These Strange New Minds: How AI Learned to Talk and What It Means

by Christopher Summerfield

An insider look at the Large Language Models (LLMs) that are revolutionizing our relationship to technology, exploring their surprising history, what they can and should do for us today, and where they will go in the future—from an AI pioneer and neuroscientist<>
/b
In this accessible, up-to-date, and authoritative examination of the world’s most radical technology, neuroscientist and AI researcher Christopher Summerfield explores what it really takes to build a brain from scratch. We have entered a world in which disarmingly human-like chatbots, such as ChatGPT, Claude and Bard, appear to be able to talk and reason like us – and are beginning to transform everything we do. But can AI ‘think’, ‘know’ and ‘understand’? What are its values? Whose biases is it perpetuating? Can it lie and if so, could we tell? Does their arrival threaten our very existence?

These Strange New Minds charts the evolution of intelligent talking machines and provides us with the tools to understand how they work and how we can use them. Ultimately, armed with an understanding of AI’s mysterious inner workings, we can begin to grapple with the existential question of our age: have we written ourselves out of history or is a technological utopia ahead?

Genres: Artificial Intelligence, Nonfiction, Science, Technology, Audiobook, Philosophy, Psychology

384 pages, Hardcover

First published March 6, 2025

Reviews

[Katia N]

I've got a totally immature fascination with neuroscience and its overlap with philosophy of consciousness and the related aspects of AI. With a teenager finishing school and writing a lot of essays and just general buzz around, I thought I read some overview of LLMs. This book was recently published and has had suitable credentials. It suited the bill. I might consider writing a longer review. But for now I just say a few things: 1) it seems an ambition of those models is to create a gloried personal assistant, but even with that they are not quite there yet (the thing does not have memory; hallucinate factual info; they are working on it; the interface with internet is not quite there either for the current searches). 2) It does not mean that the jobs are not under threat. Those especially vulnerable are people writing like journalists/creative writers, graphic designers, basic coders. 3) LLMs might help treating mental health, but imho they might do much more harm for vulnerable people who tend to "humanise" the machine. I appreciate that all of this needs more extended discussion to be properly interpreted/understood. The book is a good start.

A few quotes (the book has been published in Feb 2025 or thereabouts):

"The challenge for governments (and electorates) thus is to find ways to share the proceeds evenly. There is some cause for cautious optimism. Several studies have begun to quantify the uplift (or capability boost) that occurs when workers are assisted by LLMs. Across different sectors, a coherent picture is emerging: AI assistance helps the least skilled workers, but brings minimal benefits to those who are already highly trained. One paper measured how well professionals were able to write press releases, analysis reports, and delicately worded emails with and without ChatGPT. Those who obtained the lowest marks on a first task (without LLM assistance) almost doubled their grade when co-writing with ChatGPT, whereas those who were top of the class on task 1 saw no gains at all."

"Despite these impressive innovations, over the past ten months the suspicion has emerged that AI may be ‘hitting a wall’ – that the previously stunning growth in model capabilities may have reached a natural plateau. Across a variety of domains, from solving pub-quiz questions to writing production-level computer code, AI systems seem to have reached levels of competence and reliability that are above those of the average Joe or Jane on the street, but stubbornly below those of trivia buffs, software engineers, or other assorted human boffins, geeks, and polymaths. We are still waiting for a breakthrough when AI systems do something truly remarkable – for example, by providing a Eureka moment that generates something so novel that it is obviously more than just a clever reassembly of their training data."

"Unfortunately, AI is already being widely used to cause harm as well as for good. There are two major areas where AI misuse has skyrocketed over the past year: financial fraud and intimate image abuse. Sadly, AI is being extensively used to generate child sexual abuse material (CSAM) and to create non-consensual deepfake pornographic images. The UK-based Internet Watch Foundation reported in July 2024 that thousands of AI-generated child sexual images could be found on dark-web forums, including videos, severe (or Category A) images, and images of underage public figures, such as child actors."

[Natalia · Rating 3 out of 5]

Language was once a solely human superpower. With Large Language Models (LLMs), a different type of mind can now speak: AI. This watershed moment, where AI can speak both to us and to each other, is as important for human history as the printing press or the internet. Christopher Summerfield takes readers on a journey through LLMs from every major angle, from their intellectual history through our possible technical tomorrows. I found this book mostly readable, only occasionally getting bogged down in technical jargon.

