“What I cannot create, I do not understand.”
– Richard Feynman
Placing artificial intelligence in the arc of the history of life on Earth depends upon a firm understanding of life itself. Philosophers have grappled with this problem for millennia, but in the past few decades a growing field of contenders have begun tackling it with new approaches. At the Alife 2019 conference, practitioners of artificial life use tools such as robots, biology labs, and computer simulations to model processes that we instinctively associate with living creatures. As each model produces a lifelike behavior, our understanding of the puzzle of life should grow more complete.
The puzzle is mind-bogglingly complex, though. International Society of Artificial Life president Charles Ofria said when interviewed that, “humans make some very complex technical artifacts – some which no single human can understand. The complexity of these pales by comparison with those of nature.” As PhD candidate Penelope Rainford documented in one of the conference workshops, critical structures of life are operating in populations spanning dozens of orders of magnitude, from single organic molecules to the group of a hundred thousand trillion, trillion atoms that we call a blue whale. If we consider their sizes, extending from proteins to ecosystems, life on earth rises fifteen orders of magnitude in linear scale. The understanding of life is the understanding of an immense tower of hierarchical structure.
The Alifers build their tower on three pillars they characterize as hard, soft, and wet. Hard robots scurry about like animals. Software systems exhibit behavioral signatures of life. And wet concoctions in beakers host the chemical reactions and cell-like structures of synthetic life. If they succeed in completing the tower, they will be able to make the convincing claim of understanding life by having conjured it from inanimate parts, perhaps in new forms or in new substrates.
The conference participants are quick to reassure that life has not sprung from their laboratories and computers just yet. The comforting theme of this year’s event is the application of artificial life to help solve societal challenges. Professor Barry McMullin’s keynote address reminded participants that the models of the Limits to Growth project had effectively forecast population, pollution, and industrial output nearly half a century ago. Without some awareness and meaningful intervention, the same models forecast dire consequences. McMullin said that the overshoot of climate change was, “not like going off a cliff. It is like venturing into a minefield.” As a hopeful counterpoint, organizer Harold Fellerman said that the youthful spirit of this year’s conference was ready to apply the insights of artificial life to the situation, both in theory and in practice. Participants were already doing their part. By one estimate, thirty-five percent of the CO2 emissions that a PhD candidate produces are in conference travel. A number of sessions at Alife 2019 connected from remote sites via videoconferencing to reduce the conference’s impact on emissions. On the theoretical side, Alifers presented work illuminating social learning, evolution of skills, cultural evolution, and agent-based models of human behavior. While no single talk illuminated a path to avoid the Limits to Growth projections, many offered pieces of the puzzle of our collective behavior.
Part of the perceived magic of life is the apparent improbability of so many pieces coming together randomly. Astrophysicist Fred Hoyle is reported to have said that the, “probability of life originating on Earth is no greater than the chance that a hurricane, sweeping through a scrapyard, would have the luck to assemble a Boeing 747.” After acknowledging Hoyle’s accomplishments at the beginning of his talk, Professor Roberto Serra set to work systematically chipping away at the astrophysicist’s misunderstanding of the probability of the emergence of the processes of life. Serra focused on the complex processes within the realm of cell replication. As he explained, early protocells would not have had the benefit of the entire contemporary array of these mechanisms to successfully synchronize membrane division with genetic replication. After a short treatment of the useful emergent behaviors of simple amphiphile membranes in water, Serra shared a number of models of reactions of autocatalytic sets. Synchronization between replicator and membrane growth occurred in model after model with varying reactants, concentrations, perturbations and membranes leading to the conclusion that, “synchronization is a widespread emergent property.”
This discovery that the production of lifelike behaviors is more probable and simple than one might initially expect resonated throughout the ‘wet’ sessions from the world of molecules, structures, and cells of synthetic biology. Results presented ranged in scale from chemical reactions that catalyzed their own reactants, to artificial cells, to dancing droplets, to whole ecosystems producing food or electric power.
In several sessions, presenters showed simple sets of chemicals, often inorganic, engaging in lifelike behavior. Professor Julyan Cartwright showed Stephane Querbes’s sumptuous videos of growing chemical gardens and explained how their complex behaviors emerged from adding metal salts to a solution of sodium silicate. Geoffrey Cooper shared the results of Gutierrez et al’s high-content system for robotically testing and analyzing various concentrations of a small set of compounds. The resulting droplets exhibited a wide variety of behaviors reminiscent of living systems: they would move about in Petri dishes; they would diffuse and fuse in cycles; they would pulse and rotate, attract and connect, divide and interact with walls of their dish. Professor Shinpei Tanaka explored and explained the behaviors of droplets that would propel themselves as individuals and engage in complex dances collectively. He demonstrated a computer model of the reactants that exhibited much the same behavior. Silvia Holler showed moving droplet systems able to transport and release living and non-living objects. She demonstrated her colleague Jitka Cejkova’s droplet system that navigates mazes. Time and again, the presenters demonstrated intriguing behavior emerging from systems of a few compounds.
