Euconoclastic blog series
Stephen Casper, email@example.com
People will sometimes argue for the existence of a creator god by likening the world to a watch: If you were walking on the sidewalk one day and saw a watch on the ground, it would be much better to assume that this watch was made by a watchmaker instead of being some Boltzmann miracle. Similarly, if you stumbled upon a planet with complex life, you surely must assume that there’s some life-maker and not that life just spontaneously arose.
Not so fast.
This argument has some problems. Among them are the anthropic principle, the ability of simple Darwinian structures to evolve complexity, or counterevidence, but these ones aren’t my focus here. Think life is like a watch? Nope. It’s more like a magical burrito.
I’ll get back to that in a bit, but first I want to define life. There are a lot of definitions out there, but let me toss mine into the heap. I think that it’s useful to think of any system that performs computation and whose computational hardware undergoes evolution as being alive.
By computation, I mean the process by which a system predictably achieves different states (outputs) depending on environmental inputs in a way that lends itself to a conceptual decomposition of heterogeneous steps and modules. Let me explain.
1. I say “predictably” because completely random processes do not lend themselves to the sort of structured information transfer that computation is, but interesting computations need not be completely deterministic. If a certain input leads to a certain one of two outputs 51% of the time, that’s a process that carries information.
2. I say “achieving different states (outputs) depending on environmental inputs” because that’s pretty uncontroversially part of anyone’s definition of computation.
3. I say something that “lends itself to a conceptual decomposition” because computation isn’t interesting because of substrate-specific phenomena going on under the hood of a computer — it’s interesting because of information being processed through abstractable means. This is also pretty uncontroversially part of (or implied by) anyone’s definition of a computer.
4. Finally, I emphasize “heterogeneous steps and modules” because if steps and modules were allowed to be homogeneous, we would be considering some very uninteresting processes to be computation. For example, a line of n dominos or anything that they could be an analogy for could have an output in the form of the last domino that tells whether the first was pushed, but I think it would be inane to say that this is an n-step computational process (although it could be a single buffer-type step/module in a larger system).
By evolution, I mean an optimization algorithm that involves taking a population of candidate solutions, randomly selecting them with bias toward ones that are more fit according to some stochastic survival function (which can be static or dynamic over time), creating mutated offspring from them (sometimes with sex), and repeating. Importantly, evolution isn’t a top-town optimization process. It’s a bottom-up one.
So why do I like defining life as being something that computes and undergoes evolution? I just think it encapsulates well what biologists are/should be interested in. Computation is key to a system being interesting, and evolution is the hallmark of biology. If something doesn’t compute, according to my definition, it doesn’t really react to its environment according to some process from within — it’s just subject to its environment, and I don’t think that’s worth considering to be life. If something doesn’t undergo evolution, then I think it should be considered just a computer (a superset of lifeforms) because it doesn’t have an organic nature.
In the end, this is just another definition, and it can’t really be more right or wrong than any other, but I think it highlights some useful features and avoids the lame temptation to just define life as being something that’s in Eukarya, Archaea, or Bacteria. Notably, this definition doesn’t care about substrate and includes viruses and some data structures/functions that are inside a larger computational environment. This definition does NOT include static data structures inside a genetic algorithm, most software, a paper you make copies of, pryons, etc.
If we accept this definition, it makes biology the study of computers whose function is garnered through bottom-up evolutionary processes. There is no designer in evolution, and consequently there is nothing giving any pressure toward making living systems easily-understandable or engineerable. And THAT’s what makes biology hard. It’s the scientific pursuit of understanding and engineering systems that were designed by a process that couldn’t care less about understandability or engineerability.
Now allow me to contrast organic design and rational design. Organic design is guided by evolution, and rational design is implemented by an algorithm that optimizes ins a less blind way. And importantly, if you show me a computer, I could reliably tell if it’s organically designed or rationally designed because organic design does not lend itself to easy abstraction while rationally-designed models maintain their explanatory power through abstraction.
Two good examples of how complex organic systems can be are the challenges of understanding gene regulation and creating synthetic cells.
Cells have many ways of regulating genes: enhancers, promoters, ribosome binding sites, terminators, coding sequences, codon bias, splicing, RNA folding, tRNA concentrations, release factors, chaperone proteins, steric regulation, allosteric regulation, dimerization, untargeted degradation, targeted degradation, phosphorylation, dephosphorylation, and…you get the point. Biologists have yet to come close to understanding all of the regulatory mechanisms working in tandem in chaotic, organic cells. But evolution just blindly hums along, feeling its way through the chaotic fitness landscape without needing to understand or model what it’d doing.
Next, consider a recent paradigm for synthesizing rationally-designed synthetic life. It’s hard as hell. Some of the best we’ve done to date isn’t even that impressive: basically just in vitro DNA translation in soap bubbles. When we try to rationally engineer life, it just doesn’t tend to go so well because evidently, being alive tends to require some serious complexities.
Organic design makes a fantastic mess. It’s the reason why life is full of biodiversity, maladaptation, pseudogenes, transposons, cancer, immensely complex gene regulation systems, symbiosis, and a plethora of repurposed or vestigial structures. Rational humans can’t hope to understand and appreciate all of biology’s complexities anytime soon, and even if we understood systems so well as to be able to synthesize life, our top-down design would lend itself to a lot of rules and regularities. In other words, synthetic life would be a lot like a watch. But evolution made organic life, and although it’s messy and confusing, it works pretty damn well. So hats off to evolution and the magical burritos that are organic lifeforms.
I think the difficulty of grasping and modeling all of the complex, organic complexities in biology is a good example of how feeble human intelligence is. We’re great at understanding things that lend themselves to simple models, but for real-life systems, design spaces are so complex and high in dimensionality that we can’t hope to understand them well or non-empirically characterize portions of them. Sometimes we use stochasticity in our models to account for processes that would be too difficult to model with precision, but that can only get us so far. If and when we develop superintelligence, I think it’s possible that one of its hallmarks may be powerful capabilities for efficiently understanding systems that escape rational models.