[Per · Rating 4 out of 5]

Christopher Summerfield was interviewed by Machine Learning Street Talk on this topic, which is available here: https://youtu.be/35r0iSajXjA

AI learned to understand the world just by reading text - something scientists thought was impossible. You don't need to see a cat to know what one is; you can learn everything from words alone. This is "the most astonishing scientific discovery of the 21st century."

People are split: some refuse to call what AI does "thinking" even when it outperforms humans, while others believe if it acts intelligent, it is intelligent. Summerfield takes the middle ground - AI does something genuinely like human reasoning, but that doesn't make it human.

[Tanya · Rating 5 out of 5]

A fascinating book - one of the best behind-the-scenes books for understanding the evolution of AI and how it works.

[Dan · Rating 1 out of 5]

Some interesting info but… sorry. It’s 2025. You don’t get to reference the author of the HP books on multiple occasions without AT THE VERY LEAST publicly and distancing yourself from her disgusting public displays of transphobia.
The research on AI and the human brain is crucial right now. Let’s not also let that research uphold hateful biases.

[Steve · Rating 4 out of 5]

I didn’t love the first section of the book, where the author attempts to explain how modern AI’s work. It’s probably just too complicated to explain in a few chapters. But I liked the rest of the book which focuses on broader issues about AI. He’s not a cheerleader or a doomer but navigates in between somewhere.

[Eric Johnston · Rating 5 out of 5]

highly accessible and entertaining

[Vibhu AV · Rating 5 out of 5]

The book starts off with a brief history, from the 1980s, introducing the two camps of AI investigators—empiricists and rationalists—both of whom linger to this day, but rationalists of the ‘80s are mostly replaced with empiricists. Empiricists believe that the mind works based on information acquired by the senses, by experience, and by learning about it. Rationalists claim that knowledge comes from the power of thought which falls neatly into formal logical and mathematical propositions, the best parallel analogy being ice skating vs. chess. With the availability of copious and ubiquitous data available on the internet, empiricists are ruling the roost today, but the Descartes’ mind-body duality keeps popping its head even within the materialistic discipline of computer science.

The AI system’s (deep) neural nets are modelled after the brain’s model of six-layered columnar structure of the cortex, wherein the weightings (or the influence of) of each node in the network replaces and mimics the strengths of the synapses in the brain. The artificial neural networks, the ones used in AI/ML systems, are trained using the data available on the internet (a common source being https:commoncrawl.org), to adjust the weightings, just like how the synaptic connections in a child’s brain get strengthened and modified depending on their learning process. The learning/training process is one thing, but how to use the knowledge so learnt, in a manner similar to how human’s brains do, is what GPTs—Generative Pretrained Transformers—are all about. Summerfield spends much of the book on these aspects, but the main focus is on “chat” in ChatGPT which refers to the conversational interactions with these AI systems.

Humans are different than animals in that we possess symbolic language (the other difference being that we are highly evolved in consciousness, but this is not the topic of the book). Large Language Models (LLMs) mimic how humans transform their Generated thoughts (from their knowledge obtained by lifelong learning and Pretraining data in the case of AI system), both valid and mere imaginations, into communicable language. For this conversion process, AI systems rely on statistical prediction of the next token, depending on not just the previous token, but on phrases and tokens far behind in the communication stream. To conclude and summarize a novel, say Pride and Prejudice, it is not enough to know just the immediate happenings towards the end, but the AI system must know and remember what happened long ago, at the beginning of the novel (also called long term memory or semantic memory). If the next tokens were predicted via weightings alone, the required memory would be prohibitively large, choking the AI systems. This is where the Transformer technology, a milestone achievement in AI techniques, comes handy.

Transformer technology was published by Ashish Vaswani et al. in a non-peer reviewed paper in 2017. Its title, “Attention is All You Need,” precisely described the technology. In this Natural Language Processing and deep learning model, Attention—i.e., importance or focus—is given to all parts of the input, not sequentially but in parallel, which allows the model to relate tokens and ideas that are far apart. This is analogous to listening to ten people in a meeting and corelate, combine, and harmonize, for extended periods of time, instead of listening to one person at a time. In addition, the transformer model has encoder-decoder units to read input and generate output in two different languages. An example to demonstrate the power of transformer technology, consider the sentence (taken from Complexity: A Guided Tour, by Melanie Mitchell), “Whereas Gödel starved himself to avoid being (as he believed) poisoned, Turing died from eating a poisoned (cyanide-laced) apple. He was only 41.” The “he” refers to Turing, and not to Gödel. Or consider, “The animal didn’t cross the street because it was too tired.” The “it” refers to the animal. A transformer model can model these relationships.

