SECRET LIFE

The principle of energy conservation underpins much of physics and chemistry, to include mechanics, thermodynamics, and relativity (Mach, 1910). By contrast, the idea of statistical information provides a common language for genetics, development, and behavior, where information encodes “coordinates” for free energy landscapes – memory required by search strategies directed at appropriating metabolic resources (Krakauer, 2011). This memory is often encoded in genetic sequences that express enzymes, signaling molecules, and receptors, all of which can orient toward or process metabolic free energy. Whereas our understanding of energy can be reduced ultimately to symmetry principles (Brading and Castellani, 2003), information is derived from its antithesis, symmetry breaking (Gell-Mann, 1995). Stored information records which of several alternative adaptive configurations have been selected historically.

Adaptive matter requires metabolic energy for biological work – growth, reproduction, and repair. These requirements extend upward to populations, ecosystems, and even cities. The patchy distribution of free energy in space and time, combined with the fact that energy is transformed and thermalized when metabolized, requires efficient and accurate procedures for energy seeking. This scarcity problem is the reason why life can be thought of as a suite of evolved inferential mechanisms dependent on both memory storage – information – and “computation” for adaptive search.

Evolution by natural selection – a population-based search mechanics – produces outcomes in which information is restricted and concentrated ensuring that energy is distributed nonuniformly and anticompetitively: sequestered within cells, bodies, and communities, where it can be preferentially utilized and monopolized (Krakauer et al., 2009).

Organizations – cells, organisms, and populations – with accurate information about the whereabouts of metabolic energy sources endeavor to keep this information to themselves and share informative signals only with those with whom they have found a means to cooperate. These preferred signals include those generating the immunogenic self versus nonself, mating types, restriction systems, species isolating mechanisms, and, in some cases, languages.

The best way to restrict the flow of information is to protect it or to encrypt it (Shannon, 1949). This is the subject of this chapter: the many ways in which evolved information flows are restricted and metabolic resources protected and hidden, the thesis of living phenomena as evolutionary cryptosystems.

Information and computation have transformed the way we look at the world beyond statistical correlations, the way we can perform experiments through simulations, and the way we can test these hypotheses. In Turing's seminal paper on the question of machine intelligence (Turing, 1950), his approach consisted in taking a computer to be a black box and evaluating it by way of what one could say about its apparent behaviour. The approach can be seen as a digital version of a cogito ergo sum dictum, acknowledging that one can be certain only of one's own intelligence but not of the intelligence of anyone or anything else, including machines. This may indeed amount to a kind of solipsism, but in practice it suggests tools that can be generalised. This is how in reality we approach most areas of science and technology, not just out of pragmatic considerations but for a fundamental reason. Turing, likeGödel before him (Gödel, 1958), showed that systems such as arithmetic that have a certain minimal mathematical power to express something can produce outputs of an infinitely complicated nature (Turing, 1936). Indeed, this means that while one can design software, only trivial programs can be fully understood analytically to the point where one is able to make certain predictions about them. In other words, only by running software can one verify its computational properties, such as whether or not it will crash. These fundamental results deriving from both Gödel's incompleteness theorem and Turing's halting problem point in the direction that Turing himself took in his ‘imitation game’, which we now identify as the Turing test.

Turing's halting problem implies that one cannot in general prove that a machine will ever halt or that a certain configuration will be reached on halting; therefore, one has to proceed by testing and only by testing. These fundamental theorems and results imply that testing is unavoidable; even under optimal conditions there are fundamental limits to the knowledge one can obtain simply by looking at a system, or even by examining its source code, that is, its ultimate causes. Even with a knowledge of all the causes driving a system's evolution, there are strong limits to the kind of understanding of it that we can attain.

