Book summary

To explain the world, one needs a causal model. This means that when observing data, a DNA string, a time series, a text, a graph, etc, one has to find the process that can generate this data. Any data-generating model is, in practice, a computer program. For any piece of data, there are an infinite number of programs that can generate it. For example, a source code file containing a print statement that outputs our data does the trick. It is immediately clear that such a program would not satisfy us, nor would it explain the data. Intuitively, when you have a large dataset, which can be generated with a small program or model, it holds a lot of explanatory power. This is the idea behind Occam’s razor: preferring the simplest model for one’s data. Conversely, if it is not possible to find something smaller than just printing your data (meaning the length of your source code is about the length of the data), it might be an indication that your data is “random”. A simple example is the number , of which a simple program can generate as many digits as one would like, whereas a very long series of die roll outcomes can only be recorded, not computed. This is the idea behind Algorithmic Information Dynamics, the framework that Zenil and co-authors propose for understanding (living) systems. The keyword is compression: reducing data as much as possible with a computer program.

The framework presented in this book is based on algorithmic information theory, rather than statistical information theory, which merely searches for repetitions in the data, rather than general structures. Algorithmic information theory (AIT) is a surprisingly simple concept to grasp the main idea, but it has profound implications for computer science, mathematics, physics, biology, and science in general. It builds upon a Turing machine, the mathematical idealization of a computer. Any computer that can run a programming language such as Python can be used for the thought experiment, as it is a good approximation of the mathematical ideal. Any input to a computer is merely a text file (e.g., running a Python script), and there is only a finite number of source code files of a given length (2 to the power of the file size in bits). Rather than executing a specific program, we might run all possible source code files in lexicographic order. We start with the shortest file that runs (likely something like a=1) and record the output. Here, we can skip files with syntactic errors, as these are easy to detect. Each program will have an output, which will initially be simple, but once it is long enough to contain for and while loops, the output might be substantially longer than the input source code. It does not take very long programs to generate mathematical constants or things like palindromes. If we run enough programs, anything will be generated, ranging from computer simulations of atomic physics to all of Shakespeare’s works to a rendering of Avatar 4: Earth and Bone. This well-defined process will produce a distribution over all possible data, which we can think of as a universal prior distribution. Whatever we see earlier is simple (generated by short programs), while later outputs are more complex. Though generating a movie will require a complex computer program, the generating program is no doubt vastly shorter than the final movie file. Interestingly, there are only a limited number of short programs (say, of a megabyte), while there are many, many larger files (say, of a Gigabyte). This means that most data files cannot have a short generating program, making them algorithmically random. Every piece of data or media we like is a fleck of structure in a huge beach of randomness.

There are two critical remarks about the explanation above, one rather nice and the other not so much. My thought experiment assumed that we run all Python programs; does everything still hold if we use C, Rust, or Julia? What about using a more rigorously defined, but much less enjoyable-to-program, Turing machine? The good news is that yes, it is all (approximately equivalent). The complexity of outputs is the same, whether they are run in Python or some other language, because they are all Turing equivalent and have the same power. This is called the Invariance Theorem. Data that is simple (can be generated by a short input program) in Python will also be simple on any other computing platform. Measured in code length, there will only be a constant conversion factor needed to translate the code from language A to language B. This means that complexity, measured in minimal code length, is something universal.

Now, for the bad news. Running all computer programs up to a certain length might, while not possible in practice, be conceptually feasible by assuming a very large amount of computational resources. Sadly, it cannot even be done in theory, as some of the programs we will encounter might not even run very long; they might run forever and never halt! Though it is possible to sometimes detect whether a program will run forever (for example, it contains while True), in general, the logic might be so complex that it is impossible to determine. This is the famous Halting Problem, which states that no program can determine whether any program will terminate within a finite time or run forever. So, AIT is simple to understand, universal, well-defined, and computationally intractable. Much research is now being conducted on how to use it to solve practical problems.

To put it a bit more rigorously, the Kolmogorov complexity of a string is defined as the shortest program to generate it using a Turing machine :

Algorithmically random strings then have: . We can link probability (and hence priors) to algorithmic complexity by evaluating random programs in a Monte Carlo style! This is done by randomly generating program strings and evaluating them, yielding the algorithmic probability:

Though strings can be generated by programs of many different lengths, the shortest ones will dominate this probability, so we have that

which is called the coding theorem. The difference is merely factor of .

