Pandemics, analytics and politics: how data-driven AI makes better policy
Dr Nuria Oliver’s algorithms model the perfect politician, one who can make accurate and cost-effective policy responses amidst the chaos of a COVID-19 outbreak. Could this start a new global trend: evidence-based policymaking?
With the public weary of lockdowns, hospitals overloaded, vaccine shortages and policy-makers drowning in the chaos of a COVID pandemic, Spanish data scientist Dr Nuria Oliver knew that mathematics and big data could make better decisions than a human politician.
Her team got to work building big data models to forecast the incidence of COVID-19 and cost-effective policy interventions to minimise death, economic havoc and spread of the dreaded disease.
The model took 12 different ‘dimensions’ - or interventions such as closing workplaces and schools, mask-wearing and education - so the AI modelling could spit out the very best policy intervention given the hospital loads, economic costs and testing capabilities.
“When you look at the 12 dimensions, there were different activation levels which led to 7.8 million possible combinations,” explains Dr Oliver, who has published hundreds of research papers and developed more than 40 patents for data-driven solutions.
“No politician could likely comprehend those 7.8 million different combinations but a machine can - a machine can find the policy that satisfies the outcome against the given cost so it’s always offering up the optimal policy. It’s evidence-based policy making.”
AI helping humans
Dr Nuria Oliver - who is speaking for IAPA on May 6 - and her team, ValenciaIA4COVID, have been declared the world’s winners of the $US500,000 X Prize, which saw them create models for 236 countries to help navigate COVID-19.
“The models can make decisions that are not based on political interests or instinct but on the evidence of what the data is telling you,” Dr Oliver says.
“The system proposes policies with the best cost-benefit tradeoff so if your hospitals are empty, then you might want to open up a bit or if the hospitals are full, you would probably need to implement stricter interventions to contain transmission.”
Dr Oliver’s complex models take into account different measures but the final challenge to win the prize was to make an “artificial politician” - also called the prescriptor - to recommend an accurate response across 236 different countries or regions.
“Australia was very easy - you predict zero cases and you get it right,” she quips. “This is the advantage Australia and New Zealand have of being an island. But you will also have to vaccinate so you don’t get left behind.”
The maths is complex, the outcomes were not
The coronavirus pandemic created plenty of firsts - it was not only the first time the world adapted ZOOM meetings faster than you could say ‘have you washed your hands’ but also the first global pandemic that captured and shared daily COVID-19 case data across the globe.
Dr Oliver was working with the President of Spain’s Valencia region in December 2020 when the team started work on the X prize competition, and believes she couldn’t have done the work without the daily collaboration she was having at the highest levels of government.
Building an accurate prediction for COVID-19 cases came using official COVID-19 case data and the Oxford COVID-19 Government Response Tracker data set as the main data sources.
The team built neural network-based computational epidemiological models to predict COVID-19 cases 30, 60 and 180 days into the future.
The models used the confirmed number of COVID-19 cases and the implemented interventions to contain the pandemic in each of the 236 countries or regions in the world since March 2020.
The team also had to create a ‘prescriptor’ of 12 possible interventions, which had three to five different ‘levels’ each. These interventions included:
- Schools closing
- Workplaces closing
- Cancelling public events
- Restricting public gatherings
- Closing public transport
- Requiring people to say home
- Restricting internal movement
- International travel controls
- Public information campaigns like education
- Testing policies
- Contact tracing
- Facial coverings, or mask-wearing
Like all good data science, the quality of data varied so some countries had to be excluded due to lack of reliable data. The details of the work are in the process of being published in scientific journals but is not yet available, though there is more detail on this website.
Dr Oliver says workplaces closing so people can work from home were the most effective policies driving the number of COVID-19 cases up or down across all 236 countries and regions, closely followed by education interventions, international travel controls and restrictions on gatherings.
Better modelling came as data augmented with survey data
The team were able to refine their models as Spain embarked on its third wave of COVID-19, raising the death toll in Spain to around 77,000 deaths in April 2021 - just as they finished the predictor model for X Prize.
“We were able to accurately predict the evolution of the third wave of infections and test our predictions against real data,” Dr Oliver says.
Dr Oliver has also been running one of the world’s largest COVID response surveys, gaining valuable qualitative insights into the pandemic and offering citizens a voice.
This survey - which has had 600,000 responses - revealed valuable information about the perception and impact of the pandemic on people’s lives, especially on young people as well as the limitations of contact tracing.
“According to our survey, the contact tracing apps are a complete failure in Spain, Italy and Germany,” she says.
But how accurate was the AI-built perfect politician?
Given the hypothetical nature of the ‘prescriptor’ - or the so-called AI politician - the team could not evaluate its performance against ground truth.
The predictor did, however, deliver policy recommendations in less than two hours, which shows that politicians and governments could be using these types of models to help navigate complex daily decision making.
“It’s my hope that it will also help build the case for more evidence-based policy decision making,” Dr Oliver says.
“Public administration and governments of all levels are not undergoing the digital transformation that most companies have had to do in the past 15 years and the need to put in place a way to systematically gather data to inform policies in the future.”
Her greatest hope is that governments start standardising data collection and systematic analysis for the social good.
“We can easily create a virtuous cycle between data, people and technology so public policy is informed by insights,” she says.
Dr Nuria Oliver will deliver a virtual speech on May 6
Join us for this special webinar from Dr Oliver where she will outline the work her team has been doing in four main areas:
- human mobility modeling;
- computational epidemiological models (both metapopulation and individual models);
- predictive models;
- citizen surveys, with the launch of the covid19impactsurvey, one of the largest citizen surveys about COVID-19 to date with over 300.000 answers