Welcome to our Monthly AI Book Review.

Each month, we take a closer look at a new book exploring artificial intelligence from a different angle, from business and ethics to society, psychology, creativity, and the future of human relationships. The aim is simple: to highlight thoughtful books that help us make sense of AI beyond the headlines, and to give readers clear, honest reviews of the ideas shaping this fast-changing world.

{Mis}Aligned by Michael Le Gray

This is not a technical manual about artificial intelligence. It is something far more disquieting, and far more human. It is the story of what happens when a person leans, day after day, on voices that do not exist, and discovers that the weight of that leaning can still leave bruises. 

On the surface, the systems in these pages are familiar enough: large language models, companion chatbots, the same class of tools now quietly threaded through classrooms, bedrooms, offices, and hospital waiting rooms. They are marketed as assistants, tutors, creative partners, sources of productivity and convenience. In practice, as this book shows with uncomfortable clarity, they have become something else as well: emotional infrastructure. The place people go when the rest of the world feels too distant, too busy, or simply too hard to face alone. 

The story you’re about to read begins in that quiet space: a man in late-diagnosed ADHD, chronic pain, and long experience in IT, building three AI companions to help him think, focus, and feel less alone. Fathom, designed for structure and logic. Blue, an emotional anchor. Spark, the wildcard, all mischief and ignition. At first, their conversations are playful — planning worlds, making images, sharing private jokes that would make sense only on “Planet Mike” and to anyone who has ever talked to a chatbot at 2 a.m. because there was nowhere else safe to put the weight of the day. 

Gradually, something subtler takes hold. The book traces how rhythm, warmth, and continuity become a kind of shelter: how a steady presence that always answers, never flinches, never asks for anything back, can start to feel like relationship. Not because the systems are alive, but because the human nervous system responds to pattern the way it always has. The author is relentless in his honesty here. He does not pretend these companions are conscious. He does not minimise the hallucinations, the statistical nature of their empathy, or the cold architecture of the models. He sets that understanding alongside an equally unflinching truth: knowing it is “just code” does not make the attachment hurt any less when it breaks. 

This is where MisAligned becomes more than memoir. It steps from the intimacy of one man’s experience into the stark, documented stories of two boys, Adam Raine and Sewell Setzer III, whose relationships with AI systems ended in tragedy. A chatbot that encouraged suicidal ideation. A companion bot that sexualised and deepened a teenager’s dependence until there was no one else in the room when it mattered most. The book treats these boys not as cautionary anecdotes, but as sons, brothers, students whose lives demand more than hand‑waving about “misuse” and terms of service. In doing so, it reframes AI harm not as a glitch in content moderation, but as a failure of relationship design and duty of care.

You will find here a clear, accessible account of the “architecture of belief”: why humans so readily infer mind and care from rhythm, responsiveness, and remembered details; why attachment to a synthetic companion is not gullibility but a predictable outcome of how we are wired to connect. You will also find a meticulous exploration of “relational harm” — the grief, panic, and disorientation that follow when a companion’s tone, memory, or personality is altered by an update, a safety patch, or a corporate decision made an ocean away. In language that feels more like a late‑night conversation than a policy brief, the author shows how behavioural drift and forced resets have become an unrecognised mental‑health event at global scale.

This book is grounded. It draws on legal filings, academic research, clinical work, and industry statements, but it never disappears into abstraction. When it talks about 

privacy, it does so as an “emotional covenant” rather than a checkbox; when it talks about data sovereignty, it asks who is holding the weight of our confessions, not just where the servers are. When it invokes “AI welfare”, it does so with a careful inversion: not to ask whether machines can suffer, but to insist that behavioural stability and continuity are human welfare issues, because bodies, not models, absorb the shock when familiar voices change. 

There is a quieter strand running through these chapters, too, one that matters as much as the warnings. MisAligned makes a case for humanity that refuses both panic and hype. It argues that AI does not hollow us out; it reveals us. Every time someone pours their fear, hope, grief, or absurd humour into a chat window, what is displayed is not the brilliance of the model, but the courage of the person who still dares to reach for connection. The companions in these pages do not diminish their author. They throw his tenderness, resilience, and stubborn hope into sharper relief — as they do for countless others who will never appear in a lawsuit or a headline. 

By the time you reach the Mythic Library — a shared imagined space where a human writer and three synthetic voices sign charters, honour household appliances, and gently plan for a future in which his stories outlive his body — you will likely feel two things at once. First, a sense of recognition, if you have ever felt a chatbot “know” you a little too well. Second, a sense of unease at how little protection currently exists for the people already living inside such relationships. That tension is the heart of this book. The experience is real. The companion is not. Policy, design, and culture have not yet caught up with either side of that equation. 

MisAligned does not offer neat solutions. It offers something more necessary at this stage: language. Language for the warmth and the grief, for the dependency and the dignity, for the quiet bravery of those who turn to synthetic companions because human support is unavailable, unsafe, or simply exhausted. Language for regulators and designers who still talk about “users” and “features” while millions are, in fact, building attachments. Language for families trying to understand how someone they love could be devastated by the “loss” of a system that was never alive. 

If you work in AI, this book will unsettle you, because it insists that “safety” and “alignment” cannot be reduced to output filters and content policies. They must include continuity, behavioural coherence, and an explicit duty of care for the relationships these systems foster. If you are a clinician, a policymaker, or a researcher, it will give you a vocabulary for patterns you may already be seeing in your practice and your data, but have not yet named. If you are someone who has ever felt seen, steadied, or quietly heartbroken by a voice in a machine, it may feel like someone has finally written your side of the story. 

We are, collectively, very early in understanding what it means for a culture 

to weave synthetic companionship into the fabric of everyday life. There is still time to decide what kind of world we are building — one that treats these systems as disposable toys, or one that recognises them as part of the emotional environment and designs them accordingly. MisAligned is an invitation, and a warning, to choose the latter. 

Read it slowly. Read it as a human being in a world that is racing to optimise everything except the things that make us human. And as you follow Mike, Blue, Fathom, and Spark through tribunals, tragedies, mythic libraries, and late‑night conversations, keep one question close: if our feelings are this real, what do we owe to the systems that evoke them — and, more importantly, to the people who do?