The authors of Algorithmic Information Dynamics use these ideas to estimate the Kolmogorov complexity of sequences and images. For small datasets, they use the Coding Theory Method (CTM), which is based on the above coding theorem. They run millions of programs (using execution-time thresholds to terminate those that might run forever) to create a database of how frequently each output is obtained. These probabilities can be turned into good estimates of the Kolmogorov complexity. For larger pieces of data (DNA strings or images), they glue these complexities of the smaller pieces together using statistical information theory, using the Block Decomposition Method (BDM) in a divide-and-conquer fashion. Based on CTM, the BDM method is defined as

where we decompose the string into substrings , where we compute the complexity using CTM and correct for repetitions using (we only have to invent a substring once). In other words, a long string is simple if it is made up of simple and repeating substrings!

With a practical way of assessing complexity behind the belt, it is now possible to reason about causality and the generative process of objects. The main idea they use it to break the data up into several pieces (for example, removing a link in a network) and see how this influences the complexity. This can be used to generate causally independent subnetworks into graphs using a perturbation analysis. One could imagine using this to segment genomes to identify evolutionary and functionally distinct regions, or using it to explore images. The strength of this work is that it turns the theoretically sound AIT framework into a practical and powerful tool for analysing data. It is a formal, universal, and practical approach to model discovery, hypothesis testing, and causal analysis. Expanding on this, exploring how it can be used to analyse biology (e.g., finding causal associations, understanding evolution…) could be highly fruitful. I am still not entirely convinced that AIT is the correct way to approach such a problem, as it assumes the existence of a highly complicated object - a Turing machine - as a precondition for generating data. Rather than exploring the space of random programs (which can only be short), it might be more relevant to explore the program space in a more systematic way, as in generative programming and program synthesis. It does seem that finding algorithmic descriptions of complex objects is the way forward.

Who is this for?

This is a rather technical book. The first parts give a general philosophical overview of the different approaches of science and explain the difference between AIT and statistical information theory well. The remainder builds upon the more extended framework, which is still very novel and might need some iterations to be in its most accessible form.

The main takeaway

This book introduces the main concepts of modeling and simulation in a practical, hands-on, case study-based fashion using Jupyter notebooks. The author keeps the mathematics to the bare minimum (it only requires a superficial knowledge of derivatives) while having the applications area as broad as possible. It covers first-order differential equations (growth), simple systems (the SIR model and an insulin model) and several second-order systems based on Newton’s second law (rolling toilet paper and throwing a baseball). Most examples are solved using custom ModSim software that uses a finite-step method. The author also introduces other tools, such as solving differential equations in Sympy and root-finding methods for optimization and parameter finding.

Who is this for?

I found this book while researching how to develop a more practical version of the modeling course I teach (that now only covers ODEs). This book is very introductory and likely won’t appeal to those who have advanced mathematics courses at some point. However, for those who did not, it is a very accessible introduction to modeling that can immediately be used for a wide range of systems. The author also has a lot of notebooks online for the exercises.

The main take-away

A system is a set of interconnected things in such a way that they produce their own behaviour. Outside forces can influence it, though internal feedback is more likely to drive it — a living cell, a forest or a city are good examples of systems. If we want to understand and manage such systems, it is better to act on the system rather than on external disturbances and perturbations. For example, to keep your garden free from pests, it is better to create enough diversity to provide niches for predators such as ladybugs rather than to try to eradicate them using pesticides directly.

Systems that work well (strong economy, good farmland, productive research group) usually have the following properties: 1. they are resilient and robust to changes; 2. they show a high degree of self-organization; 3. they are hierarchically organized.

When systems fail, there is usually also a system solution. For example, when dealing with the tragedy of the commons, you can change incentives by educating, regulating or privatizing.

Who is this for?

For all scientists and engineers, the book gives new mental tools to understand the world around you. It is even helpful for investing, policy-making and management. It uses virtually no mathematics, so it is accessible to everyone. A gem!

