The impact of the Covid-19 pandemic has proved that effective infectious disease surveillance systems are essential for global health. Last year, BlueDot and Metabiota separately demonstrated the disruptive potential of artificial intelligence (AI) in disease surveillance and pandemic prevention. We should not, however, lose sight of AI’s limitations either.
Founded in 2013 and 2008 respectively, BlueDot and Metabiota both utilise AI to monitor disease outbreaks around the world. Both companies use AI algorithms to analyse various data sources and public documents, such as airline ticketing data and statements from public health bodies, using natural language processing to analyse sources written in any language. The collected data enable the companies to build up a picture of where outbreaks are occurring, how severe they are, and where the diseases are most likely to spread next.
Both companies’ prediction records show that this system works. The World Health Organisation’s first public reference to Covid-19 came in the form of a tweet on 4 January 2020, in which it referred to ‘a cluster of pneumonia cases in Wuhan’. Both BlueDot and Metabiota had already reported the outbreak on 31 December 2019. Even more impressively, after its initial announcement, BlueDot predicted which ten cities Covid-19 would spread to next. Eight of these predictions were correct. In addition, BlueDot predicted in 2016 that the Zika epidemic would spread to Florida six months in advance, and in 2014 correctly forecast that Ebola would spread beyond West Africa.
Both companies have been involved in pandemic success stories. Software-as-a-service (SaaS) platform Bizzabo, a specialist in event success, made effective use of Metabiota’s analysis of the risks posed by Covid-19 when it presented to board members on 12 February 2020. With little more than a month’s head start, Bizzabo had developed an online virtual event solution by 20 March. Bizzabo’s revenues more than doubled last year, enabling it to acquire three companies this year and open an office in London.
BlueDot played a part in Taiwan’s successful initial repulsion of Covid-19. Dr Hao-Yuan Cheng of the Epidemic Intelligence Centre in Taipei claims that BlueDot’s automatic news collection, modelling and risk assessments were useful for implementing informed border controls and monitoring what was happening around the world during the pandemic. Until late last April, Taiwan had recorded only seven deaths despite having a population of 23.5 million.
The limitations of AI and big data
These results seem impressive, but expectations should be proportionate. Problems with the data on which these surveillance systems are based, as well as with AI’s use as a developing technology, could produce imperfect conclusions about the diseases under surveillance.
The first issue is that data sources such as statements from public health organisations are only as good as the organisation’s understanding of the disease in question. In the case of Covid-19, when faced with this new coronavirus, the scientific community’s understanding of how it spread changed over time. An immature understanding of a new disease could limit the usefulness of an AI-based surveillance system.
In addition, AI is still a developing technology and has mistranslated foreign languages in recent years. GlobalData’s 2021 report on AI states that, in 2017, a Facebook AI algorithm translated ‘good morning’ in Arabic as ‘attack them’ in Hebrew, resulting in the wrongful arrest of a Palestinian man. Companies looking to follow in the footsteps of BlueDot and Metabiota must be vigilant in this respect.
The road ahead for AI and disease surveillance
Pandemic control has become an understandable priority for governments. AI has proven itself a powerful tool for tracking outbreaks, but it is not yet a silver bullet.