We are surrounded by noise. The controlled explosions of internal combustion engines combine with the roar of rubber on asphalt to create the drone of road and highway traffic. Weed-whackers, lawn mowers, and leaf blowers can turn sunny suburban summer days into a buzzing confusion. Indeed, the entire universe is filled with the faint din that is the cosmic microwave background (CMB) radiation, leftover electromagnetic radiation from the Big Bang. The CMB was discovered when Penzias and Wilson (1965) went looking for the source of annoying hiss plaguing their new radio telescope, hiss that threatened to obscure signals from distant stars and galaxies. Noise seems to be entirely destructive, thus something to be eliminated, if possible.

Noise can be beneficial, however, in at least two ways. One is familiar, the other paradoxical and far less well known. The familiar way is simply as a source of variety. For example, genes undergo random mutation from processes both external to the organism (e.g., cosmic rays) and internal (Dobrindt and Hacker, 2001). This genotypic variation is the source of heritable variation in phenotype, which is of course essential for the process of natural selection (Wagner, 2014). Phenotype can even vary within isogenic populations due to variation in gene expression (Fraser and Kaern, 2009). This phenotypic noise is thought to provide an evolutionary advantage for some microorganisms, as it increases the chance that some will survive under stressful conditions. The CMB noise, though destructive from the standpoint of the users of radio telescopes, plays a constructive role in generating the tiny variations in energy density in the early universe that are the seeds of structure formation. The random fluctuations we call noise gave rise to stars and galaxies and galactic clusters. This much is familiar, at least to those working in the relevant areas of biology or astrophysics.

Considerably less familiar is the role that noise can play in nonlinear systems, in particular systems with one or more thresholds, points at which small differences in input give rise to disproportionate differences in output. Converting an analog signal into a digital signal involves sampling the signal at regular intervals and writing down a digital approximation of the amplitude at each point.

Life has been described as information flowing in molecular streams (Dawkins, 1996).Our growing understanding of the impact of horizontal gene transfer on evolutionary dynamics reinforces this fluid-like flow of molecular information (Joyce, 2002). The diversity of nucleic acid sequences, those known and yet to be characterized across Earth's varied environments, along with the vast repertoire of catalytic and structural proteins, presents as more of a dynamic molecular river than a tree of life. These informational biopolymers function as a mutualistic union so universal as to have been termed the Central Dogma (Crick, 1958). It is the distinct folding dynamics-the digital-like base pairing dominating nucleic acids, and the environmentally responsive and diverse range of analog-like interactions dictating protein folding (Goodwin et al., 2012)-that provides the basis for the mutualism. The intertwined functioning of these analog and digital forms of information (Goodwin et al., 2012) unified within diverse chemical networks is heralded as the Darwinian threshold of cellular life (Woese, 2002).

The discovery of prion diseases (Chien et al., 2004; Jablonka and Raz, 2009; Paravastu et al., 2008) introduced the paradigm of protein templates that propagate conformational information, suggesting a new context for Darwinian evolution. When taking both protein and nucleic acid moelcular evolution into consideration (Cairns- Smith, 1966; Joyce, 2002), the conceptual framework for chemical evolution can be generalized into three orthogonal dimensions as shown in Figure 5.1 (Goodwin et al., 2014). The 1st dimension manifests structural order through covalent polymerization reactions and includes chain length, sequence, and linkage chemistry inherent to a dynamic chemical network. The 2nd dimension extends the order in dynamic conformational networks through noncovalent interactions of the polymers. This dimension includes intramolecular and intermolecular forces, from macromolecular folding to supramolecular assembly to multicomponent quaternary structure. Folding in this 2nd dimension certainly depends on the primary polymer sequence, and the folding/assembly diversity yields an additional set of environmentally constrained supramolecular folding codes. For example, double-stranded DNA assemblies are dominated by the rules of complementary base pairing, while the self-propagating conformations of prions are based on additional noncovalent, environmentally-dependent interactions.