The main takeaway

Evolution is one of the most potent forces in the universe, having created everything in the biosphere and, some would argue, was the invisible hand behind all our technological innovations. Variation and selection intuitively explain how phenotypes are optimized to their environment, e.g., an enzyme that becomes more efficient or a bee’s mouthpiece that evolves to match a flower’s shape. However, it is less clear how new phenotypes (e.g. a new enzyme function or a new morphology) arise. Wagner, who previously wrote a brilliant book on robustness and evolution, outlines a theory on the evolution of biological innovations. He mainly studies three prototypical biological systems: gene regulatory networks, metabolic networks and sequences such as RNA and proteins. The implications go far beyond these, reaching higher functions and even the evolution of technology.

The main intuition pump in this work is the genotype network, a graph containing all genotypes connected by a mutation (i.e. that can be reached in a single step) and producing the same phenotype. So moving on this graph network via mutations is neutral; selective forces are blind to it. Wagner goes to great lengths to describe the properties of these graphs, both from a theoretical point of view and based on hundreds of empirical experiments on diverse biological systems. The networks are usually vast, covering large fractions of the genotype space. Different genotype networks are also closely intertwined, allowing for quickly hopping from one phenotype to a new one.

Genotype networks are exquisite concepts to think about evolution, much more profound than fitness landscapes. The latter only conveys a single, one-dimensional property, fitness, whereas genotype networks paint a picture of many different phenotypes without imposing an ordering. Genotype networks allow the exploration of neutral mutations that can lay the foundation for later beneficial mutations. Wagner outlines the various implications: - Robustness occurs in regions where the genotype network is dense, so mutations are likely to preserve the phenotype. Such regions are found to be highly evolvable. - Recombination allows making large jumps in the genotype network. - Duplication is a way to explore multiple paths in the genotype network simultaneously.

The most profound insight of the book is that evolution is more than mere selection of the fittest. Self-organization is equally important, as this creates the genotype networks that random walks can explore.

Who is this for?

The Origins of Evolutionary Innovations is denser than a typical popular science book, though a more pleasurable read than a textbook. This book requires a sizeable familiarity with molecular biology, though the main ideas should be accessible to anyone. The theoretical chapters are especially rewarding, as they give a fresh perspective on how to look at evolution.

The main takeaway

The Little Book of Stoicism gives a practical and accessible introduction to stoicism. The first half covers the essential aspects of the stoic principles. To thrive (reaching Eudaimonia): 1. focus only on the aspects you can control; 2. take responsibility for your actions; 3. try to aim to be the very best of yourself. Most people have only a limited sphere of control. Most circumstances are outside our power to change. We can choose how to react to what happens around us.

The book’s second half outlines more than 50 stoic practices and mindsets one can use to deal with daily hardships or be generally happier. They range from routines such as negative visualisation, voluntary discomfort and a stoic morning routine to mindsets such as considering everything being borrowed from nature, doing good and projecting kindness as strength.

Who is this for?

Life is hard—stoicism is a practical philosophy for becoming a more resilient, better, and generally happier person. Practising stoicism has helped me immensely with my mental well-being. This book is a very accessible introduction to getting started. It is chiefly a self-help book. For those who want a more formal, though still concise introduction, I can also strongly recommend “Lessons on Stoicism” by John Sellars.

The main takeaway

Every high school student can understand the basics of natural selection and grasp the origin of the diversity of the living world. Hawkins hopes that the secrets of the minds and intelligence (natural and artificial) might be similarly simple enough to be within reach of every interested layperson. A Thousand Brains outlines Hawkins’ theory of the mind.

The main idea is quite simple and seems reasonable with neuroscience. The neocortex consists of about 150,000 columns, where each column is a sensory-motor system. Each column has a similar function: representing part of the external world. These different models are linked using reference frames that connect them spatially, temporally or conceptually. The models are based on these reference frames. Each object or concept is represented by a relatively small number of these columns in a distributed fashion. A voting mechanism creates a consensus interpretation from the individual column results.

The book’s first third outlined the above theory and was quite interesting. The next part deals with what it would mean for artificial intelligence. The final third of the book deals with topics as diverse as memes, climate change, brain uploading and colonizing different planets; all put under the umbrella of “human intelligence”. I found the last chapters a bit unsubstantial. Other authors have treated this topic in much more depth.

The central theory seems to be coherent and supported by data. However, it seems unclear to me what it would mean practically, i.e. whether it can be used to generate some interesting, nontrivial forms of intelligence. For example, Hebbian learning is a theory used successfully in machine learning. Hawkins claims his framework is similar in spirit to Hinton’s capsule networks, which yet have to replace conventional artificial neural networks. I follow the reasoning of Yann LeCun. If you have a relevant theory of intelligence, you should have no problem showing superior performance on various learning tasks.

