AI Weekly: AI researchers publish toolkit to promote AI that helps achieve sustainability goals
Missed a Future of Work Summit session? Head over to our Future of Work Summit on-demand library to stream.
While discussions about AI often focus on the technology’s commercial potential, more and more researchers are exploring ways to harness AI to drive societal change. Among others, Facebook’s chief AI scientist, Yann LeCun, and Google Brain co-founder, Andrew Ng, have argued that mitigating climate change and promoting energy efficiency are preeminent challenges for AI researchers.
In this vein, researchers from the Montreal AI Ethics Institute have proposed a framework designed to quantify the social impact of AI through techniques such as computationally efficient machine learning. An IBM project is providing cropping recommendations from digital farm “twins” that simulate future soil conditions of real-world crops. Other researchers are using AI-generated imagery to help visualize climate change, and nonprofits like WattTime are working to reduce household carbon footprints by automating when electric vehicles, thermostats and appliances are active depending on where renewable energy is available.
Seeking to spur new explorations in the field, a group from the Stanford Sustainability and Artificial Intelligence Lab released this week (to coincide with NeurIPS 2021) a reference data set called SustainBench for tracking Sustainable Development Goals (SDGs) including agriculture, health, and education using machine learning. As the co-authors told VentureBeat in an interview, the goal is three-fold: (1) lower barriers to entry for researchers to help achieve the SDGs; (2) provide metrics to evaluate SDG tracking algorithms, and (3) encourage the development of methods where improving AI model performance facilitates progress towards the SDGs.
“SustainBench was the natural result of the many research projects that [we’ve] worked over the past half-decade. The driving force behind these research projects has always been the lack of large, high-quality labeled datasets to measure progress towards the United Nations Sustainable Development Goals (UN SDGs), which compelled us to propose creative machine learning techniques to overcome label scarcity. “, said the co-authors.”[H]After accumulating enough experience working with datasets from various sustainability domains, we realized earlier this year that we were well placed to share our expertise on the data side of the machine learning equation. … Indeed, we are not aware of any previous datasets focused on sustainability. with similar size and scale as SustainBench.
Progress towards the SDGs has historically been measured through civil registration, population-based surveys and government-orchestrated censuses. However, data collection is costly, leading many countries to wait decades between taking action on SDG indicators. It is estimated that only half of the SDG indicators have regular data from more than half of the countries in the world, which limits the ability of the international community to track progress towards the SDGs.
“For example, at the start of the COVID-19 pandemic, many developing countries implemented their own cash transfer programs, similar to IRS direct cash payments in the United States. However…data on household wealth and income in developing countries are often unreliable or unavailable,” the co-authors said.
However, AI innovations have shown promise in helping to fill data gaps. Data from satellite imagery, social media posts and smartphones can be used to train models to predict things like poverty, annual land cover, deforestation, agricultural cultivation patterns, crop yields and even the location and impact of natural disasters. For example, the governments of Bangladesh, Mozambique, Nigeria, Togo, and Uganda have used machine learning-based poverty and cropland maps to direct economic aid to their most vulnerable populations. during the pandemic.
But progress has been hampered by challenges, including a lack of expertise and a lack of data for low-income countries. With SustainBench, the Stanford researchers—along with contributors from Caltech, UC Berkeley, and Carnegie Mellon—hope to provide a starting point for training machine learning models that can help measure SDG metrics and have a wide range of insights. applications for real-world tasks.
SustainBench contains a suite of 15 benchmark tasks across seven SDGs taken from the United Nations, including good health and well-being, quality education, and clean water and sanitation. Beyond that, SustainBench offers tasks for machine learning challenges that span 119 countries, each designed to promote the development of SDG measurement methods on real-world data.
The co-authors caution that AI-based approaches should complement, rather than replace, ground-based data collection. They point out that ground truth data is needed to train models in the first place, and that even the best sensor data can only capture some – but not all – of the results of interest. But AI, they still believe, can be useful for measuring sustainability metrics in regions where ground-truth metrics are scarce or unavailable.
“[SDG] indicators have huge implications for policy makers, but “key data is scarce, and often scarcer where it is most needed,” as several members of our team wrote in a recent Science review article. By using abundant, cheap, and frequently updated sensor data as inputs, AI can help fill these data gaps. These input data sources include publicly available satellite imagery, street-level imagery, Wikipedia entries, and cell phone records, among others,” the co-authors said.
In the short term, the co-authors say they are focused on raising awareness of SustainBench within the machine learning community. Future versions of SustainBench are being planned, potentially with additional datasets and AI benchmarks.
“Two technical challenges present themselves to us. The first challenge is to develop machine learning models capable of reasoning on multimodal data. Today, most AI models tend to work with unique data modalities (e.g. only satellite images or only text), but sensor data often comes in many forms… The second The challenge is to design models that can take advantage of the large amount of unlabeled sensors. data, compared to sparse ground-truth labels,” the co-authors said. “On a non-technical level, we also see a challenge in getting the broader machine learning community to focus more effort on sustainability applications…As we mentioned earlier, we hope that SustainBench will enable machine learning researchers to more readily recognize the role and challenges of machine learning for sustainability applications.
For AI coverage, send topical advice to Kyle Wiggers – and be sure to subscribe to the AI Weekly newsletter and bookmark our AI channel, The Machine.
Thanks for reading,
VentureBeat’s mission is to be a digital marketplace for technical decision makers to learn about transformative technology and transact. Our site provides essential information on data technologies and strategies to guide you in the direction of your organizations. We invite you to become a member of our community, to access:
- up-to-date information on topics that interest you
- our newsletters
- gated thought leader content and discounted access to our popular events, such as Transform 2021: Learn more
- networking features, and more
Become a member