In "Life 3.0" Tegmark delves into artificial intelligence, a topic that has generated an incredible amount of junk about superintelligence and singularity. His take on artificial intelligence is possibly the most competent and (excuse the pun) intelligent in recent years.

He starts out with a simple definition of the two stages of life: non-intelligent life (life that can only do what it is genetically programmed to do) and intelligent life (life that can learn to do things for which it was not programmed). Life 2.0 beings can learn to perform tasks that they didn't know how to perform when they were born. Furthermore, Life 2.0 as instantiated in Homo Sapiens is capable of storing information for other humans to use, thereby exponentially increasing what humans can learn over the course of their lifetime. The field of Artificial Intelligence has been working on machines that learn since its very beginning, and the current generation of neural networks has been capable of "learning" to do a lot of things (technically speaking, they didn't learn anything by themselves: they were "trained" by humans). He then calls "Life 3.0" life that can "design" not only its software (the tasks it can perform) but also its hardware.

Tegmark wastes a chapter telling us what is coming soon out of A.I., in various fields from finance to health care, with the inevitable litany of the jobs that will be lost (five years later, the USA has record-low unemployment, and it's easy to calculate that A.I. has stolen exactly zero jobs worldwide while creating quite a few high-paid jobs for software engineers).

(My take on the past, present and future of A.I.: "Intelligence is not Artificial").

Chapter 4 ("Intelligence Explosion?") is where his scientific and philosophical meditations get more interesting. Tegmark wonders whether someone could use super-intelligence to obtain absolute domination of the world, creating the perfect surveillance state and the perfect police state, or if the super-intelligence itself could take over the world. (Note that the "security bracelet" that Tegmark envisioned in this 2017 book was implemented in 2020 in China to control the covid pandemic) In case the transition from subhuman intelligence to super-human intelligence happens rapidly, the likely outcome is a single super-intelligence. The case in which the transition happens slowly leads to the possibility of many super-intelligences cooperating in a Nash equilibrium, which is what we have today with nations and corporations competing with each other. The latter case is more likely to happen than the former because of the limit imposed by the speed of light on the transmission of information. Given that the advent of a super-intelligent machine is possible, Tegmark wants us to start thinking about what kind of world we want. That's actually an important consideration: after thousands of years of philosophy and politics, the human race doesn't quite have a clear goal in mind. Most of human history has been spent fighting each other, trying to conquer and dominate, and defending ourselves from others who wanted to conquer us and dominate us. We've had precious little time to think about what we want the world to be like in a thousand years or in a million years. (And probably, if you focus, you don't even really care for the world a thousand years from now: sadly, for most people survival "is" the meaning of life).

In particular, nobody has polled people to see if they want a super-intelligence to be created. It is difficult to choose between scenarios in which a) no intelligent machine help humankind, b) several intelligent machines perform useful work for us, c) a super-intelligent machine replaces human politicians and runs a better society for us. If we choose b), at some point we'll have all the intelligent machines that we need, and at that point we'll have to stop technological progress. If we choose c), and we assume that the machine is better than us at choosing policies, we can't expect to vote: the machine knows better. Tegmark even wonders if the machines working for us day and night should have rights. After all, they would be slaves, intelligent slaves, which evokes images of the African trade slave and of Roman slaves.

Nobody has polled people to check if they want humans to continue to exist or if they prefer to be technologically replaced by something more efficient (e.g. longer living). Futurists wish to upload their mind into the eternal body of a machine. Regular people content themselves with raising children and wishing them a better life: how about instead raising machines programmed for a much better life? Would people accept as their descendants not people but machines? If machines (invented by humans) eventually take over the planet, they could be viewed as conquereros but also as, simply, descendants. The family tree of many Italians would show that their ancestors came from northern or eastern Europe and even north Africa, but most Italians view themselves as descendants of the Romans, so the genetic factor is not usually the one we use to determine descendants over the long run.

In the following chapter we are even asked to think whether we want life to spread into the whole universe. The religious person would probably be stuck because no holy scripture prescribes the mission to export life to every corner of the universe. The atheist would probably hesitate, wondering "What's in it for me? What difference does it make to my life? Does it even make sense to talk about good and evil when discussing living matter versus inert matter? Why is living matter better than nonliving matter?"