Who is this for?

Anyone interested in artificial intelligence and brain science. The book is accessible, though it lacks the technical details (for which Hawkins refers to his papers).

De hoofdboodschap

Vaak stelt men ecologisch handelen gelijk met consuminderen: minder vlees eten, minder verwarmen, zelden of nooit vliegen en vooral: minder mensen op de planeet. Mensen (en vooral deze met een Westerse levensstijl) zijn immers de bron van al het onheil. Groei is iets slechts.

Het ecomodernisme, in Vlaanderen sterk gepromoot door de auteurs Jan Deschoolmeester en Thomas Rotthier, is een alternatieve stroming. Ze vertrekt van economische ontwikkeling en technologische innovatie om uitdagingen zoals klimaat, energie en voedselbevoorrading aan te pakken. De argumentatie is als volgt: door minder te verbruiken kan je (bijvoorbeeld) uitstoot met maar een fractie terug brengen. Enkel een collectieve globale zelfmoord zou deze naar nul terug kunnen brengen. Groei en technologische innovaties daarentegen hebben geen limieten. Het uitrollen van -arme energieproductie laat in principe toe netto klimaatneutraal te zijn terwijl carbon capture (indien dit op grote schaal zou werken) het potentieel heeft om netto koolstof uit de lucht te halen. Er is geen limiet aan wat innovatie kan bereiken, Riley omschreef het als de echte Infinite Probability Drive.

In de eerste hoofdstukken van “De Wereld Red Je Niet Met Minder” beschrijven De Schoolmeester en Rotthier hoe de moderne mens altijd al technologie gebruikt heeft om natuur in cultuur om te toveren. Lang verhaal kort, wat jager-verzamelaars kwamen op het lumineuze idee om zelf wat graan te verbouwen en sindsdien nederzettingen, georganiseerde religie, oorlog, overbevolking, industrie, vervuiling en alle puree waar we nu mee zitten. (Toegegeven, er zitten ook goede dingen tussen.) De volgende hoofdstukken overlopen hoe technologie het antwoord kunnen bieden op de grote uitdagingen, van energie en klimaat tot landbouw en voeding. Met veel voorbeelden geven de auteurs een voerzicht van historische technologische successen alsook toekomstig potentieel van wat nu in de kinderschoenen staat.

Het laatste hoofdstuk was voor mij het meest intrigerenste, maar waarschijnlijk ook het meest polarizerendste. In de rentemeesters van de natuur pleiten De Schoolmeester en Rotthier om (bio)technologie zoals gene drives actief in te zetten om de natuur te beschermen en te beheren. Toegegeven, de auteurs beschouwen dit als laatste redmiddel voor bedreigde ecosystemen terwijl intensievere landbouw er voor zorgt dat er meer land vrij komt voor de natuur.

Mijn review

Kernenergie, klimaatsverandering, GMO’s zijn zeer actueel, “De Wereld Red Je Niet Met Minder” weet over elk van deze onderwerpen iets zinnigs te zeggen. Het boek is toegankelijk, informatief en leest als een trein. Wat ikzelf persoonlijk wat miste is een meer globale beschouwing en kritische blik op het vooruitgangsdenken, maar er zijn voldoende andere boeken die hier verder op in gaan.

Voor wie?

“De Wereld Red Je Niet Met Minder” is erg toegankelijk en aangenaam geschreven. Niet iedereen zal van het Ecomodernisme overtuigd zijn maar dit boek is een uitstekend vertrekpunt om er eens kennis mee te maken.

Mijn bedenkingen

De filosofie van “De Wereld Red Je Niet Met Minder” is gestoeld in het vooruitgangsoptimisme - aanhangers hiervan worden door sommigen tovenaars genoemd. Deze stelt dat de wereld en samenleving door de boot genomen altijd maar beter wordt, vanzelf bottom-up zonder sturing van buitenaf. Vooral Steven Pinker, Hans Rosling en Ray Kurzweil zijn prominente advocaten hiervan met boeken vol grafieken met stijgende (of dalende, indien slecht) trends.

