PALO ALTO (CBS SF) -- What if you could use artificial intelligence to measure global poverty?
Well, a group of Stanford University researchers say they've been doing just that and that the implications could be far-reaching.
The researchers, with a diverse array of expertise including computer science, electrical engineering and earth systems science, found that they can use artificial intelligence to identify global poverty zones by comparing daytime against nighttime satellite images.
During the day, the satellites pick up information about a locations' infrastructure, including man-made structures and roads, while at night it picks up information about where electricity, in the form of emitted light, is being used.
The low-tech alternative for measuring poverty currently entails conducting time-intensive and costly ground surveys in remote locales.
In a paper released by the researchers last week, they explain that poverty data is scarce and often unreliable in developing countries, leading to major obstacles for those trying to provide sustainable development, food security, and disaster relief.
"Remote sensing data such as high-resolution satellite imagery, on the other hand, is becoming increasingly available and inexpensive," the Stanford researchers state in their paper. "Unfortunately, such data is highly unstructured and currently no techniques exist to automatically extract useful insights to inform policy decisions and help direct humanitarian efforts."
The World Bank agrees. Andrea Coppola, senior country economist at The World Bank noted last year that two-fifths of countries fail to conduct a household survey every five years and that even when household surveys are conducted, many of the poorest people tend to be located areas that are difficult for survey takers to reach. Data quality is also generally low.
Coppola notes that, "On the contrary, satellites gather data at a constant rate throughout the year, regardless of physical or social hazards."
So, the Stanford researchers came up with a new type of machine learning to analyze millions of high-resolution satellite images and provide accurate data necessary to pinpoint areas of poverty.
The researchers designed a computational model, created an objective, and then fed it raw data. Then, they let the computer take over.
The daytime imagery allows the model to predict distribution and intensity of nighttime lights to help predict relative prosperity.
"We demonstrate that these learned features are highly informative for poverty mapping, even approaching the predictive performance of survey data collected in the field," the paper states.
Since the model has more than 50 million tunable, data-learned parameters, and is constantly updating based on details gleaned from raw data, the Stanford team doesn't actually know how the model is doing what it does.
"To a very real degree we only have an intuitive sense of what it is doing," Stanford associate professor of Earth system science, David Lobell says. "We can't say with certainty what associations it is making, or precisely why or how it is making them."
This is the hallmark of machine learning, as defined in 1959 by computer scientist Arthur Samuel, as a field of study that gives computers the ability to learn without being explicitly programmed.
"The beauty of machine learning in general is that it's very useful at finding that one thing in a million that works. Machines are quite good at that," Lobell explains.
The researchers said the model was incredibly accurate and would only get more accurate as it continues to receive more data.
But, satellite coverage of impoverished areas tends to be spotty, and the researchers say more satellite imagery is needed.
The researchers are also considering what they could do with data collected on mobile phone activity and other prosperity indicators.
Next weekend, the team will present their research at the Association for the Advancement of Artificial Intelligence's 30th Conference on Artificial Intelligence.
By Hannah Albarazi - Follow her on Twitter: @hannahalbarazi.
for more features.