Tegmark then poses another meta-question to the ones who may be reluctant to endorse technological progress towards more intelligent machines: what happens if we don't develop such machines. Here of course the risk is that one day we'll be faced with an existential danger that humans alone are not smart and powerful enough to confront. Without vaccines, diseases would kill a lot more people. Without ambulances, a lot of people would die of heart attacks and accidents. Without power plants we wouldn't have heating and air conditioning. Future technology may cause problems, but not having technology may cause even bigger problems. Ask those who lived one hundred thousand years ago, or one thousand years ago, or even one century ago.

Chapter 6 is even more speculative, as Tegmark tackles the potential for life to spread everywhere. Tegmark seems fascinated by the potential for life to spread into the universe, and one has to wonder what makes him so excited. He won't be around. Nobody alive today will be. Why should we care that the universe become full of life? Why should we think that it is a good thing? Good for whom? Anyway, Tegmark points out that exploiting the matter of our solar system would increase resources for life a million times, and settling our galaxy would increase resources for life a trillion times. We are introduced to ideas such as Freeman Dyson's artificial biosphere (an entire planet used by Earthlings as a source of food and energy), Stephen Hawking's black hole power plant (exploiting both the black hole's radiation and its rotational energy), the sphaleron process (that produces energy by turning quarks into leptons). These are all ways to extract energy from matter. Then the limit becomes the speed of light: how many galaxies a civilization can settle if it expands in all directions at the speed of light. Tegmark imagines a future in which a super-intelligence conquers galaxies at the speed of light and transfers humans by reproducing them atom by atom, again at the speed of light. If you are already annoyed by the fast life of the city, this future is not for you. Tegmark continues his extrapolation into the future and reaches the end of the universe: life may expand all over the universe, but it would still die out if the universe ends. Here we are introduced to five possibilities: the heat death (or "big chill"), the big crunch (the opposite of the big bang), the big rip (a variation on the heat death), the big snap and death bubbles (he doesn't really explain the last two). You'll be surprised to find out that he thinks that the heat death is the most likely scenario, based on what we know today. Incidetally, he also thinks that intelligence only happened on this planet (he discussed it in "Our Mathematical Universe"). However, just like Freeman Dyson before him, Tegmark believes that life can substantially alter the scenario. First we can create a super-intelligence, and then this super-intelligence can generate an explosion of life all over the universe, and this could alter the outcome.

Chapter 7 is devoted to goal-directed behavior, which is pervasive in nature. Tegmark sees it as a general property of matter: nature always tries to maximize something, whether time, energy or entropy, and anything created by natural processes therefore has a "goal" (which is typically a subgoal to achieve one of those global maximizing goals). On one hand nature tries to maximize entropy, which would lead to the heat death, but on the other hand nature also creates self-organizing systems that fight that trend. As Schroedinger explained in "What is Life?", life decreases entropy locally (in violation of the second law of thermodynamics) because it increases it globally, i.e. by exploiting its surrounding environment. This self-organizing systems start replicating, and replication fights dissipation (the creation of entropy) in what becomes a sort of cosmic battle. This brings us to humans, who are such a case of self-organizing system: we use the environment to survive and improve our lot. We perceive this cosmic fight as "feelings": if hungry, we look for food; if thirsty, we look for water; and so on. Now Tegmark points out that a lot of matter exhibits a goal-directed behavior: it exists in order to achieve some kind of goal. This matter is part of the natural world. Humans have learned how to build machines that exhibit goal-directed behavior, and so in the future there could be more artificial goal-directed systems (machines designed by humans) than natural ones. The natural ones have no choice about what goals to pursue: nature decided for them what needs to be maximized. But human-designed systems can potentially have any kind of goal, even goals that counter the global goals of nature. Nature, to achieve its goals, has also created along the way a system, the human mind, that can design systems to fight nature's goals.