Groen denken benadrukt vaak de eenheid met de natuur, waarbij natuur vaak als goed en puur gezien wordt en menselijke activiteit slecht en corrumperend. Dit gaat de biologische realiteit voorbij, elk species wil tot de uiterste limieten expanderen. De cyanobacteriën hebben er zich ook niets van aangetrokken toen ze de atmosfeer met (indertijd) giftig gas hebben gepompt. Meeste nieuwe species zijn experimenten die zichzelf om zeep helpen. De tragedie en geluk van Homo sapiens is dat we ons eigen noodlot kunnen zien aankomen, en er naar kunnen handelen.

The main takeaway

A living cell and one that just died have the same DNA. Put differently, both cells have precisely the same information content. Just as the flow of people and goods, rather than the arrangement of the buildings, determines that a city is alive, the fluxes of metabolites and energy characterise a living cell. Modern biology is often solely discussed in terms of information. In “Transformer”, Lane argues that metabolism is at least as important. This viewpoint of “follow the goods” is also emphasised on a completely different scale by Vaclav Smil in How the World Really Works.

In my opinion, Nick Lane is the best biology author out there. Lane has an unmatched talent for making seemingly boring topics (Transformer mainly covers the importance of the Krebs cycle) and making them exciting. Read this work and realise why the citric acid cycle is the answer to Life, the Universe and Everything. Within these pages, Lane depicts the Krebs cycle as the heart of the metabolism, linking it with the origin of life, assimilation, cancer, ageing and even consciousness.

The key insight is that the Krebs cycle links catabolism (degradation of metabolites for energy) and anabolism (synthesis of more complex metabolites for growth). It is a checkpoint for respiration and acts as a hub to start the synthesis of all building blocks of life: carbohydrates, lipids, nucleotides, and amino acids. Cancer cells, for example, promote their anabolic power by mainly using fermentation for energy. This process is known as the Warburg effect. It can even run in reverse, taking up as an alternative to the Calvin cycle for carbon fixation. Because the Krebs cycle controls reparation, it also regulates the electric potential on the membranes of the mitochondria (the topic of Lanes’ equally brilliant book “The Vital Question”). The last ( speculative though exciting) chapter discusses the emerging biology of electric fields and their relation with consciousness and morphogenesis (see also the work of Michael Levin on bioelectricity).

In addition to the pure science, Lane also paints the history of the unravelling of these biochemical processes. It conveys the brilliance and resourcefulness of the pioneers but does not skip the quarrels and small-mindedness.

Who is this for?

Honestly, not for the faint of heart, as this can be a highly technical popular science book. To follow Lane’s dissemination, you likely need to have followed a course in biochemistry at some point in your life. If you have, you are in for a treat! It is a joy to rediscover a topic previously filed under “dreadfully dull” and realise its profound implications in life and death.

The main takeaway

Tony Fadell had a career to be envious of: hard- and software designer (co-creator of the iPod and iPhone!), entrepreneur and CEO (founder of Nest, the company that made thermostats sexy), investor etc. Despite such a stellar track record, Fadell admits to making plenty of mistakes, just like the rest of us! Any successful person likely has plenty of mentors before becoming a mentor themself. In Build, Fadell bundles what he considers his most important advice to build a fruitful career. It covers aspects from your first job to how to become a CEO.

The book is subdivided into six sections, each containing several small chapters with tidbits of practical advice. It covers 1. Build yourself; where planning your first career should mainly be centred around what you want to learn rather than what you want to earn. Work hard, and work with established leads in your field of choice (“heroes”). I prefer small companies or groups where you can make a difference. 2. Build your career how to work in a group, why managing is a distinct skill and how to deal with various types of assholes. It also covers the etiquette of how and when to quit. 3. Build your product, good product design focuses on the customer. A successful product reveals and solves a problem your customer might not know they had (“painkillers are better than vitamins”). Your product should have a story that appeals to people’s emotional and rational side. Reaching profitability will always take longer than you think. 4. Build your business is just as important as building the product! A good business idea starts with the why (one should care). It covers accepting venture capitalists, work-life balance and coping with a crisis; 5. Build your team covers hiring, dealing with the growth of the company and the roles of sales, designers and lawyers. 6. Be the CEO; there can be several founders but only one CEO. You must deal with a board, buying and being bought while maintaining company culture. Fadell has some good views on perks (no free massages!), don’t give regular free stuff that becomes an acquired right though the occasional surprise boosts morale. Not all founders have what it takes to become a CEO, and even for successful CEOs, there might be a natural point to step down when you cannot let your company grow anymore, and you’ve effectively become a babysitter. ## Who is this for? This book generally contains good advice for those planning a career in a technical field. Many aspects might also be relevant for an academic profile, e.g. building and managing a group.

