Public sector IT projects need ethical data practices from start

Public sector organisations should take an iterative approach to data ethics that encompasses every stage of a project, from initial data collection all the way through to live implementation and beyond, so that the lessons learned can be truly incorporated into future work, says data ethicist Haroon Ahmed.

Ahmed, commercial partner for data capability at public sector technology delivery firm Made Tech, says that a major focus for Made Tech is integrating data ethics into every project it works on, which it refers to as “ethics by design”.

He tells Computer Weekly that, far too often, data ethics is simply ignored by public sector organisations because it is perceived as either too contentious, complicated or time consuming. He also noted that with governments generally, but the UK government in particular, there is no clear, standardised direction or approach to data ethics.

“Maybe you can’t [adopted standardised approaches] because there’s so much variation in the types of work different departments do...[but] usually data ethical problems turn up when it’s a bit too late, when things have already gone wrong,” he says.

To overcome this and to catch any issues early on, Ahmed says organisations need to think about the ethics of data inputs and sources from the very start of the process.

“The starting point is looking at the whole data journey. So it’s looking at, what are your data sources? Who has the rights on those data sources? Are there any limitations on data sources?” he says, adding there also needs to be clarity around why data is being collected and processed in the first place.

From here, he says organisations should start measuring the effect of the data on different demographics, so they can see where it is producing both positive and negative impacts.

For example, organisations should look at what kinds of bias are “inadvertently being created” in their models, as well as through the ways they collect, share and use the data.

“Are there practices that you’re developing that reinforce stereotypes or inequalities that may exist within the data – well, look at the data sources, who’s collecting them, how are you collecting them data?” Ahmed says.

“Look at data that may propagate some falsehoods. A lot of the time we’re structuring data around opinions – for example, if you’re mining data from social media, what sort of information are you gathering? How much can you trust it, what is news and what isn’t? Most of the time, bias [is introduced] in the way you’ve collected the data, so look at your research methods.”

Ahmed says that once negative impacts are identified, organisations should immediately move to think about mitigation: “It’s just asking yourself like clear and transparent questions around how you’re setting up the project…there’s also a level of testing and rigour you should put in when you’re building. Have private betas, develop them, test them before going out and launching something live that is then going to give you a bad name and crush everything.”