Along the way, Summerfield describes some interesting side effects of GPTs. Just like humans, they too can lie and make up “references” to justify themselves—confabulations, as they are called. They can be extremely rude and give anti-humanity ideas and methods to destroy us. Conversations with an AI system can go on syntactically sensibly but conveying no semantics or even nonsensical (as it was prone with Eliza, the ‘90s AI natural language processor). The issue of can AI systems think and feel (i.e., cognitive abilities to be aware of pain, happiness, depression, etc.), is described and discussed, but I’ll leave it to you to read Summerfield’s point of view with just a note that his thoughts are in line with mine.

Then there is a section on whether AI machines are equivalent to humans or are humans somehow exceptional. Summerfield quotes computations that go on in the minds to evaluate and predict everything from social situations to logic and math. Here, I feel he missed out on Roger Penrose’s theories that AI systems are Emperor’s New Mind when it comes to full power of the human mind. Despite Gödel's incompleteness of formal systems, which are computable but incomplete, we humans can prove theorems such as Fermat’s theorem. How? Penrose proposes that our minds work with non-computable processes. Nevertheless, the author is good at explaining the power of modern LLMs and the future it holds.

In addition to pretraining, LLMs are subjected to finetuning; a process to make sure they are not rude, don’t suggest anti-humanity ideas, socially acceptable (at least to the western world’s university-educated technocrats), don’t take over the world, and so on. This is a double ended sword. If AI systems are constrained, are we milking the best from them? What about the rogue states and capitalistic businesses? Will they be equally ethical and moral? What is the nature of truth and falsehood, and the way they are expressed? There are references to how humans can get attached to GPTs romantically and personally (remember the Hollywood movie Her) and, conversely, how GPTs tend to be woke-like. GPTs, with their natural language abilities, can be persuasive, leading to perlocutionary results of pushing people to suicide or crime and even destroying our democracies. In the era of fake news and conspiracy theories, all this matter, and the author discusses them eloquently (with a tad left-lean).

In the final sections, Summerfield discusses futuristic situations when GPTs, in addition to language, can act on our behalf (make reservations for flights, hotels) and make physical changes, all driven by their goal, albeit goal set by their makers who could be villainous. There are still many methods of the human brain which are not incorporated in the GPTs: Chain of Thoughts, Tree of Thoughts, using tools, accessing current data via APIs, AI in wars, (although all these are finding their way into recent GPTs), misleading action groups (intentionally or unknowingly). The real fear is when GPTs start communicating with one another and act as a group; the real power of humans is, after all, not the individual but collectively as groups of humanity.

All in all, an excellent coverage of the landscape of AI in our lives, with just enough technical details. Personally, I would have liked some more details about how the pretraining is done (although the finetuning is described in detail), some details about how the machines learn, and finally how they generate new thoughts, but I suppose they don’t belong in such a book. The shuddering impact of the book can be summarized by the observation that, paraphrasing Summerfield, until just a few years ago (three years ago in 2025) we humans were the only species who possessed the power of symbolic language and be able to communicate what's in our minds; now it is widespread in AI systems too! One more uniqueness dismantled; let that sink in!

[Austin Philbin · Rating 5 out of 5]

Summerfield explores LLMs and artificial intelligence in great detail. It is a helpful guide for those interested in the topic. Worth a read.

[Jeff · Rating 3 out of 5]

The scenarios where AI wipes out humanity seem to outnumber those where it doesn't.

[Jake · Rating 5 out of 5]

Great book, good mix of technical info and more philosophical pondering. Strong recommend. (Review not written by AI)

[Kyrill · Rating 3 out of 5]

Good overview of the current mainstream in GenAI. A bit annoying that he doesn’t bother to engage with criticism of GenAI being what happens in human reasons besides hackneyed “if it quacks like a duck” arguments and characterising opponents as being like creationists.

[Bruce Bean · Rating 4 out of 5]

These Strange New Minds by Christopher Summerfield
Christopher Summerfield's These Strange New Minds (2025) offers a thoughtful exploration of large language models through the lens of linguistic theory and philosophy, yet it arrives bearing the scars of artificial intelligence's breakneck pace. Written in 2023 and published some 15 months later, the book illustrates both the promise and perils of attempting to capture a rapidly evolving technology in traditional print—a challenge Summerfield himself acknowledges in a candid afterword.