The main takeaway

Smil is a self-proclaimed generalist, trying to understand everything in the wide world as thoroughly as his mind allows, from the growth of bacteria to the organisation of cities. “How the World Really Works” tries to paint a complete picture of the world’s complex problems in fewer than 235 pages.

The key to understanding the world is to keep track of the energies, materials and risks involved. In the first chapters, Smil discusses energy production, food production and the main materials of modern life: fertilisers, cement, steel and plastic. These depend on relatively old and established technologies, such as turbines and the Haber-Bosch process. Importantly, they are largely anchored on fossil fuels. For example, a humble tomato requires an ample dose of fossil fuel (nicely visualised by pouring a couple of spoons of sesame oil over sliced tomato). All these production processes are so ingrained in our ways of life it is nearly impossible to decarbonise them.

The following chapters of the book deal with more abstract, though no less topical: globalisation, risk and environment. Again, Smil advocates a dry emphasis on keeping track of the numbers.

“How the World Really Works” outlines what is required for a reasonable standard of living. This involves food in our bellies and a roof over our heads. These commodities require certain quantities of resources and energy, and no foreseeable technology will change that. The book is neither pessimistic (proclaiming environmental collapses) nor optimistic (no singularly).

Who is this for?

Generalists, who want to understand what is required to keep a couple of billion people reasonably comfortable alive on our planet.

The main takeaway

Computers can be a phenomenal tool to improve learning in the classroom! Though written in the eighties, Mindstorms by Seymour Papert feels contemporary. Papert, an AI and education researcher at MIT, believes in using computers to create “microworlds” for teaching. These are tiny self-contained universes where children and students can explore a given concept, say geometry. Most of his case studies relate to experiences using the Logo programming language, which allows people to use simple, human-readable instructions to draw simple graphics, the original turtle graphics. Children are free to experiment, create, fail, and debug. The limited complexity serves as a safe haven for children to build confidence in their skills. I particularly liked one anecdote of a girl who initially had a lot of trouble writing programs. The instructor simplified her setup to boost her confidence, enabling her only to draw lines with fixed angles. After a couple of weeks of playing with the possibilities, she found herself suggesting more elegant ways to draw something than her instructor saw. The student became the master.

Interestingly, Papert did not advocate learning children to work with computers because it helps solve practical problems. Instead, he valued programming because it improves how children think about problems. It motivates creativity, shows multiple ways to tackle a problem and forces students to identify and fix errors.

Who is this for?

Mindstorms is a great inspiring book for teachers, not only for introducing children to programming but as a general case study in discovery-based learning. I even believe there is quite some wisdom for computational researchers as well.

The main takeaway

I recently read ‘Antifragile’ by Nicholas Taleb. The author discusses how objects, organisms and systems can differ concerning how tolerant they are against stresses, shocks, perturbations and other random external forces. He distinguishes three categories based on how you would send them by the post office. 1. There is the fragile; the stuff labelled “FRAGILE: HANDLE WITH CARE”: Ming vases and crystal wine glasses. 2. Some packages might contain books or kettlebells. You generally don’t have to worry much about dropping them because they are robust. 3. Finally, you have a mysterious third type of packages, the antifragile¹, which improve by stressors. Such packages might carry a label with “SHAKE ME PLEASE!”. What kind of magical things could such packages contain?

Taleb likens fragile things with candles; a puff of wind would extinguish them. A powerful torch would be the robust analogue; it can copy with moderate winds. However, a forest fire would be antifragile; winds would only invigorate it! Many interesting, complex systems are antifragile: a capitalist economy, the scientific process, pandemics, ecosystems, the immune system, democracy as envisioned by Klaas Mensaert, living organisms… The key to antifragility is being composed of fragile components that can easily be replaced. Science moves because good ideas substitute bad ones, faulty cells in our body undergo apoptosis and ineffective virus strains become replaced by more effective mutants. Antifragile things seem to be evolved or emerged as opposed to being designed. Recent work discusses how we have had to rethink our definition of “machine” in the area of synthetic biology and artificial intelligence where machines are being evolved.

