Asked to describe AI for Earth and his role at the company, Joppa spoke of the program’s key areas of focus.
“AI for Earth is a cross-company program at Microsoft that works across all of our business divisions, in particular our cloud and AI divisions to deploy Microsoft’s 35 years of AI research and technology in four key areas of environmental sustainability that we care deeply about. And, those areas are agriculture, water, biodiversity, and climate change. I founded and run the program itself.”
Joppa described some examples of specific projects where AI for Earth has launched Microsoft’s technology into the environmental space using aerial imagery to classify land cover types such as forest, water, and urban areas. Classified data (e.g., labeled with identifiable characteristics) is more efficiently ingested by machines and computer algorithms, a necessary component of cutting edge machine learning processes.
Joppa spoke of the difficulties of these deployments, and the necessity for many teams to improve their respective technologies, whether it be software, cloud infrastructure, or computer hardware, to make projects run efficiently and in a cost-effective manner.
“We’ve been using our deep learning library and framework called CNTK, or the Cognitive Toolkit, to build up convolutional deep neural networks and to deploy that algorithm at scale on our cloud-based [artificial intelligence (AI)] infrastructure… Our aspiration is to produce a one-meter resolution land cover map for the United States. And at one-meter resolution that’s about twenty terabytes or so of data and about ten and a half trillion pixels that we have to process. So, we’ve got to figure out a way to be able to do that both quickly and cost efficiently… Project Brainwave… takes advantage of a new kind of processing unit, a new chip type called FPGAs, or field-programmable gate arrays, and using those and deploying a deep learning algorithm on top of those we are able to produce a national land cover map in about ten minutes and for about 42 dollars… The algorithms that had to be built came out of our peer research division with dedicated researchers and PhD interns.”
Though there have been many great advancements in AI technology in recent years, Joppa noted that it will be difficult to widely deploy these technologies in the environmental space until greater focus is put into data preparation.
“We’re launching satellites and other instrumentation platforms for monitoring environment or weather… So, there’s a lot of raw data coming in about the environment, but to be able to take advantage of the AI advances, the infrastructure and the algorithms that the technology industry in particular has been producing over the past decade, we need those data to be labeled… The computers need to learn and to learn they have to be taught, and labeled data are what we teach machines with.”
Aside from data, Joppa sees many more points in the process that will also need drastic improvement to capitalize on gains in AI’s power. In order to deal with the massive amounts of data that AI demands, computation infrastructure will need to rapidly improve and remain cost-efficient.
According to Joppa, this issue is particularly acute in the environmental space, as the number of environmental questions that can and should be asked is substantial but answering these questions requires the processing of a lot of data, which becomes very expensive — which in turn limits the number of issues that can be studied due to cost. In a similar vein, connectivity issues in remote areas need to be addressed so that data collection in these areas can occur as rapidly and conveniently as in well-connected areas.
Additionally, Joppa said that because many improvements in AI and machine learning are made within businesses, yet much of the environmental research is done in the academic sector, it is vital that there is a clear pipeline from academia to businesses and vice versa. Without this dialogue, the gap between what technology can do and what it is doing for environmental research will only continue to grow.
Asked what he sees as the role of businesses in protecting the environment, Joppa stressed making connections between sectors so that they can work together to utilize each other’s available resources.
“I’ll constrain my answer to the technology sector, and Microsoft in particular, and AI for Earth in extreme particular. I think from a technology sector perspective we bear some responsibility for meeting those other sectors, whether it’s the public sector, higher education, government, etc., halfway. For understanding that not everybody has the same background that we do… Not everybody runs a Fortune 500 company and has the resources to put on large, expensive solutions for solving particular problems … One of the ironies of the day is that we ask some of our least resourced organizations, like nonprofits, to solve some of our biggest societal challenges… It doesn’t matter what sector you come from; we all have a role to play. Understanding that it’s on you to make it as easy as possible for all of the other sectors to engage with you, and it’s on the other sectors to do what they need to do to reach out and form those connections as well… What’s interesting about [Microsoft’s] investment is that it was a fifty million-dollar investment over five years. And that five year timeline is really important because… in the technology sector we don’t usually do things on a five year timeline… It gives us that runway and that timeline to think strategically and carefully about how we build those connections and lets the other sectors know that we’re sincere about keeping those connections strong over time.”
Joppa sees many incentives for companies to be environmentally mindful. Apart from societal impact, which he sees as one of the most important, he detailed how environmental problems are perfect testing grounds to stress test and design products.
“I think first and foremost we think it’s the right thing to do… [Microsoft forms] a large part of the digital fabric of society and that comes with it a certain responsibility to do the right thing… Working on some of society’s biggest challenges, and environmental sustainability in particular, means that you are dealing with big problems with big data, and big problems with big compute. And when you put big data and big compute together and you’re a technology company, you’re talking about a big opportunity… to stress test your tech and showcase what our technologies are capable of doing, and if we’re successful have a positive and significant societal impact at the same time… It gives us ideas about new technologies and new use cases that we wouldn’t necessarily have had before… [AI for Earth has] influenced many of those products in their development cycles, and Microsoft has even shipped new products ourselves as a result of this work.”
Another potential incentive for companies to protect the environment is investor pressure, something that Joppa said companies really listen to. Though he doesn’t necessarily see it as the public or an investor’s responsibility to place this pressure, he does see it as a powerful tool if the environment is a topic that an individual is passionate about.
“If they care about it, if they feel like they have a responsibility, then they definitely should [place pressure on companies]. Companies respond to these sorts of things. When we launched AI for Earth, one of the first sets of calls that I took were from some of our major investors calling us to congratulate us, and to ask more questions. Investors are an important factor in how businesses make decisions, and companies listen to them. I think that if investors have a topic that they care passionately about, and that more importantly they think has the potential to impact the bottom line of the company that they’re investing in, then for sure they should apply that pressure. I do believe that environmental sustainability is one [topic] investors are starting to pay attention to from a bottom line perspective”
Tyler Johnson is a second-year student intending to study Computer Science at Stanford University. Himanshu Gupta is the former advisory fellow to Vice President Al Gore and a dual degree student in MBA/MS in Climate Science at Stanford University.
Originally published at sej.stanford.edu.