BT AI chief criticises UK’s centralised approach to COVID-19 contact tracing app
The UK's approach to delivering a contract tracing app that is aimed at reducing the COVID-19 infection rate has been criticised by BT's head of AI and data science research.
Detlef Nauck said today at the CogX Leadership Summit that the UK would have been better off following the path of other countries that have adopted a decentralised approach, because of data privacy concerns.
As lockdowns begin to ease, governments around the world are looking to contact tracing as one mechanism by which they can help track the spread of the COVID-19 virus. This is being done manually through the recruitment of contact tracers that call, email or text people that develop symptoms to find out who they have come into close proximity to - but also through smartphone applications that use Bluetooth technology.
The apps typically adopt one of two approaches. The decentralised approach alerts users via Bluetooth to when they have come into contact with someone experiencing COVID-19 symptoms, but all the data remains on each citizen's individual phones.
The UK's approach - a centralised one - will see the NHS collect all the data centrally. It has been argued that this approach has its benefits for mapping outbreaks and getting better control of the virus.
And whilst NHSX - the NHS's digital unit that is responsible for the development of the app - has said that it will be publishing the key security and privacy designs alongside the source code for peer review, privacy advocates have heavily criticised the approach.
Commenting on the centralised approach, BT's AI chief said:
A better version is the decentralised version where all the data stays on your phone all the time and you would only notify a server that you have symptoms - and then everybody who also uses the app can download your notification and see if they have been in contact with you. They wouldn't know who you are and nobody would be able to see all the contacts that had happened.
It has been said that for the app to be effective in reducing the spread of COVID-19, it would have to be downloaded by a significant proportion of the population. Nauck added that the centralised approach could reduce peoples' desire to download it. He said:
It will certainly put off some people that are worrying about this. And so it would have been better to go with a decentralised app because then at least you can say there are no privacy concerns, as the data is on your device. That takes away a whole discussion around this problem area - there are already organisations that want to sue to understand what the data privacy repercussions are.
The privacy assessment that has been done has been lacking and came out very late. So you just have to be on the money with these things to build the trust and convince people that the data will be safe and used responsibly. You could do things like Australia where they have put laws in place to make sure that the data gets deleted after the crisis is over and can only be used for this one purpose.
Mission creep is a big problem in these kinds of areas. Once the data is available centrally, what else could be done with it?
Nauck went on to say that a decentralised approach also doesn't have to mean fewer opportunities for smart analysis of the data. He explained that complementary analytics technology could be used on the smartphone to aid a more sophisticated approach to understanding an individual's risk exposure. He added:
A lot can be done if you commit to keeping the data on the phone, because you can use smart applications on the phone to manage the data and to mix it with other information about yourself that only you know and you don't want anyone else to know - to curate the data and make better decisions.
That plays into this responsible use of AI, where you want to make sure that the technology isn't biased, is fair, it's transparent and people understand how it's going to be used. And that it guarantees privacy and security and protects the data of individuals.
Principles for contact tracing app development
In addition to the data privacy concerns, Nauck also said that these apps are biased towards the more affluent parts of the population - because they rely on people having access to a smartphone. This was a point echoed by Claire Steves, senior clinical lecturer at King's College London (KCL), which has been carrying out extensive research on COVID-19 symptom tracking across the UK population via an app.
Nearly 4 million people across the UK have downloaded the KCL app and are regularly reporting their symptoms and how they're feeling to the study. KCL has also made it possible to report the symptoms of other people in the app, in order to broaden access.
Steves said that the contact tracing apps, which should be designed to try and help everyone in the population, should be guided by three core principles. She explained:
It's really important that users are fully involved in any AI process that's going on with their data, rather than it being a black box that's done by either researchers or scientists or Google, or whoever. The users really need to be involved in that process and have ownership of it. It's about that issue of trust and making sure that their data is treated appropriately and with their consent.
The second thing is around an issue we find in healthcare - there are so many records systems out there. There is quite a lot of difficulty still in linking these records. Really if AI is going to reach its maximum potential, we need to have the governance systems in place to fully link them.
The third is making sure that any technology that we develop is equitable across society. We've already talked about apps not being available to everybody. But it also generally tends to favour those that are able to take advantage of it. But of course healthcare problems are more prevalent in individuals who are more disadvantaged. And so if we really want to make a difference and protect everybody, we have to make sure that we make the technology work for the most deprived populations and the most vulnerable.
Gloss