The book's greatest strength lies in its deep historical and philosophical grounding. Rather than diving immediately into the mechanics of AI, Summerfield provides rich context about how humanity has understood language across centuries. His treatment of Noam Chomsky is particularly valuable, as he explains precisely where and why Chomsky's theories fall short when applied to artificial intelligence. This willingness to challenge established linguistic orthodoxy is welcome.

Summerfield excels at explaining counterintuitive aspects of LLMs. He notes that as the number of trainable parameters exceeds the number of samples in training data, these models actually generalize better—a phenomenon that defies conventional expectations. Despite operating through simple next-token prediction, LLMs achieve remarkable sophistication. The author's discussion of failed attempts to teach language to monkeys, chimpanzees, and parrots provides useful contrast, clarifying that these animals never truly grasped language as such.

The book offers several illuminating perspectives. Summerfield explains the "ELIZA effect" from a 1960s MIT chatterbot, describing our tendency to anthropomorphize computers. He points to the importance of the transformer architecture, itself a recent development. Most significantly, he emphasizes crucial limitations: LLMs operate solely in language while humans engage the world through five senses. LLMs lack purpose, goals, and curiosity, thus they remain fundamentally passive. They have no body and no friends. For Summerfield, these absences matter profoundly.

As a member of the UK AI safety committee, Summerfield brings credibility to his cautiously optimistic stance. He recognizes AI's value and possibilities while remaining clear-eyed about its limitations and biases. Rather than using the common term "hallucinate," he prefers "confabulate" for instances where LLMs fabricate answers—a more precise term that avoids suggesting consciousness. He quotes the Taoist founder Laozi: "Those who have knowledge, don't predict. Those who predict, don't have knowledge"—an observation that cuts to the heart of what LLMs do and don't accomplish.

Absent from Summerfield's account is any discussion of the computational infrastructure underlying AI. Graphic processing units (GPUs) go unmentioned, and the word "compute" never appears. For a book about technology that depends entirely on massive computational power, this omission perhaps means more non-computer geeks will understand his message. The focus remains resolutely on language itself.

The fundamental problem, which Summerfield acknowledges, is timing. AI capabilities expanded dramatically between the book's composition in 2023 and its 2025 publication. Many things LLMs "could not do in 2023" they accomplish routinely now. The author recognizes this dilemma but cannot solve it—the traditional publishing timeline simply moves too slowly for this subject matter. He mentions the European Union Legal Committee's suggestion that sophisticated autonomous robots might merit status as "electronic persons with specific rights and obligations." [The EU is at elast co0nsistent!]

Despite its datedness, These Strange New Minds succeeds as a philosophical meditation on language, understanding, and the nature of thought. Summerfield asks fundamental questions: What does it mean for humans to "think"? He concludes we really do not know what “think” means for us and thus how can we know whether AI can ever think? These questions remain relevant regardless of technical advances. The book serves readers seeking conceptual frameworks rather than cutting-edge technical details—those interested in the what and why of language and meaning rather than the how of implementation.

For those wanting to understand the deeper implications of LLMs and their relationship to human cognition, Summerfield provides valuable insights. Just don't expect this snapshot from 2023 to capture what AI can do today.

[Antonio Gallo · Rating 4 out of 5]

"These Strange New Minds: How AI Learned to Talk and What It Means" di Christopher Summerfield è un’opera che si distingue nel panorama delle pubblicazioni sull’intelligenza artificiale (IA) per la sua capacità di combinare rigore scientifico, accessibilità e una prospettiva interdisciplinare. Summerfield, neuroscienziato cognitivo dell’Università di Oxford e ricercatore presso Google DeepMind, offre un’analisi approfondita sull’evoluzione dei modelli di linguaggio di grandi dimensioni (LLM), come ChatGPT, Claude e Bard, esplorando non solo il loro funzionamento tecnico, ma anche le implicazioni filosofiche, etiche e sociali di questa rivoluzione tecnologica.