Now Tegmark takes a long detour into the goals of machines. His concern is how humans can be sure that machines will have and retain the correct goals. An artificial intelligence that is not a human has trouble understanding the goals that drive human behavior. Tegmark uses the example of a firefighter rushing into a burning building to save someone. The machine can infer that the firefighter was cold and wanted to warm up. Even if the machine learns the correct goals (the goals that will help humans), we have no guarantee that the machine will stick to them: the more intelligent the machine, the more likely that the machine comes up with "better" goals, which may or may not be beneficial to humans. Tegmark then enters the dangerous territory of ethics: someone needs to figure out which goals are the "right" goals for a machine. It is difficult enough to find two humans who can agree on a set of ethical principles, even if we limit ourselves to the big ones. For example, the USA is divided between opponents and supporters of the death penalty. Some countries punish with the death penalty the use of drugs that are legal in other countries. There are Muslim countries where you will be jailed if not killed if you dare say that Mohammed was not a good man, an opinion shared by many of us who are not Muslims. Not long ago it seemed natural that women were not allowed to vote. Slavery was ubiquitous in all societies. Moral beliefs have changed so much over the centuries, and change so much from country to country, that it's hard to claim anyone knows which moral beliefs will matter for future generations.

Tegmark asks an even bigger ethical question: what should the goal of the universe be? What should we strive to turn the universe into? Later in the book, Tegmark points out that it's not the universe that gives meaning to our lives, but it's our lives that can give meaning to the universe.

Finally it occurs to Tegmark that nobody has a definition of "intelligence", so we don't really know what a "super-intelligent" machine will be like. But he's already busy dealing with consciousness, the topic of chapter 8, and these days it's a lot easier to define consciousness (something that we all feel inside) than intelligence. The mysteries of consciousness are two: 1. how does matter become conscious, and 2. why does some matter become conscious. Here we are given a very partial survey of studies on consciousness. He gets close to endorsing panpsychism, but then opts for Giulio Tononi's theory that consciousness arises from "integrated information" and that consciousness is "the way information feels when it's processed in certain complex ways". This leads him to believe that it's only the structure of computation that matters, not the substrate in which the computation happens. Just like computation is "substrate-independent", i.e. that the same computation can be performed on different kinds of "computers", consciousness too can arise in different "substrates". Turing proved theoretically how to build one such universal computer, later implemented as the electronic computer, and Tegmark shows how to build a universal computer out of NAND gates, and Tegmark assumes that the human brain is a universal computer, and other thinkers have suggested that other physical systems could be made into universal computers. Tegmark concludes that computation is a pattern and it does not matter in which substance that pattern is implemented, just like information is a pattern and it can be implemented in substances like paper and silicon. (No evidence is provided that all these "universal computers" can compute the same things, e.g. that painting a picture on canvas is the same as painting the same picture on sand).

If a conscious machine emerges, Tegmark makes two interesting point: it may not share with us one of our fundamental feelings, the fear of death; and all artificial intelligences will share the same "memory" (since they can transmit information to each other). If you and i had the exact same memories, would i feel that i am myself, distinct from you?

In the preface Tegmark summarizes the current debate on the future of artificial intelligence: there are "digital utopians" who believe that digital life is the inevitable and desirable next step in the evolution of life; there are techno-skeptics who think that we are far away from having truly intelligent machines; there are those who think that artificial intelligence poses a threat to the human race; and there are those like Stuart Russell who study how to build beneficial A.I. In 2015, following a conference in Puerto Rico, Max Tegmark established the Future of Life Institute with Jaan Tallinn (Skype cofounder) and Anthony Aguirre (UC Santa Cruz astrophysicist). Its mission was "To catalyze and support research and initiatives for safeguarding life and developing optimistic visions of the future". The advisors included astrophysicists (Stephen Hawking and Martin Rees of Cambridge University, Saul Perlmutter of UC Berkeley, Alan Guth of the MIT), biologists (Christof Koch of the Allen Institute for Brain Science, George Church of Harvard University), philosophers (Nick Bostrom of Oxford's Future of Humanity Institute), economists (Erik Brynjolfsson of the MIT), high-tech entrepreneurs (Elon Musk of Tesla and SpaceX) and Artificial Intelligence scientists (Stuart Russell of UC Berkeley). In the last chapter of the book Tegmark returns to the topic of "friendly AI" and presents the "Asilomar Principles"

I have nothing against the good intentions of making sure that A.I. will be beneficial, but one wonders why only A.I. has to be beneficial. We are threatened on multiple fronts by software programs but somehow we should worry only about the software programs that don't exist (the super-intelligent ones) and not about the ones that exist and pervade our digital lives.

All in all, this thick stimulating book covers an impressive range of ideas, discussed with admirable competente and lucidity.

TM, ®, Copyright © 2022 Piero Scaruffi