State government’s first priority is simple: to help people and to do so by creating an environment in which its constituents can flourish. For over 200 years, state and local governments have been serving people and have greatly increased their services. As these services have increased, so has the demand of constituent needs. Artificial intelligence (AI) can equip states to surge ahead faster than ever before as services continue to increase and critical needs arise. By leveraging prediction and augmented decision-making, governments can address the needs of taxpayers proactively while quickly adapting to evolving circumstances. Over the course of this blog series, we will address five critical ways that AI can help your state solve its hardest problems—beginning with the most immediate need facing states and localities today: crisis response.
The COVID-19 pandemic gave a stark reminder to the world of the inevitability of a public health crisis. Unfortunately, the pandemic also highlighted just how ill-prepared much of the world was to respond. Likewise, the U.S. was left flat-footed at the state and local level. COVID-19 was not the first crisis state governments have had to face and it will not be the last. State governments must enhance their crisis response for the future immediately—and AI can help.
When tackling a crisis, policymakers are often confronted with a scarcity of reliable information—as is the case regarding community spread and COVID-19 variant surges in respective communities. Many available forecasts provide less than four weeks notice at the state and county level. Additionally, these forecasts often miss surges in community transmission until it is too late to change course. As a result, policymakers are left without enough time to create and implement targeted, localized strategies in time to prevent outbreaks.
DataRobot helped combat this problem head on by applying AI to evaluate and predict resource allocation and identify the most impacted communities from a national to county level. On average, DataRobot forecasts had a 21 percent lower rate of error than all other published competing models over a six to eight week period. Additionally, our AI predicted disease impact up to six months in advance by leveraging more than 450 variables including mobility and unreported infections at the local county level.
For example, the COVID-19 vaccine trials required long-term forecasting to identify which hospital locations across the nation were most suitable to enroll participants. At the time, much of the available data was limited and inaccurate as a result of forecasting based on traditional—and outdated— epidemiological modeling. DataRobot applied an innovative, hybrid modeling approach, which combines mechanist epidemiological modeling with the enhanced predictive power of AI to identify COVID-19 hotspots months in advance by optimizing parameters daily to match rapidly changing dynamics of viral spread.
When collecting data at the county level, it is critical that vast differences in the size of counties across the nation are taken into account—whether the county has a few hundred people or a few million. DataRobot models successfully identified areas with high per-capita infection rates across the board, not only the densely populated urban counties. This approach allows end-users to make better informed decisions when assessing risk and other criteria such as age or ethnic diversity.
Traditional predictive models do not account for anomaly detection on data reporting issues (e.g., reporting backlogs). When these backlogs are not caught and corrected, these reporting issues can cause modeling errors resulting in untrustworthy forecasts. This was prevalent in the reporting anomaly in Houston, Texas in 2020, which led open source models (and the general public) to believe a large spike in COVID-19 cases were on the horizon when there was simply a backlog of reporting that went undetected until it was too late. The DataRobot platform and team provided long-term forecasting, localized intelligence, and continuous learning that addressed backlogs of data, allowed policymakers to rapidly adapt public health policy, and integrated variants of interest and concern across multiple geographic scales to create a best-in-class AI-driven solution for infections, cases, and deaths.
Augmented decision intelligence can help control outbreaks by informing public policy decisions such as travel and testing policies and optimizing the allocation of resources such as funding, vaccines, tests, ventilators, plasma, and personnel. To optimally provide for those most at-risk, real-world timelines must be aligned with real-world needs. AI will be critical for future crisis response and meeting public health needs.
About the author