Historically, scientists have defined the “macrostate” of a system, and the associated “level” or “scale,” in a purely informal manner, based on insight and intuition. For example, in evolutionary biology often the macrostate is an entire species, a quantification of a set of coevolving organisms that ignores within-species diversity, fundamental dependencies between organisms, and complicated subunits such as tissues and cells that can also reproduce. Similarly, in economics the macrostates of the world's socioeconomic system are often defined in terms of firms, industrial sectors, or even nation-states, neglecting the internal structure of these highly complex entities.

One might view the reliance of many sciences on such vague human “insight” into how to even describe a physical system – a reliance that has no formal justification – as troubling. How do we know that these choices for the macrostates are the best ones with which to analyze the system? How do we even quantify the quality of a choice of macrostate? Might there be alternatives that are superior to our choices? A superior choice might, for example, allow greater accuracy in our prediction of the evolution of the system and/or reduce the computational cost of making such predictions. Given the possibility that superior choices might exist, can we solve for the optimal macroscopic state space with which to investigate a system?

This question, of how best to compress amicrostate of a system into a macrostate, is the general problem of state space compression (SSC). To address this problem, we must first decide how to quantify the quality of a proposed map xt → yt that compresses a dynamically evolving “fine-grained” microstate xt into a dynamically evolving “coarse-grained” macrostate yt. Given a definition of the quality of any compression and a microstate dynamics xt, we can try to solve for the best map compressing xt into a higher-scale macrostate yt. The dynamics of such an optimally chosen compression of a system can be viewed as defining its emergent properties. Indeed, we may be able to iterate this process, producing a hierarchy of scales and associated emergent properties, by compressing the macrostate y to a yet higherscale macrostate y′.

Organisms reproduce themselves, which means that their structures and processes must be copied to pass on to later generations. The question addressed in this chapter is whether they do this by encoding everything in digital format or whether analogue information is also important. The complete genome sequence of an organism is often compared to the digital information in a computer program or data file. Strong versions of biological reductionism also assume that this information is sufficient to define the organism and its development. This idea, popularised today in the idea of a genetics program, has its origins in the mechanistic philosophy of René Descartes. In his treatise on the formation of the fetus in 1664, he wrote:

This statement can be seen to foreshadow the Central Dogma of molecular biology (Crick, 1970), which in turn can be seen to echo the idea of the Weismann Barrier (Weismann, 1893): that the genetic material is isolated from both the rest of the organism and its environment and that it is sufficient in itself to specify the development of an organism.

THE QUESTION

The question addressed in this chapter is: Are organisms encoded as purely digital molecular descriptions in their gene sequences? By analysing the genome alone, could we then solve the forward problem of computing the behaviour of the system from this information? I argue that the first is incorrect and that the second is impossible. We therefore need to replace the gene-centric digital view of the relation between genotype and phenotype with an integrative view that also recognises the importance of analogue information in organisms and its contribution to inheritance across generations. Nature and nurture must interact. Either on its own can do nothing.

The term “dynamical system” encompasses a vast class of objects and phenomena – any system whose state evolves deterministically with time over a state space according to a fixed rule (Nykamp, 2015). Since living systems must necessarily change their state over time, it is not surprising that many attempts have been made to model a wide range of living systems as dynamical systems (de Jong, 2002; Ermentrout and Edelstein-Keshet, 1993; Izhikevich, 2007; Kaneko, 2006; Nowak, 2006; Sumpter, 2006), although identifying an accurate time-evolution rule and estimating the relevant state variables can be hard. On the other hand, even simple state-update rules can give rise to complex spatio-temporal patterns, reminiscent of certain features associated with living systems, such as complex behavioral patterns, self-organizing behavior, and self-reproduction. This has been demonstrated extensively using a class of simple, discrete dynamical systems called “cellular automata” (CA) (Gardner, 1970; Knoester et al., 2014; von Neumann, 1966; Wolfram, 1984, 2002). CA consist of a lattice of identical cells with a finite set of states. All cells evolve in parallel according to the same local update rule, which takes the states of their neighboring cells into account (Figure 14.1).

