The Westmont Inspired Computing Lab just got a “seed” grant to launch a greenhouse powered by Artificial Intelligence! Our plan is to see if a machine learning algorithm with access to sensor data can beat a human at growing food in drought conditions. This work is inspired by a paper that I coauthored with colleagues about the idea of Computational Agroecology and is in partnership with Joy Patterson.
Why build a greenhouse?
It is widely understood that today’s industrial agriculture, which is central to our civilization, but reliant on non-renewable resources (e.g., fossil fuels, chemical fertilizers, and pesticides) and causing increasing damage to the ecosystem effects (e.g., erosion, oceanic dead zones, greenhouse gas emissions), cannot persist indefinitely [The Ecology of Agoecosystems, Vandermeer 2011]. However, none of the existing alternative agricultural systems is currently able to meet the challenge of feeding the 7.4 billion people in the world, much less the added billions that demographers project. We believe that we have an opportunity at this juncture to change the way future food systems are structured through the design of technologies to implement sustainable new ecosystems for food production. This is a broad problem, but locally in Southern California we have the additional pressing problem of dealing with a very limited water supply.
Scientists in the field of agroecology have identified approaches to reconcile human food needs with the broader planetary ecosystem and its limits. Agroecology has failed to scale due to the information and logistical challenges of adapting agroecological techniques to millions of unique potential sites of food production. Especially in urban environments where there is quite a bit of space to grow food, but a great variety of site specific details. This is in contrast to industrial agriculture’s approach of using external inputs to standardize sites and increase yields: think hundreds of acres of monoculture.
We have proposed a new approach, Computational Agroecology, that seeks to make agroecology viable at a global scale. Our approach entails computational systems that enable the design of new ecosystems for food production using agroecological techniques.
Internet of Things
This greenhouse will combine technology developed for the Internet of Things (IoT) with artificial intelligence and hydroponics to automatically learn watering policies that compete with hand-made policies to grow more food, faster, with less water.
We are going to prototype a small temporary greenhouse that has 5 eight-foot rain gutters in it, arranged in a stair step pattern. Each gutter will contain soil and will be planted with lettuce seeds for this experiment. Each gutter will have an associated water tank and solenoid that will release water into the gutter to hydrate the plants as dictated by a micro-controller. Each gutter will have a competing algorithm controlling it whose strategy was learned from the previous round of lettuce growing.
Ultimately we would like to deploy many greenhouses around Santa Barbara, controlled by Raspberry Pi’s, to enable people in urban environments to have a very local food supply.