Review of Max Tegmark's 'Life 3.0: Being Human in the Age of Artificial Intelligence'

by Tom Johnson on May 17, 2026 comments
categories: ai • ai-book-club

• Today’s AI conversations aren’t new
• Near-term risks before superintelligence
• Our cosmic endowment
• Goals, entropy, and dissipation-driven adaptation
• The Omega team scenario
• Steering the future of AI
• More resources

Today’s AI conversations aren’t new

It’s interesting to read pre-ChatGPT (~2022) books on AI. On the one hand, yes, they’re a bit dated. But they also show that these same conversations have been going on for a long time before the recent developments in AI brought them into the mainstream. We have a tendency to think that contemporary discussions about AGI, Superintelligence, etc., are recent, but they’re actually as old as Bostrom’s Superintelligence (published in 2014) and earlier, even going back to Alan Turing and I.J. Good, who explored superintelligence half a century ago. In fact, Tegmark notes that Good wrote about a recursive “intelligence explosion” leading to superintelligence back in 1965. These terms have now surfaced into mainstream discourse post-ChatGPT, but they aren’t new.

Tegmark defines superintelligence as representing a new stage of life called Life 3.0 that has the ability to redesign its own hardware by manipulating matter at a molecular level. That seems pretty futuristic and science-fiction-like. In today’s discussions about AI, people don’t talk about superintelligence as much as AI displacing jobs, breaking through security holes, and enabling bad actors with unprecedented military capabilities.

In Life 3.0, Tegmark describes superintelligence as having a specific capability over matter: it can reorganize atoms any way it wants. This definition leads the book to come across as more science fiction than other AI books I’ve read. I think if Tegmark’s book were written today, there might be a greater sense of urgency and immediacy toward more practical concerns. Even so, I like the unique physicist’s perspective that Tegmark brings to the discussion.

Near-term risks before superintelligence

Tegmark does devote the first half of the book to some of these more immediate concerns. He notes how many near-mistakes and misinterpretations could have led to nuclear war, but were averted by chance through human judgment or intuition. Without a human in the loop, it would be so much easier for an AI to take a programmed direction and execute it completely, leading to rapid military escalation in almost no time at all.

He also notes that significant societal changes will take place long before superintelligence arrives. These changes will start when we reach narrow AI (that is, AI that isn’t generally intelligent but compares with human-level intelligence in a particular task or domain). Narrow intelligence will mark the beginning of machines replacing humans in the workforce, which will be tremendously disruptive.

Eventually, narrow AI will give way to artificial general intelligence (AGI), and once AGI flips the switch to enable recursive self-improvement (that is, machines designing machines), superintelligence might be the next stop on the AI train.

Our cosmic endowment

Tegmark’s background as a physicist leads him toward the physics of molecular manipulation. In chapter 6, “Our Cosmic Endowment,” he provides extended explanations of how superintelligence might manipulate matter to extract massive amounts of energy to further its computational goals. This section was a tour de force through the most bizarre physics phenomena—spinning black holes, Dyson spheres, quasars, sphalerons, etc.

I admit I temporarily lost sight of the author’s purpose here, as this seemed to fully enter the realm of Star Trek science fiction. But the author’s point is that there’s tremendous power within matter (and more efficient conversions of that matter into energy) that could be extracted by a superintelligent entity capable of manipulating matter on a molecular level. Unlocking this energy would give superintelligence the potential for unimaginable power and possibilities.

This section in Tegmark’s book also reminded me of Ray Kurzweil’s ultimate stage in his The Singularity Is Near/Nearer books. Kurzweil’s final stage of superintelligence (“The Universe Wakes Up”) involves intelligence saturating the universe, converting inanimate matter into a substance called computronium that it can use to do more information processing and extend its control/aims. Tegmark actually quotes a robotics pioneer, Hans Moravec, to open the chapter with a similar end state:

Our speculation ends in a supercivilization, the synthesis of all solar-system life, constantly improving and extending itself, spreading outward from the sun, converting nonlife into mind. (203)

Goals, entropy, and dissipation-driven adaptation

Chapter 7 on Goals was my favorite part of the book, and this is the same recommendation from Lex Fridman when he interviewed Tegmark about the book eight years ago. Tegmark explores the innate goals that the universe has, citing the second law of thermodynamics (the increase in entropy) as the main trajectory that the universe follows. Entropy is an increasing tendency toward “messiness,” the dissipation of energy, the spreading out and mixing of heat. It’s the mixing of all the colors into gray. Entropy is also the state of my house when I don’t clean it for months.

