SILICON VALLEY — Hope and anxiety coexist over how artificial intelligence will reshape society. Optimists expect AI to drive the development of humanoid robots, easing human labor and improving living standards. Pessimists warn it will create structural problems, including military threats, exposure to cyberattacks, mass job losses and a widening wealth gap.
One benefit often cited is AI's potential to prevent natural disasters caused by the climate crisis and to contain infectious diseases. Climate change, wildfires, floods and epidemics are global challenges, yet the damage has been concentrated in less developed and developing countries with weaker finances and technology. AI can sharply reduce the cost of building climate observation and forecasting systems, allowing those countries to redirect resources from disaster and disease response to economic development.
Google Research, the company's research arm, held its Research@ event at its Mountain View, California, headquarters on the 14th, presenting 10 research results from this year that it said expanded human creativity. Among them was Groundsource, released in March.
Groundsource is an AI system that forecasts urban flooding. Using Google's Gemini model, it analyzes public reports, local news and Google Maps to build a dataset based on past flood records, and issues flood alerts up to 24 hours in advance. According to the World Meteorological Organization (WMO), about 85% of flood-related deaths worldwide are caused by flash floods. Flash floods typically occur within six hours of heavy rainfall, turning city streets into torrents and killing more than 5,000 people each year.
Extreme weather is making flash floods increasingly severe. Early warning systems (EWS) are essential to limiting the damage, but they remain out of reach for poor countries with limited budgets. In developing countries, fewer than half have access to multi-hazard early warning systems, and billions of people receive no advance warning at all, even when lives are at stake.
To address the problem, Google began large-scale machine-learning experiments in flood forecasting in 2018, starting with a pilot project in Patna, India. It built a dataset from 20 years of public reports and more than 2.6 million flood events, then trained its AI on the data.
The result was Groundsource, which can forecast urban flash floods up to 24 hours ahead. Through Flood Hub, Google provides flash-flood information to 2 billion people in more than 150 countries. At an online briefing in March, Google cited a case in which the Southern African Development Community checked flood alerts on the beta version of Flood Hub and then contacted South African authorities to request monitoring of the events, saying the example showed how AI can help less developed countries prevent natural disasters.
Earth AI, Google's geospatial analysis platform, is also helping less developed countries prevent and treat infectious diseases. Google said the World Health Organization's Africa office and the National Institute of Biomedical Research (INRB) in the Democratic Republic of Congo are using Earth AI to curb the spread of outbreaks. When Ebola broke out in Congo this year, the WHO Africa office used Google AI to analyze satellite imagery and maps, identifying 48 settlements at risk of exposure and a population of 45,500 people, enabling an early response. The INRB used Earth AI to estimate the risk of the virus spreading to uninfected areas. Congo, which declared an Ebola outbreak on May 15, had recorded a cumulative 9,212 confirmed cases as of the 13th.
Last month, Google published a report titled "How Google AI is reshaping global crisis resilience." Weathernews, the world's largest private weather news company, integrated Google's models into its own platform to support forecasting in Vietnam, the Philippines and Indonesia. After identifying flood-prone areas in Vietnam, it issued a warning three days before Typhoon Matmo made landfall. Google said joint pilot projects with the WMO and the national weather agencies of the Czech Republic, Nigeria, Uruguay and Vietnam improved the accuracy of forecasts for local river conditions.
Seventy years ago, advances in computing allowed the United States and Europe to build modern weather forecasting. Countries in Africa, Southeast Asia and South America, meanwhile, were so consumed by floods, droughts and epidemics that development and investment were out of the question. Analysts say that while pouring national resources into disasters and disease has long held back growth in poorer countries, AI could resolve those constraints and change the structure of their economies. In a report in August, Indermit Gill, a World Bank researcher, said AI would give less developed countries more accurate medical imaging and better weather forecasts, and that the shock of job displacement would be limited because their economies are centered on agriculture and other primary industries — meaning they could gain more from the spread of AI than advanced economies.
Still, some argue that simply building AI models will not solve the problem. What matters, they say, is whether the countries that need the technology have the authority to use it on their own terms. If a single country or company controls a model — as the United States restricted access to Anthropic's most advanced AI models amid its technology rivalry with China — poorer countries could face even greater difficulty addressing disasters and public health crises. Even models developed for the public good cannot be assumed to be free of dependence on a particular state or firm.
The Institute for Climate and Sustainable Growth at the University of Chicago struck an optimistic note in a report last month, "Weather Forecasts for Health — The case for a decision-first approach." "AI weather models can deliver results equal to or better than the best physics-based forecasts at far lower cost, using only laptop-level computing power," the report said. "The arrival of AI has removed constraints on the cost and time of running models and on generating forecast outputs." But the institute also called for work on demand articulation. "This requires clearly defining what decisions are to be made, who has the authority to make them, and what authority and resources they hold," it said. "On that basis, systems for forecasting, evaluation and delivery can be built." Raising forecast accuracy alone is not enough, it added — decision-makers must be able to operate the models in practice.
