Deep Swamp is a triptych of semi-inundated environments that gather together wetland life forms and artificially intelligent software agents. The agents, Nicholas, Hans, and Harrison, patiently watch their swampy territories and modify the conditions in them. Every few minutes they adjust the light, water flow, fog, and nutrients to try to engineer their environments for different goals. Harrison aims for a natural looking wetland, Hans is trying to produce a work of art, and Nicholas simply wants attention. Using a deep learning approach, each software has learnt what their goal might look like by parsing thousands of tagged and categorized images downloaded from online collections. Over and over they try new combinations of settings, photograph themselves, and then choose permeations that bring them closer to their programmatic desires.

As territories flood and melt, dehydrate and erode, paradigms of environmental protection and conservation give way to those of management, engineering, and strategic intervention. In this sense, as landscape architects Bradley Cantrell, Laura Martin, and Erle Ellis argue, conserving wilderness has become more about maintaining autonomous ecological processes rather than the preservation of historic conditions. If new “wilderness” is the absence of explicit human intervention, what would it mean to have autonomous computational systems sustain wild places?

Practices of environmental engineering, whether by machine or human intelligence, raise thorny questions of optimization. Optimization is the problem of making the best or most effective use of a situation or resource. It is the attempt to find the best solution from all feasible solutions. However what is considered “best” or “most effective” elicits the question of best for who? As ecological calamity is met with environmental engineering, what should environments be optimized for?