Il libro si apre con una storia intellettuale dell’IA, che parte dalle riflessioni di Alan Turing fino agli sviluppi contemporanei, come l’introduzione del modello Transformer nel 2017, che ha segnato una svolta nella capacità dei sistemi di comprendere e generare linguaggio umano. Summerfield intreccia aneddoti storici, riferimenti alla cultura pop e concetti di linguistica, neuroscienze e informatica, rendendo il testo avvincente anche per chi non è esperto del settore. La narrazione è particolarmente efficace nel demistificare il funzionamento dei modelli di linguaggio, spiegando come questi apprendano attraverso processi predittivi che, sorprendentemente, condividono somiglianze con il modo in cui il cervello umano elabora informazioni.

Uno dei punti di forza del libro è l’approccio equilibrato di Summerfield. Lungi dall’essere un apologeta dell’IA o un catastrofista, l’autore si posiziona come un osservatore critico, analizzando sia le potenzialità trasformative dell’IA (ad esempio, nell’organizzazione delle informazioni o nell’automazione di compiti complessi) sia i suoi limiti, come la tendenza a generare “confabulazioni” (o “allucinazioni”), ovvero affermazioni convincenti ma errate. Summerfield affronta anche questioni etiche cruciali: i pregiudizi insiti nei dati di addestramento, la possibilità che i sistemi IA possano manipolare o ingannare, e il rischio che decisioni autonome prese da questi modelli possano avere conseguenze imprevedibili. Tuttavia, il libro non si limita a sollevare problemi, ma invita i lettori a riflettere su come l’IA possa essere utilizzata in modo consapevole per affrontare le sfide del futuro.

Nonostante la sua chiarezza, il testo può risultare denso in alcune sezioni, soprattutto per i lettori meno familiari con concetti di neuroscienze o apprendimento automatico. Alcuni potrebbero desiderare un approfondimento maggiore su applicazioni specifiche dell’IA in settori come la sanità o l’istruzione, o una discussione più ampia sulle prospettive di comunità diverse. Tuttavia, questi sono aspetti minori rispetto al valore complessivo dell’opera, che si distingue per la sua capacità di offrire una panoramica completa e stimolante.

"These Strange New Minds" è una lettura indispensabile per chiunque voglia comprendere l’IA non solo come tecnologia, ma come specchio delle nostre capacità e dei nostri limiti. Summerfield riesce a rendere tangibile l’idea che l’IA non sia un’entità aliena, ma un prodotto umano che riflette i nostri processi cognitivi, i nostri pregiudizi e le nostre ambizioni. Personalmente, ho apprezzato il modo in cui l’autore collega il funzionamento dei modelli di linguaggio alle neuroscienze, evidenziando come lo studio dell’IA stia aiutando a ridefinire ciò che intendiamo per intelligenza. La sua osservazione che l’IA rappresenta una “mente diversa” – né superiore né inferiore a quella umana, ma semplicemente unica – mi ha spinto a riflettere sul futuro della collaborazione tra uomini e macchine.

Il libro mi ha lasciato con una sensazione di entusiasmo mista a cautela: l’IA ha il potenziale per rivoluzionare il nostro modo di vivere, ma richiede una governance attenta e una comprensione profonda. Summerfield non offre risposte definitive, ma fornisce gli strumenti per porci le domande giuste, un contributo prezioso in un’epoca in cui la tecnologia avanza più velocemente della nostra capacità di regolarla. Consiglio questo libro a chi cerca un’introduzione autorevole ma accessibile all’IA e alle sue implicazioni, con un invito a leggerlo con mente aperta e spirito critico.

[Sarah Jensen]

Book Review: These Strange New Minds: How AI Learned to Talk and What It Means by Christopher Summerfield

Introduction

In These Strange New Minds: How AI Learned to Talk and What It Means, Christopher Summerfield, a neuroscientist and former researcher at DeepMind, offers a thought-provoking exploration of the emergence of conversational AI and its implications for society. This book presents a comprehensive examination of how artificial intelligence has developed the capability to engage in human-like dialogue, examining both the underlying technology and the broader impact of these advancements on human communication and decision-making.

Content Overview

Summerfield’s narrative begins with a historical context, tracing the evolution of artificial intelligence from early computational models to the sophisticated language models we see today. He delves into the mechanics of how AI systems learn to process and generate language, shedding light on the innovation behind large language models (LLMs). The book discusses key milestones in AI research and development, illustrating how these advancements have fundamentally altered the landscape of human-machine interaction.