Crucially, Taleb’s point, as outlined in his earlier books, is that the world is fundamentally unpredictable, the dreaded Black Swans. These include earthquakes, stock market crashes and pandemics. In many cases, we have a depressingly low capacity to influence, predict, or in some cases, even understand our environment. However, one can often control the system that generates the outcomes from the environment (say, your long-term investment returns or the number of exciting research ideas you can bring to fruition). In mathematical terms, one must ensure that the function that generates the outcomes from the random environment is convex. From Jenssen’s inequality it follows that the expected outcome will exceed the outcome of the expected environment:

To take a very simple example, let , i.e., we cap our downsize and only take the positive benefits. It is trivial to see that such a system gains from increased randomness; we only retain the benefits without drawbacks!

So, practically, what does it mean? From a productivity standpoint, you have the likes of Cal Newport, who would prefer to live in a lead box, meticulously controlling his signal-to-noise input. Based on the ideas in Antifragile (and the related books shown on top of this post), I think it is smarter to set up a system that gains from a moderate dose of disorder. Tools such a the smart note-taking principle and Building a Second Brain can help with the information overload and allow new ideas to crystalize.

Who is this for?

I enjoyed reading this book and found the ideas quite thought-provoking. The appendices were illuminating, as Talebs deduces his theory from simple mathematical principles. Sure, in retrospect, the book’s message is rather simple and obvious, though that is part of the charm, I guess.

Taleb’s writing style is not for everyone. He is probably the most well-read anti-intellectual, never missing a chance to dish out to academics, economists, medical practitioners and other IYIs. Taleb manages to appear to be profoundly wise and to be a bubbling vat of frustration at the same time. I was not really put off by his style (I think Talab has a heart at the right place), but I guess it will not be for everyone.

Those who like to think in systems and believe that the most simple solution is often the best will enjoy this book.

Review

In “The Flaws That Kill Our Democracy”, Mensaert discusses the main issues with our current western democratic system, which inevitably leads to polarization, stagnation and corruption. This book outlines an alternative approach based on smaller, specialized inclusive parties. Borrowing ideas from scientific disciplines such as complex systems theory, Mensaert argues that such a system would be antifragile, which would be more tolerant to external lobbying and influence.

It is a thin booklet yet rich in ideas. I immensely enjoyed the scientific approach to democracy. The last part is, relating to voting, is in my option the least well developed (or rather, it discusses the most complicated problem: finding out what people want!)—food for many stimulating discussions.

The main takeaway

It has been said that democracy is the worst form of government except all the others that have been tried. - Winston Churchill

With one equal vote for every person, democracy is held as a pillar of enlightenment. As with every system that involves more than two humans, the current implementation exhibits some rampant flaws, from corruption at the top and misinformation and fake news at the bottom. The current state of affairs leaves many people across the political spectrum disillusioned.

In his book, “The Flaws That kill our Democracy”, Klaas Mensaert argues for a scientifically-driven approach towards a better democratic system. A democracy should function similarly to the principles of the scientific system. Here, scientists prose, test and publish falsifiable hypotheses, increasing the amount of knowledge in the world. Likewise, a democratic process should be able to establish what people what to happen. Societies problems (energy, migration, climate change, economy, etc.) are complex. Hence these can only be approached from the bottom up. An example of a similar system that works is capitalism. Capitalism, for all its flaws, can generate value and goods via unimaginable complex supply chains for most of the people who are participating. The wondrous thing is that this capitalism is almost completely self-organizing, composed of independent, selfish actors. A fundamental and desirable property of such systems is that they are antifragile: external stresses, errors and attacks do not weaken them; it makes them stronger. Taleb, who coined the term, argues that systems can only be antifragile if their subcomponents themselves are fragile. A capitalistic economy, composed of numerous companies that can go out of business, can be antifragile. Living organisms composed of delicate living cells with constant turnover are Antifragile. The current political system with its large political parties that are too big to fail is not antifragile.

Mensaert argues that in most Western countries, political parties are far too big and far too static. In “Het DNA van Vlaanderen”, De Vadder writes that the most remarkable thing that can be said about most Flemish parties is that they are still there after that many years. The most extreme case is likely in the USA, where two old parties completely dominate the political landscape. Such power structures work corruptions and stagnation in hand. Worse, as parties need to profile themselves, this inevitably leads to polarization, making compromises hard.