In any dynamical system, the time-evolution rule determines the future trajectory of the system in its state space given a particular initial state. The aim of dynamical systems theory is to understand and characterize a system based on its long-termbehavior, by classifying the geometry of its long-term trajectories. In this spirit, CA have been classified according to whether their evolution for most initial states leads to fixed points, periodic cycles, or chaotic patterns associated with strange attractors (Li and Packard, 1990; Wolfram, 1984) (Figure 14.1B). Despite the simplicity of the rules, this classification is undecidable for many CA with infinite or very large numbers of cells (Culik and Yu, 1988; Sutner, 1995). This is because determining the trajectory of future states is often computationally irreducible (Wolfram, 1985), meaning that it is impossible to predict the CA's long-term behavior in a more computationally efficient way than by actually evolving the system.

Constructor theory (Deutsch, 2013) is a new mode of explanation in fundamental physics, intended to improve on the currently most fundamental theories of physics and explain more of physical reality. Its central idea is to formulate all laws of physics as statements about what transformations are possible, what are impossible, and why. This is a sharp departure from what I call the prevailing conception of fundamental physics, which formulates its statements exclusively in terms of predictions given the initial conditions and the laws of motion. For instance, the prevailing conception of fundamental physics aims at predicting where a comet goes, given its initial conditions and the laws of motion of the universe; instead, in constructor theory the fundamental statements are about where the comet could be made to go, with given resources, under the dynamical laws.

This constructor-theoretic, task-based formulation of science makes new conceptual tools available in fundamental physics, which resort to counterfactual statements (i.e., about possible and impossible tasks). This has the potential to incorporate exactly into fundamental physics notions that have so far been considered as highly approximate and emergent – such as information and life.

The constructor theory of information (Deutsch and Marletto, 2015) has accommodated the notion of (classical) information within fundamental physics and has unified it exactly with what currently goes under the name of ‘quantum information’ – the kind of information that is deemed to be instantiated in quantum systems. This theory provides the conceptual basis for the constructor theory of life (Marletto, 2015a). Here constructor theory is applied to produce the explanation of how certain physical transformations that are fundamental to the theory of evolution by natural selection – such as accurate replication and self-reproduction – are compatible with laws of physics that do not contain the design of biological adaptation. It also shows what features such laws must have for this to be possible: in short, they must allow the existence of information media, as rigorously defined in the constructor theory of information.

Before going into the details of these theories, one has to introduce the foundations of constructor theory. Constructor theory consists of ‘laws about laws’: its laws are principles – conjectured laws of nature, such as the principle of conservation of energy.

The concept of information has penetrated almost all areas of human inquiry, from physics, chemistry, and engineering through biology to the social sciences. And yet its status as a physical entity remains obscure. Traditionally, information has been treated as a derived or secondary concept. In physics especially, the fundamental bedrock of reality is normally vested in the material building blocks of the universe, be they particles, strings, or fields. Because bits of information are always instantiated in material degrees of freedom, the properties of information could, it seems, always be reduced to those of the material substrate. Nevertheless, over several decades there have been attempts to invert this interdependence and root reality in information rather than matter. This contrarian perspective is most famously associated with the name of John Archibald Wheeler, who encapsulated his proposal in the pithy dictum ‘it from bit?’ (Wheeler, 1999).

In a practical, everyday sense, information is often treated as a primary entity, as a ‘thing in its own right’ with a measure of autonomy; indeed, it is bought and sold as a commodity alongside gas and steel. In the life sciences, informational narratives are indispensable: biologists talk about the genetic code, about translation and transcription, about chemical signals and sensory data processing, all of which treat information as the currency of activity, the ‘oil’ that makes the ‘biological wheels go round’. The burgeoning fields of genomic and metagenomic sequencing and bioinformatics are based on the notion that informational bits are literally vital. But beneath this familiar practicality lies a stark paradox. If information makes a difference in the physical world, which it surely does, then should we not attribute to it causal powers? However, in physics causation is invariably understood at the level of particle and field interactions, not in the realm of abstract bits (or qubits, their quantum counterparts). Can we have both? Can two causal chains coexist compatibly? Are the twin narratives of material causation and informational causation comfortable bedfellows? If so, what are the laws and principles governing informational dynamics to place alongside the laws of material dynamics?