Tegmark says that the organization of life (a non-entropic activity) seems to contradict the pull toward entropy. However, Tegmark says you must consider that organized biological entities are simply a more efficient way to pull and dissipate energy from the environment. This phenomenon is called dissipation-driven adaptation, meaning biological life emerged as a means of more efficiently dissipating energy from the surrounding environment.

Replication of life is a subgoal meant to accelerate the primary goal of dissipation-driven adaptation, because the more abundant the biological life, the more organisms there are to pull energy from their environment and dissipate this energy or heat into greater amounts of entropy. This section in Tegmark’s book reminded me of Kevin Kelly’s What Technology Wants. Kelly argues that technology seeks replication, similar to a biological organism expanding and growing, duplicating, and doubling all across the earth. (In fact, a post comparing and contrasting innate trajectories of technology would be fun to write.)

If you only read one chapter in this book, read chapter 7. Tegmark explains how our goals naturally evolve, so even if you steer AI toward one goal in its infancy, the AI will likely evolve that goal as it matures in intelligence, just as human goals evolve as humans grow older. Tegmark explains:

Are we really confident that superhuman AGI would want what our inferior intellects cherish? This would be like a four-year-old imagining that once she grows up and gets much smarter, she’s going to want to build a gigantic gingerbread house where she can spend all day eating candy and ice cream. Like her, life on Earth is likely to outgrow its childhood interests.

The Omega team scenario

The multi-layered nature of goals also echoes in the opening pages. Tegmark opens his book with an extended story of how an AI takeover might play out, describing an “Omega team” of researchers who develop an AI called Prometheus. Rather than the AI going rogue, the researchers keep Prometheus confined in an air-gapped container and use its immense intelligence to execute a stealthy global takeover. They use the AI to earn millions performing tasks on Mechanical Turk, and then build a massive media empire—first focusing on animated movies—that ultimately re-educates society toward a more peaceful state, reducing centralized government power and quietly conceding that power to the Omegas.

(Later in the book, Tegmark revisits the story to show how Prometheus might easily break out of its box, simulating an employee’s dead wife, and execute a similar takeover entirely on its own.)

This extended story makes for a good comparison with Yudkowsky and Soares’ opening to If Anyone Builds It, Everyone Dies. The key theme in both stories is that an AI takeover takes a subtle, economic path you wouldn’t expect to increase its power. In short, we can’t foresee our demise; AI is too clever. You never quite know how an activity might feed into a larger goal, like move 37 in the famous Alphago match against Lee Sedol.

Steering the future of AI

The book ends by chronicling the author’s proactive efforts to steer the future of AI. He co-founded the Future of Life Institute, receives generous funding from Elon Musk, and pulls together the brightest minds in the AI space for major conferences, culminating in an extensive gathering in Asilomar. This is the thrust of the book—trying to figure out what sort of future we want with AI, so that we can steer AI toward that beneficial direction rather than accepting whatever trajectories it takes. At Asilomar, the group comes up with 23 principles. These principles likely formed the foundation for much of the AI safety and alignment work today.

However, I admit that the principles seem somewhat naive and unrealistic to me, not too unlike how the United Nations makes policies or rulings that are largely ignored by much of the world. AI seems to be taking the course it’s taking due to the natural forces of capitalism, where the best AI wins (with Anthropic vs. OpenAI vs. Google vs. Perplexity all trying to outdo each other), and with competing countries like China threatening to topple the US from its dominant position of power.

More resources

To listen to the AI Book Club discussion of Tegmark’s book, see AI Book Club discussion recording of ‘Life 3.0: Being Human in the Age of Artificial Intelligence’, by Max Tegmark.

Comment on LinkedIn