and simple team exercises that organisations can adapt to their own work.

The seven films will broadly follow the architecture of the book:

1. Preserve an independent human position before the machine frames the problem.

2. Keep track of where information and ideas came from during AI-assisted work.

3. Use machine reasoning without transferring final responsibility to the machine.

4. Treat hallucinations as material to inspect or explore, not as facts to accept.

5. Increase verification effort when the consequences of being wrong become larger.

6. Understand what changes when AI systems operate repeatedly or at scale.

7. Develop a practical way of remaining the final decision-maker in a machine-assisted workflow.

The films will use examples rather than lectures wherever possible. I want the viewer to see a situation unfold: a manager accepting a beautifully written but weak analysis, a researcher losing track of a source, a creative professional discovering that an apparent machine error contains an interesting idea, or a team scaling an automated process before anyone has decided who is responsible when it fails.

The production itself will be deliberately small. Rather than assemble a conventional film crew, I will use an AI-assisted production workflow built around tools such as HeyGen, ElevenLabs, and Descript Pro. These tools make it possible for one researcher-author to produce narration, editing, visual sequences, and multilingual-ready media at a quality that would previously have required a substantially larger production budget.

The scripts, argument structure, examples, and educational design will remain human-directed.

How will this funding be used?

I am seeking a $10,000 microgrant to produce the complete seven-film series and its supporting educational materials.

$3,500 — Video and audio production

This will cover the software required for synthetic narration, video generation, editing, transcription, captioning, and post-production across the seven films.

$2,500 — Website, distribution, and hosting

The films themselves will be publicly accessible, including through YouTube. This allocation covers the project website, domain and hosting costs, storage, distribution infrastructure, and the delivery of downloadable learning materials without a paywall.

$2,000 — Graphic design and educational materials

The book contains frameworks that work better visually than verbally. This portion will fund the design of diagrams, worksheets, downloadable PDFs, visual decision tools, and team exercises accompanying the films.

$2,000 — Script adaptation and production work

Writing a book chapter and writing a ten-minute film are very different tasks. This funding supports the concentrated work required to reduce the longer arguments in the book into short scripts without stripping away the reasoning that makes them useful.

The total requested budget is $10,000.

Who is on your team? What's your track record on similar projects?

I am currently the sole researcher, author, and project lead.

My professional background is in fluid mechanics, engineering simulation, and complex systems modelling. Much of my academic work has involved building mathematical or computational representations of systems that cannot be understood reliably by looking at one variable in isolation.

That background strongly influenced Adjoint Thinking.

I became interested in human-AI interaction less as a philosophical question than as a systems problem. When a person and a generative model work together, information moves between two very different reasoning systems. Each changes the state of the other. Errors can propagate. Assumptions can become hidden. Feedback can improve the result, but it can also amplify a mistake.

The book grew from trying to describe that interaction in a useful way.

I therefore bring two things to this project: the original framework itself and experience translating technically difficult ideas into structured models, explanations, figures, and research communication.

The challenge now is different. The ideas have already been developed in written form. The purpose of this project is to make them accessible to people who are unlikely to read a full book about cognition and AI but may watch a ten-minute film and change one important habit as a result.

What are the most likely causes and outcomes if this project fails?

The most obvious failure would be making seven films that are intellectually correct but unpleasant to watch.

The source material is conceptual, and some of it is dense. If I simply convert book chapters into narrated summaries, the result will feel like a lecture. The scripts therefore need to be built around situations, conflicts, examples, and decisions. Every episode has to earn its ten minutes.

A second risk is producing something that appeals mainly to people who are already preoccupied with AI.

That would miss the point. The people most affected by these problems are not necessarily AI specialists. They are managers, researchers, writers, analysts, designers, teachers, entrepreneurs, and students who increasingly use generative systems as part of ordinary work.

The series will therefore avoid depending on one particular model, vendor, or interface. The films should remain useful even as today's AI products are replaced.

A third risk is distribution.

Publishing a film does not mean that anyone will see it. I plan to treat distribution as part of the project rather than as something that happens afterward. Short extracts, diagrams, arguments, and examples from each film will be adapted for professional networks such as LinkedIn and used to lead viewers toward the complete free series.

The project will have failed if it produces attractive videos but no change in behaviour.

The outcome I care about is much smaller and more concrete: that someone stops before accepting an AI-generated conclusion, writes down their own interpretation first, checks where an important claim came from, or decides that a particular judgement should not be delegated.

That is the kind of change Adjoint Thinking is intended to produce.