The Alifers make a compelling case that the tower of understanding that they are constructing is made of known inanimate parts and processes. And they have shown that the improbable construction of an airplane by a hurricane is not the model for the construction of the tower. An abundance of signature behaviors of life can spring from the interaction of small sets of these parts – salts dropped into solutions grow into elaborate shapes, and droplets dance, divide and navigate mazes. The bricks and mortar of the tower, the molecules and reactions of life, have a tendency to self-organize. They snap together as if equipped with magnets. And the larger structures that emerge have their own self-constructing, autocatalytic properties that generate behaviors and structures at ever larger scales. These properties make it less mysterious that the tower of life can emerge from inanimate molecules and processes down below. Unfortunately, the scales of life telescope upward through fifteen orders of magnitude. The most optimistic of chemists, biologists, and physicists at the conference did not see a path from inanimate molecules to living organisms occurring in a laboratory even years hence.
The top-down approach
But the new tower of artificial life does not have to be built from the foundation up. Professor Kate Adamala presented her group’s provocative approach of creating whole synthetic cells the size of bacteria in populations of 100 million per experiment. She described her synthetic cells as, “bioreactors that have some but not all of the functions of a live cell.” Her group at the University of Minnesota uses a microfluidic system to inject the content of formerly living cells into a manufactured lipid bilayer membrane of tunable permeability. The material may come from multiple taxonomic domains, may contain sequenced genetic material, and engages in effective protein synthesis. Adamala showed a list of drugs that her system had produced in therapeutic concentrations. While the cells do not function indefinitely and complete a life cycle that includes cell division, they are durable enough to be handled and compelled to interact and even merge in a controlled process resembling sexual reproduction. She explained, “We might be able to build synthetic live cells before we gain total understanding of natural life chemistry. Artificial evolution in a test tube will go on whether we understand all processes of it or not. We like to think that we know where every atom in a synthetic cell belongs, but at certain level of complexity (that approaching life) we might not be able to hold on to that understanding. Still, being able to engineer living system from scratch will teach us a lot about natural life, and hopefully we’ll be able to make Dr. Feynman proud.”
Adamala viewed the possibility that a living cell line would emerge from the “relatively fragile system,” as a positive but unlikely occurrence. Adamala showed a technique by which populations of cells which each contain a portion of a metabolic cycle could be literally stacked to create a complete pathway. She illustrated using her lab’s ‘mating’ process to construct a genetic circuit from populations containing portions of the circuit. She described creating a population of ‘pluripotent’ synthetic cells being driven to differentiate into subpopulations producing different protein outputs. By looking down from the scale of whole cells, Adamala hopes to shed light on essential components of life as we know it, such as ribosome function – an activity difficult to study in living cells because of their propensity to die when altered. Interviewed, she said that a, “faster ribosome would confer a tremendous advantage.”
The traditional software pillar and theory of artificial life made a strong showing at the conference. Charles Ofria noted the promise of the AvidaEd development platform to provide a common workspace for education and modeling of Alife systems. He described the field as being on the verge of understanding the hierarchical transitions that are signatures of life, such as the development of multicellularity. Olaf Witkowski of Cross Labs also was optimistic about growing understanding spanning several disciplines, including artificial life. Despite his talk’s provocative title of “When AI Takes Over: Transfer of Control in Distributed Information Flows,” Witkowski was relatively sanguine about the risks of artificial intelligence. He explained, “Alife… may be the most promising way to tackle the next challenges in AI. We built pumps before we could understand the heart, and sonars before we understood bats. In that sense, Alife goes beyond AI in seriously attempt to understand intelligence in living systems.”
So the tower of understanding grows in Newcastle at Alife 2019. Researchers busily lay the mortar and fill in the structure in scales from molecular to planetart. Many direct their efforts toward the conference’s theme of addressing societal challenges. Others study life’s first principles. In the unfolding sixth extinction, some gap in understanding between these researchers and the agents directing the global economy hastens life toward punctuated equilibrium. Whether the researchers’ insights will accelerate that punctuation or forestall it remains to be seen. Perhaps the answer will be more clear when the community achieves the goal to which Feynmann aspired – understanding by creating. Organizers and presenters say that despite the field’s early aspirations, truly open-ended evolution has not arisen to the satisfaction of the consensus. Participants prognosticated that artificial life in all its forms was still years, perhaps many, in the future. Organizer Rudolf Füchslin questioned whether there was a clear metric for open-endedness beyond the “absolutely amazing” reaction that a synthetic life-form might elicit from researchers if it were ever presented. Charles Ofria offered an open-ended, koan-like definition of artificial life itself as, “that which we see in nature but not yet in computational systems.” Apparently, just as life itself continually infiltrates inanimate matter toward an indeterminate boundary, artificial life must march ever deeper into computational space.