Mechanical Turk

A few years ago, Amazon came out with a project called Mechanical Turk. It's a service by which people can engage the labor of others to do small tasks that computers are not very good at. For example, if you wanted to look up the addresses of certain companies on the web, it would be a very challenging task for a computer, but not that hard for a human. However, it would also not be that interesting, and you probably wouldn't expect to get a job looking up addresses full-time. Enter Amazon's Mechanical Turk.

This Spring, once I put up the initial version of the Garden and told a few friends about it, progress was a bit sluggish, to say the least. After all, real evolution takes place over millions of years and millions of organisms. Even dog-breeding takes many generations to start to see a new breed emerge. So, although I can sit at my computer and click a few hundred plants, not a whole lot of change occurs.

So, I decided to make use of the Mechanical Turk to jump-start my project. I thought of it as another art supply, really. Not that expensive, but not something you'd spend money on unless you needed to.

I created a half a dozen tasks on Amazon's site. Each one described my project briefly, asked workers to vote for at least 25 plants and paid out somewhere between 25 and 50 cents. Using the Mechanical Turk, I was able to generate over 10,000 new plants, and it didn't cost me much more than buying a few tubes of paint and a brush at the local art supply store.

Because I wanted my workers to have a more streamlined work-flow than what you see in the Garden proper, I developed a completely different interface. I called it the Voting Booth, and this is where I sent workers to vote for plants. Check it out if you're interested:

The Voting Booth

People who worked for the Garden, seemed to enjoy the work. After all, voting for odd, plant-like shapes, is not uninteresting work. In fact, many people got so into the task that they voted for many more plants than was required.

It's an intriguing service and one that I hope to use again. In the next version of the Garden, the old genetic code will no longer work, so I'll need to jump-start the next version with new plants. (I do plan to keep the old Garden around, but the new one will be much improved.)

Gallery From 1st Garden Show

I finally got around to putting up a gallery of the prints I made from my favorite plants in the Garden for the June open studio at the Sunny Arms. I sold about a third of these and will likely show them again this Fall, along with whatever I come up with for the next version of the Garden.

Gallery of Garden Prints

Recombination - Bringing Sex To The Garden

I've spent the past few days working on a new plant genome. To date, I've added a new stem-drawing algorithm and made individual genes more stable. I've also enlarged a plant's overall growing space.

Today I created an interface to experiment with recombination. Up until now, all plant reproduction has been asexual. Each plant simply copies itself to create a new plant. Copy errors occur, which create variations, but ultimately, offspring are just slightly different versions of their parents. Today I experimented with crossing two different plants.

In the image below, two plants were crossed to produce a third:

Of course, the parents must have somewhat similar genes, or, when crossed, the structural changes are too drastic and the offspring look nothing like either parent. In some cases, crossing dissimilar parents results in a plant genome that doesn't draw anything at all, just like if you were to cross an orange tree with a cactus. There's a reason species design their reproductive organs such that only members of similar species can mate.

(This brings up interface design questions. Before, a visitor could click a plant and move on. For recombination, two plants must be selected. And, if the visitor tries to cross two very different plants, the result may not be all that compelling.)