Machine learning is a lively academic discipline and a key player in the continuous pursuit for new technological developments. The editorial in the first issue of the journal Machine Learning, published in March 1986, described the discipline as that field of inquiry concerned with the processes by which intelligent systems improve their performance over time (Langley, 1986). A glossary of terms published in the same journal in 1998 refined this to: Machine Learning is the field of scientific study that concentrates on induction algorithms and on other algorithms that can be said to ‘learn’ (Kohavi and Provost, 1998).

Thomas J. Watson, the brilliant salesman who from 1914 to 1956 oversaw the remarkable growth and success of IBM, serving as both CEO and chairman, was famously quoted as saying in 1943, ‘I think there is a world market for maybe five computers.’ With more than one billion computers now in use worldwide (Virki, 2008), this quote is often referenced to illustrate how vastly their usefulness had been underestimated.No area of computer science is making progress more rapidly than machine learning, with computers being capable of tasks that were a few decades ago only mentioned in science-fiction stories. Watson brought to IBM from his previous employment his trademark motto ‘Think’. It would at the time have been reasonable for Watson to suppose that only humans could really ‘think’. While computers could surpass humans in adding, subtracting, multiplying, and dividing, they were hardly thought of as being good at human tasks, such as playing chess, which required thinking. This begs the question, ‘What is thinking?’ In February 1996 World Chess Champion Garry Kasparov took on the IBM computer Deep Blue in Philadelphia. Even with the IBM engineers allowed to reprogram the computer between games, the world champion won, but only just, losing one game, drawing two, and winning three. His victory was short-lived. The following year he played a rematch. With the score even after the first five of six games, Kasparov allowed Deep Blue to commit a knight sacrifice, which wrecked his Caro–Kann defence and forced him to resign in fewer than twenty moves.

The lack of a rigorous account of biological information as a proximal causal factor in biological systems is a striking gap in the scientific worldview. In this chapter we outline a proposal to fill that gap by grounding the idea of biological information in a contemporary philosophical account of causation. Biological information is a certain kind of causal relationship between components of living systems. Many accounts of information in the philosophy of biology have set out to vindicate the common assumption that nucleic acids are distinctively informational molecules. Here we take a more unprejudiced approach, developing an account of biological information and then seeing how widely it applies.

In the first section, ‘Information in Biology’, we begin with the most prominent informational idea in modern biology – the coding relation between nucleic acid and protein. A deeper look at the background to Francis Crick's Central Dogma, and a comparison with the distinction in developmental biology between permissive and instructive interactions, reveals that ‘information’ is a way to talk about specificity. The idea of specificity has a long history in biology, and a closely related idea is a key part of a widely supported contemporary account of causation in philosophy that grounds causal relationships in ideas about manipulability and control. In the second section, ‘Causal Specificity: An Information-Theoretic Approach’, we describe the idea of ‘causal specificity’ and an information-theoretic measure of the degree of specificity of a cause for its effect. Biological specificity, we suggest, is simply causal specificity in biological systems. Since we have already argued that ‘information’ is a way to talk about biological specificity, we conclude that causal relationships are ‘informational’ simply when they are highly specific. The third section, ‘Arbitrariness, Information, and Regulation’, defends this identification against the claim that only causal relationships in which the relation between cause and effect is ‘arbitrary’ should count as informational. Arbitrariness has an important role, however, in understanding the regulation of gene expression via gene regulatory networks. Having defended our identification of information with specificity, we show in the final section, ‘Distributed Specificity’, that information is more widely distributed in biological systems than is often supposed. Coding sequences of DNA are only one source of biological specificity, and hence only one locus of biological information.