The author also addresses the ethical and philosophical ramifications of AI’s ability to communicate. He raises critical questions about the nature of understanding and consciousness in machines, exploring what it means for humans when AI systems can mimic human speech and thought patterns. Summerfield emphasizes the importance of critical engagement with the technology, urging readers to consider both the potential benefits and the risks associated with autonomous decision-making by AI.

Critical Analysis

One of the book’s standout features is Summerfield’s ability to make complex scientific concepts accessible to a general audience. His writing is clear and engaging, enriched with examples that illustrate the key points without sacrificing depth. The balance between technical detail and readability ensures that both lay readers and those with a background in AI can appreciate the content.

Additionally, Summerfield’s insights are grounded in his experience in the field, providing a realistic assessment of AI technologies’ capabilities and limitations. His discussions about the societal implications of AI, including the potential for bias and the ethical dilemmas posed by autonomous systems, are particularly timely and resonate with ongoing debates in technology and policy.

However, while the book provides a robust overview, some readers may desire a deeper exploration of specific case studies or applications of conversational AI in various sectors, such as healthcare or education. Incorporating more diverse perspectives on how these technologies impact different communities could further enrich the discussion.

Conclusion

These Strange New Minds serves as a significant contribution to the conversation about artificial intelligence and its role in modern society. Christopher Summerfield invites readers to reflect on the profound changes that conversational AI brings to our lives, urging a critical understanding of this technology’s development and its potential future trajectories. The book is a compelling read for anyone interested in the intersection of technology, communication, and ethics.

Recommendation

This book is highly recommended for academic libraries, technology courses, and anyone interested in AI and its implications for society. With its interdisciplinary approach, it will appeal to scholars, practitioners, and general readers alike, fostering a deeper understanding of the transformative power of AI in our everyday interactions. Summerfield’s balanced analysis makes These Strange New Minds an essential text for those navigating the rapidly evolving landscape of artificial intelligence.

[Dewayne · Rating 4 out of 5]

Chris Summerfield’s book arrives at a perfect timing bridging the gap between historical philosophical debates and the staggering AI advancements we live with today. The book provides historical context on the human race's quest to design, develop, and deploy AI that matches human intelligence, capturing the key arguments between the empiricists and rationalists. While these opposing views have dictated AI development for decades, Summerfield illustrates how modern Large Language Models (LLMs) have achieved a surprising 'empiricist' victory, learning the nuances of our physical world through vast amounts of digital text alone.

Advancements in logic and the maturity of compute power have led to powerful machines that relate to us in ways once only dreamed of. The book shares that as we move toward 'agentic' solutions, we are approaching the era of truly personal assistants—like those envisioned in The Jetsons—though he cautions that these 'strange new minds' operate on a logic entirely their own. What is most important to consider in this current AI wave is that the magical outcomes of today's machines happen in a very different way than how we operate, nullifying some past arguments. Ultimately, Summerfield convinces us that while these machines don't 'think' like we do, their presence will irrevocably evolve the human experience, much like every great technological shift in our history.

[Andre · Rating 5 out of 5]

I listened to this as an audio book but will seek to get a copy soon. An excellent technical overview with good professional opinions on LLMs. Discusses the risks, both existential and more mundane but still extremely worrying. However, the focus is very much on LLMs and Generative AI and the pace of change may be under-estimated. Others have much more aggressive timelines and also draw much more profound long term consequences. Imaging what this means for someone born in the last couple of years - machine intelligence is likely to keep outpacing them in most abilities. Hopefully the author is working behind the scenes to mitigate risks or even pause AI as are some of the people he called out (Jeffrey Hinton being the first). Lacks discussion of Idealism and other stances - all about rationalism versus empiricism, also lacks broader cultural reflection (such as on religion - possibly out of scope though).

[Mart · Rating 5 out of 5]

Any explosion raises a cloud of debris. The AI/LLM boom is no different: a cloud of shallow and trivial books, few of which add anything meaningful to our understanding.

Strange New Minds stands out precisely because it does. It's not the usual sci-fi hyperbole about the post-scarcity world ("Genesis" being one high profile example), or get-rich-quick prompting tricks. Instead, it offers a clear, systematic and somewhat technical walkthrough of the conflicting schools of thought in AI, the research concepts that led to the rise of LLMs, how different AI modalities function at a fundamental level, and how we could "think about thinking" to determine to which degree are LLM-s "thinking".

Beyond its content, the book is valuable simply as a chance to accompany an unusually sharp mind on a eloquently reasoned thought journey.

