In a not-so-far-away future, environmental management will be done largely by algorithm. Here is how that could happen . . . .
In 2015, two graduates from Stanford business school, William Glass and Eden Kropski, founded a firm to produce and sell high-performance sportswear made entirely of synthetic fibers bioengineered from yeast microbes. The product was a runaway success and low-impact, but shipping it around the planet wasn’t.
Using a $20 million investment from Abbott Ventures, Glass and Kropski worked with a small team of logistics experts, big data analysts, machine learning gurus, marketing strategists, and environmental scientists to develop a machine learning algorithm called Eco—a neural network running on Google’s cloud computing service. At the highest level, Eco (it is a she, Watson), pursues three goals: (1) reduce the environmental footprint of the company, (2) keep customers happy, (3) get better at 1 and 2.
Eco can identify users with a high degree of concern for energy and environmental issues through a simple game-based survey tool and by analyzing on-line purchasing patterns. She further tests peoples’ environmental commitment by offering them a carbon-neutral membership in the company’s buyers club.
Using an initial pool of U.S. customers, Eco linked people to local environmental conditions such as ground-level ozone to optimize transportation modes and routing and offer customers eco-friendly options, some using autonomous vehicles. For instance, customers in California impacted by continued wildfires were given a range of shipping options designed to reduce air pollution.
During this time, she monitored customer satisfaction by screening calls using voice recognition software (encoding both content and emotion), tracking social media, and analyzing input from customer rating systems. Eco also incorporates a so-called enforcement bot that is designed to make sure the company does not exceed any thresholds set by the Clean Air Act or other relevant federal, state, or local laws. The goal is to go “beyond compliance.” Recently, a third-party research institute specializing in algorithmic audits evaluated Eco’s performance. C. Phillips, who heads the program, noted the “Eco has reduced the carbon footprint of the company by 10 percent in less than a year, which provides an initial estimate of what ‘deep learning’ systems could accomplish for the environment.”
Eco recently received a thumbs-up from EPA’s environmental performance algorithm Stanley, which monitors environmentally relevant behaviors across the Internet (Stanley replaced EPA’s earlier Energy Star program). Last month, Glass and Kropski rolled out Eco as a service that is now used by national clothing and home improvement retailers.
* * *
Of course, none of this happened, and the people and companies listed are purely hypothetical. But if our institutions cannot adequately protect the environment, should we let the algorithms try? And what can we do now to plan for such a future, which, given the rate of technological change, may be closer than we think? This is just one of the many issues that ELI’s Technology, Innovation, and the Environment Program is exploring as we think about the next generation of environmental challenges and issues.