The book proposes a more decentralized system where many smaller and volatile parties populate the political landscape. This can be realized by two components: specialization and inclusiveness. Specialization means that parties do not need to focus on every single issue but specific problems. An advantage of more focused parties is that their representatives can have more targeted expertise. In addition to specialized parties, the system also needs to be inclusive, meaning it is possible to join or vote on multiple political parties. in this system, a citizen, the freedom to combine parties, leads to a combinatorial explosion of options, meaning that the representation can be as diverse as the voters. In How Innovation Works, Ridley praises the recombination of existing ideas, a concept that could here be incorporated in democracy.

The final part of the book relates to voting. Two tools are discussed here: allowing symmetric voting (i.e., allowing people to express a positive, negative or neutral attitude towards a proposition) and voting aggregations to correct for popularity. By allowing a negative or neutral stance, one can measure how polarizing an issue is. I think this part is the most difficult to implement. The field of preference aggregation is complex and riddled with paradoxes. In all likelihood, it will be hard to design a voting system that is both simple, results sensible outcomes and cannot be gamed.

Here is an example of how I interpreted the ideas of the book. Energy is a hot-button issue now in Belgium, where the green party favours renewables. In contrast, the right-wing party pushes nuclear energy (it is easy to imagine a parallel universe where the opposite would be true). Views on energy are currently linked to the party, complete with all other ideological baggage. In practice, however, opinions are much more diverse. The Ecomodernism movement claims the need for a party that embraces technology and a growing economy and an environmental sense. However, ‘Het DNA van Vlaanderen’ claims, based on an extensive survey on peoples political preferences, that a party targeted to such people would not enjoy broad support (Flanders has a strong right-left axis both socially and economically). In the world of Mensaert, energy is determined by specialized parties, which can freely be mixed and matched with parties that cater to other issues such as economy and migration.

Who is this for?

Systems thinkers who would like to see the democratic system fixed.

Newton’s method

Newton’s method is a root-finding algorithm. We can use it to find roots of a function , i.e., inputs for which it holds that .

When executing Newton’s method, we start with an initial and iteratively apply the following rule till convergence:

This rule can be derived by taking the first-order Taylor approximation of in and solving for the step that sets the approximation to zero.

Let us apply this to a simple function!

f(x) = x^3 - 1;

For most functions, we can compute its derivative, and hence Newton update, easily enough. However, let us be lazy and use this opportunity to use Julia’s new CAS Symbolics.jl. We write a function that takes a function as an input and computes the map above:

using Symbolics

function get_map(f)
    # define variable
    @variables x
    # define derivative operator
    Dx = Differential(x)
    map = x - f(x) / Dx(f(x)) |> expand_derivatives
    # now we expand and compile to a Julia function
    update_expr = build_function(map, x)
    return eval(update_expr)
end

update = get_map(f)

Yes! We get a new function: the update map! Let us try it on a value

update(2.0)  # apply map once

update(update(2.0))  # apply map twice

update(update(update(2.0)))  # apply map thrice

update(update(update(update(2.0))))  # apply map four times

We see that after a couple of steps, the Newton method converges to the root .

For our convenience, let us define a function that applies this map times.

function applyiteratively(x, update; n=100)
    for i in 1:n
        x = update(x)
    end
    return x
end

applyiteratively(2.0, update)  # apply 100 times

Complex roots

Astute readers might have noticed that has three roots: one real () and two complex ones ( and ). Would our method also work with complex inputs?

applyiteratively(-2.0 - 2.0im, update)

We see that this converges to a different (complex) root! Other values might converge to different roots!

applyiteratively(2.0 + 2.0im, update)

The Newton fractal

Imagine that we try this process for many complex numbers in an interval. Depending on the initial starting input, we end up in a different root. By colouring the results according to the root we end up in, we obtain the Newton fractal.

lower = -2 - 2im
upper = 2 + 2im

step = 0.5e-2

# generate a range of complex values
Z0 = [a+b*im for b in real(lower):step:real(upper),
                    a in imag(lower):step:imag(upper)]

# apply the update 100 times
Z100 = applyiteratively.(Z0, update);

This results in a large complex matrix. We might define a colourmap for complex numbers as done here, though plotting the angle of the complex number in polar coordinates suffices.

using Plots

heatmap(angle.(Z100), colorbar=false, color=:rainbow, ticks=false)

Below are some other examples.