Artificial intelligence plays an increasing role in the way security services respond to disasters and humanitarian crises like floods and disease outbreaks. However, this technology relies on human expertise to analyze and develop strategies based on the data those systems produce.
That was the assessment of a group of experts in both artificial intelligence and disaster management assembled recently by the Stimson Center think tank.
“In a lot of cases, AI models present a very definitive answer, but there still is inherent uncertainty,” Aaron Opdyke, a disaster risk specialist with the World Bank, said during a panel discussion. “We still need experts to understand and interpret that answer.”
As AI becomes more complex and more commonplace, it brings with it the capability to predict disasters in a way that was not possible before. Combining weather data with topographic data from satellites and local knowledge, for example, can provide insights into the potential for a given area to flood. Similarly, a combination of geologic, topographic and construction data can forecast the potential location and scale of damage caused by a future earthquake.
However, the experts noted, AI’s results are only as accurate and complete as the data it takes in — an issue known as data fidelity. Data that is incomplete or of poor quality can reduce the reliability of predictions, especially in countries or regions where governments lack the kind of data that AI needs to reach its conclusions, Opdyke noted.
For that reason, trust has become a key hurdle to using AI more broadly, according to panelist Susan Wolfinbarger, program manager at Geospatial Consulting Group International.
“From a governmental perspective, there’s a lot of hesitation to use AI in situations because of the need for trust and reliability,” said Wolfinbarger, who previously worked for the U.S. State Department’s Bureau of Conflict and Stabilization Operations.
Government agents often prefer person-to-person discussions to guide policy rather than trusting AI, Wolfinbarger said.
“Moving from ‘Let me call my friend’ to trusting something that is a big black box is a tall order,” she added.
As governments gather data to feed into AI, it’s crucial that they act in a way that safeguards the data while being transparent about what they’re collecting and why, according to Rachael Lau, a disaster risk expert at the Stimson Center. She said AI can be an incredibly powerful tool if deployed thoughtfully and in cooperation with local experts. But it also has the potential to cause harm if deployed improperly.
“On the governance side, there’s a lot of questions we have about data sharing,” Lau said. “If there aren’t guardrails that exist, it can be misused or used for purposes for which it was not acquired.”
Wolfinbarger offered the example of gathering information from witnesses to war crimes in Sudan. Without robust safeguards, that data could be used to harm the people who chose to speak out.
“There are so many routes that things can go wrong, you can end up targeting people you’re trying to protect,” she said. “Some of these technologies can be used for monitoring, for targeting displaced people fleeing a conflict, and figure out ways to harm them. It’s just such a huge problem.”
Opdyke reinforced that idea.
“We need to be thinking about what happens to the data that we’re feeding into AI systems — what are the potential misapplications of these systems?” he said. “We really need to be thinking through the negative consequences of using these systems.”
In some cases, it may be more effective to avoid using AI entirely, he added.
“That’s a difficult question for us to think about,” he said. “But that’s an important question we need to be asking ourselves.”