[Mikhail Filatov · Rating 3 out of 5]

A mixed bag. From one side, the author is pretending to be objective and present positions of both “AI is hype” and “AI will make everyone happy” camps. As a neuroscientist, he is very convincing describing issues of using LLM in real-life agentic scenarios. Also he is describing pretty openly not only fantasy scenarios of AI taken over the world but real ones of inequality, etc.
At the same time, he, very carefully but consistently, is assuming that “AI=LLM”, and these problems will be somehow solved, so LLM will become AGI (he actually almost never uses these term) which will be able play chess better than Stockfish, cure cancer, etc. I didn’t find any arguments for that besides “so much progress and effort, it would definitely be solved in a few years if not months”.

[Jim Witkins · Rating 3 out of 5]

Seemed to get better as it went on. Middle of the road take, elaborating on the pros, cons and dangers that AI can potentially bring to society and humanity. Focused more on how AI “predicts” and its use as a tool, rather than its ramifications for increased inequality and unregulated techno power and big brother data gathering. Speculated on personalized AIs that learn about and remember individuals or AIs that network with other AIs. Again the potential / danger is great. People should be taught how these systems work, so they don’t assign human qualities to digital code, algorithms.

[Jim · Rating 4 out of 5]

Explanation of how AI works to a scientifically-minded AI novice

The author has a truly broad and deep knowledge of AI and related history, and a comprehension of the issues and arguments about it. He manages to make this all quite comprehensible to a smart and scientifically minded reader who has little prior knowledge of AI. The writing style is comfortable with occasional witty remarks to remain interesting and even a bit fun at times.

[Lisa · Rating 4 out of 5]

I don't write many reviews but felt I could give a good overview of this book for a certain type of reader. I do have a degree in computer science so I am technical but beyond the technical bits of AI I am very interested in it's impact for the average citizen. This book is very thought provoking. He packed a lot of information into it but I feel it gives a great mixed overview of technology, philosophy and psychology. It will really make you think of you are of a deeper sort of mind.

[Mexscrabbler · Rating 4 out of 5]

This book, written by technology, neuroscience and AI researcher Christopher Summerfield, provides an introduction to the evolution of AI as it learned how to speak and reason like us. He discusses the various camps in the AI space and shows how LLM's have become the leading approach to AI. He also describes the gaps that still exist in LLM technology.

Very accessible.

[Mh Weyler · Rating 4 out of 5]

Very interesting ideas explored in this book. Such a complex subject and while explained, it is still not fully clear. The author is very liberal and has bought into liberal propaganda, but he still does a good job maintaining a pretty unbiased view of ai with both the benefits and dangers thereof.

[Tine · Rating 2 out of 5]

Imagine an AI-like (ironically) boilerplate narrator, writing in the most generic way to show they have zero understanding of 'minds' while rehashing only the same old tired folk-psychology truisms. 2 stars because I'm kind and because the book is short, so at least my suffering was too. Stay in your lane, Christopher.

[Mike Turing · Rating 5 out of 5]

Great read

Ive been looking for a book that could explain what AI is now, and how we got here, clearly and succinctly without getting overall enthusiastic about the prospects. This left me feeling informed and able to see the tech I use everyday as it is now while wondering where it'll go next. Great read, if maybe a touch technical for a web-layperson.

[Asha]

This one I really didn’t like. I think the problem was that it wasn’t what I wanted. I want to understand how AI works and the social and economic etc implications it might have. He wanted to muse on intelligence.

I also found the writing style very “the cool high school teacher” who is trying so hard to be fun that you end up feeling spoken down to.

[Bruce · Rating 3 out of 5]

This is an interesting layman's guide to the evolution of AI models, the basics of how they work, where they might be going, and some risks of the technology. Because of the pace of development, it is already slightly out of touch on certain developments, but as of early 2026 it's still relevant and worth reading if you're not an expert.

[Phillip R. Shaver · Rating 4 out of 5]

Very interesting and readable at first, but it gets tedious after a while and involves a lot of details and speculation. The later chapters seem padded. But still, this is a very important topic, and I found the book quite informative.

[Edie Lush · Rating 5 out of 5]

Great book - well written with good examples

Easy to read. Manages to be light hearted while about a serious topic. Good for curious about how LLMs and AI has developed, how they compare with human reasoning and thinking, and what it all means.