This post is a continuation of my previous post “Changes in the Natural Philosophy of Language“. I will discuss some ideas introduced by the first of the three researchers who I mentioned there: Ken Stanley. His work is on the topic of evolutionary systems — that is, systems that resemble the process of biological evolution. Evolutionary analogies show up in a wide variety of open-ended systems, such as in the case of historical linguistics, as well as the transmission of beliefs and ideas among populations, so his contributions to understanding these systems have interdisciplinary implications.

I’m sure anyone who reads this will be familiar with the two main components of evolution: mutation and selection. Genetic mutation occurs over long stretches of time, where genes are non-disruptively modified and passed on through generations, contributing to biological diversity. Selection occurs when environments shift and new pressures change the likelihood that individuals with certain genetic traits will have offspring. This is a generalized explanation of the evolutionary mechanism that led from the Earth’s first single celled organisms to today’s complex life.

This push-and-pull explanation is easy enough to grasp, but as you look into the details, it becomes clear that there’s more to the story. Common examples of evolution in recorded history include cases of color adaptation in insects or the selective breeding of crops and animals, but this displays only half of the evolutionary process: selection. Mutations in complex life can take an inordinate amount of time to build up, so cycles of evolution have historically only been observed in smaller, “r-selected” (prolific) organisms such as bacteria. That may have changed somewhat with the advancement of tools and processes for genetic modification, but even so, we can hardly reproduce even a microscopic portion of the effect of life’s billions of years of evolution. The point I’m leading to here is that the evolution of an arm or an eye — or really any significantly complex structure — still falls outside the scope of our common understanding. You can’t just “select for an arm” or “select for an eye” the way you can for color or size.

What we ought have a commonly accepted explanation for, is how one can get full-on, qualitative changes from evolutionary systems; not just single-axis, quantitative changes. With computers, we can simulate evolutionary systems mathematically and study them directly without disrupting or relying on other living systems for our discoveries. Ken Stanley and Stephen Wolfram (who I will discuss in my next post) both use simulations to discover pieces of this explanation. In the case of Ken Stanley, him and his colleagues developed an “Interactive Evolutionary Computation” (IEC) called “Picbreeder“, which allows users to select from a collection of images, each with their own “genetics”, and produce a new generation of images from them with the addition of several mutations. Ken Stanley’s analysis of the data from this long-running experiment shows something special about this system, which is that users would come to find the most interesting images unintentionally, by building on each other’s discoveries. Users who intentionally explicitly seek out images from the set of chaotic, uninteresting blobs do not find them. Not only this, but the genetic lineages leading to these images often demonstrate no obviously shared features linking ancestors with the final image — except for that they are interesting in a very broad sense of the word.

In his own words (at 19m15s of an interview with YouTube content creator Sethbling):

“So, in fact, these were extreme needles in a haystack, like, the 99.9999999% of this space is just total garbage blobs. But our users were consistently finding these extreme needles in a haystack. And the question is ‘why and how?’ Because, you see, that question is not just about Picbreeder; it’s about explaining open-endedness itself. Like, how do systems like this work when there are these extremely complex and obscure discoveries lurking in vast spaces of total garbage?”

Owing to their having access to all the data for the evolutions, he explains further:

“It turned out that there was a very important principle that we discovered which is totally counter-intuitive, and actually kinda shocking, which is that the only way to find things on Picbreeder was by not looking for them. […] I called it the “Objective Paradox” eventually; it’s this idea that setting an objective — like I want a picture of an apple — will actually stop you from achieving your objective. Like, the only way you can get an image of an apple or a butterfly or a car is by not trying to get those images.”

What?

Yes; such a concept can be intuitively understood by artists (as he also discusses in the interview), who must always fight through questions about their work’s profitability and purpose – often with little to say to those who only see through the lens of objective. Others might simply notice this theme in their life or in nature, but not have any need to name it. In fact, I would bet that anyone reading this blog would have a way to relate the Objective Paradox to their own life; it’s a very human experience, after all. Ken Stanley and his team demonstrate this phenomenon reproducibly in a generalized evolutionary system, and therefore create a scientific foundation for us to categorize, understand, and interact with these unique evolutionary spaces. I personally believe that this empirical demonstration could be revolutionary on a cultural level too, because it finally gives a basic, easily explainable, scientific argument supporting human exploration through art and science – simply because it is interesting. And that is the only way to progress through certain evolutionary spaces.

This is not to say that individuals or groups should do away with goal-orientedness, because it is indeed effective at generating predictable outcomes. However, it shows that approaching exploration without a goal in mind, learning how to play, is a practical necessity when faced with complex, open-ended systems. It even shows, indirectly, that the aforementioned qualitative changes from biological evolution do NOT get selected for directly. A worm wishing for an arm just won’t get there in a straight line! This research provides us with evidence for the procedural development of these qualities through stages of completely unrelated development.

This leaves us with a burning question, though, which is: “what on Earth does ‘interesting’ mean to an evolutionary system?” Picbreeder depends on a human interpretation of visual “interestingness”, so you’ve got to wonder what sort of intelligent system could guide the process of biological evolution toward anything in particular, let alone… this! Us! Well, the next researcher, Stephen Wolfram, focuses on how mathematics itself can underpin this apparent directionality, acting as a trellis on which the creeping vines of mutation feel out new territory. I am looking forward to tackling those details in the next one.

Thank you for reading and bearing with my longwinded approach to introducing a new way to look at language — I said it was going to be paradigmatic, so I hope you’ll forgive me for the length!