Evolvability of Natural and Artificial Systems

How Organisms Come to Know the World: Fundamental Limits on Artificial General Intelligence

Artificial intelligence has made tremendous advances since its inception about seventy years ago. Self-driving cars, programs beating experts at complex games, and smart robots capable of assisting people that need care are just some among the successful examples of machine intelligence. This kind of progress might entice us to envision a society populated by autonomous robots capable of performing the same tasks humans do in the near future. This prospect seems limited only by the power and complexity of current computational devices, which is improving fast. However, there are several significant obstacles on this path. General intelligence involves situational reasoning, taking perspectives, choosing goals, and an ability to deal with ambiguous information. We observe that all of these characteristics are connected to the ability of identifying and exploiting new affordances—opportunities (or impediments) on the path of an agent to achieve its goals. A general example of an affordance is the use of an object in the hands of an agent. We show that it is impossible to predefine a list of such uses. Therefore, they cannot be treated algorithmically. This means that “AI agents” and organisms differ in their ability to leverage new affordances. Only organisms can do this. This implies that true AGI is not achievable in the current algorithmic frame of AI research. It also has important consequences for the theory of evolution. We argue that organismic agency is strictly required for truly open-ended evolution through radical emergence. We discuss the diverse ramifications of this argument, not only in AI research and evolution, but also for the philosophy of science.

The World Is Not a Theorem

The evolution of the biosphere unfolds as a luxuriant generative process of new living forms and functions. Organisms adapt to their environment, exploit novel opportunities that are created in this continuous blooming dynamics. Affordances play a fundamental role in the evolution of the biosphere, for organisms can exploit them for new morphological and behavioral adaptations achieved by heritable variations and selection. This way, the opportunities offered by affordances are then actualized as ever novel adaptations. In this paper, we maintain that affordances elude a formalization that relies on set theory: we argue that it is not possible to apply set theory to affordances; therefore, we cannot devise a set-based mathematical theory to deduce the diachronic evolution of the biosphere.

Real-World Robot Evolution: Why Would it (not) Work?

This paper takes a critical look at the concept of real-world robot evolution discussing specific challenges for making it practicable. After a brief review of the state of the art several enablers are discussed in detail. It is noted that sample efficient evolution is one of the key prerequisites and there are various promising directions towards this in different stages of maturity, including learning as part of the evolutionary system, genotype filtering, and hybridizing real-world evolution with simulations in a new way. Furthermore, it is emphasized that an evolutionary system that works in the real world needs robots that work in the real world. Obvious as it may seem, to achieve this significant complexification of the robots and their tasks is needed compared to the current practice. Finally, the importance of not only building but also understanding evolving robot systems is emphasised, stating that in order to have the technology work we also need the science behind it.

Formal Definitions of Unbounded Evolution and Innovation Reveal Universal Mechanisms for Open-Ended Evolution in Dynamical Systems

Open-ended evolution (OEE) is relevant to a variety of biological, artificial and technological systems, but has been challenging to reproduce in silico. Most theoretical efforts focus on key aspects of open-ended evolution as it appears in biology. We recast the problem as a more general one in dynamical systems theory, providing simple criteria for open-ended evolution based on two hallmark features: unbounded evolution and innovation. We define unbounded evolution as patterns that are non-repeating within the expected Poincare recurrence time of an isolated system, and innovation as trajectories not observed in isolated systems. As a case study, we implement novel variants of cellular automata (CA) where the update rules are allowed to vary with time in three alternative ways. Each is capable of generating conditions for open-ended evolution, but vary in their ability to do so. We find that state-dependent dynamics, regarded as a hallmark of life, statistically out-performs other candidate mechanisms, and is the only mechanism to produce open-ended evolution in a scalable manner, essential to the notion of ongoing evolution. This analysis suggests a new framework for unifying mechanisms for generating OEE with features distinctive to life and its artifacts, with broad applicability to biological and artificial systems.

Defining and simulating open-ended novelty: requirements, guidelines, and challenges

The open-endedness of a system is often defined as a continual production of novelty. Here we pin down this concept more fully by defining several types of novelty that a system may exhibit, classified as variation, innovation, and emergence. We then provide a meta-model for including levels of structure in a system's model. From there, we define an architecture suitable for building simulations of open-ended novelty-generating systems and discuss how previously proposed systems fit into this framework. We discuss the design principles applicable to those systems and close with some challenges for the community.

Primordial evolvability: Impasses and challenges

While it is generally agreed that some kind of replicating non-living compounds were the precursors of life, there is much debate over their possible chemical nature. Metabolism-first approaches propose that mutually catalytic sets of simple organic molecules could be capable of self-replication and rudimentary chemical evolution. In particular, the graded autocatalysis replication domain (GARD) model, depicting assemblies of amphiphilic molecules, has received considerable interest. The system propagates compositional information across generations and is suggested to be a target of natural selection. However, evolutionary simulations indicate that the system lacks selectability (i.e. selection has negligible effect on the equilibrium concentrations). We elaborate on the lessons learnt from the example of the GARD model and, more widely, on the issue of evolvability, and discuss the implications for similar metabolism-first scenarios. We found that simple incorporation-type chemistry based on non-covalent bonds, as assumed in GARD, is unlikely to result in alternative autocatalytic cycles when catalytic interactions are randomly distributed. An even more serious problem stems from the lognormal distribution of catalytic factors, causing inherent kinetic instability of such loops, due to the dominance of efficiently catalyzed components that fail to return catalytic aid. Accordingly, the dynamics of the GARD model is dominated by strongly catalytic, but not auto-catalytic, molecules. Without effective autocatalysis, stable hereditary propagation is not possible. Many repetitions and different scaling of the model come to no rescue. Despite all attempts to show the contrary, the GARD model is not evolvable, in contrast to reflexively autocatalytic networks, complemented by rare uncatalyzed reactions and compartmentation. The latter networks, resting on the creation and breakage of chemical bonds, can generate novel ('mutant') autocatalytic loops from a given set of environmentally available compounds. Real chemical reactions that make or break covalent bonds, rather than mere incorporation of components, are necessary for open-ended evolvability. The issue of whether or not several concrete chemical systems (rather than singular curiosities) could realize reflexively autocatalytic macromolecular networks will ultimately determine the relevance of metabolism-first approaches to the origin of life, as stepping stones towards true open-endedness that requires the combination of rich combinatorial chemistry controlled by information stored in template replicators.

From evolutionary computation to the evolution of things

Evolution has provided a source of inspiration for algorithm designers since the birth of computers. The resulting field, evolutionary computation, has been successful in solving engineering tasks ranging in outlook from the molecular to the astronomical. Today, the field is entering a new phase as evolutionary algorithms that take place in hardware are developed, opening up new avenues towards autonomous machines that can adapt to their environment. We discuss how evolutionary computation compares with natural evolution and what its benefits are relative to other computing approaches, and we introduce the emerging area of artificial evolution in physical systems.