Just under half (49%) of UK data leaders have said their organisation’s data is disproportionately weighted to specific ethnic demographics, increasing the risk that their artificial intelligence systems contain ethnic bias.
According to research by Big Data LDN, one of the UK’s largest data events, all of the 250 respondents surveyed conceded that their organisation’s datasets contained some form of bias.
Artificial intelligence (AI) systems are only as good as the data they are trained on. When data on a certain demographic is limited – whether it’s race, age or gender – then the accuracy of the output for that group is reduced.
This has resulted in a number of high-profile cases of algorithmic bias. In 2018, e-commerce giant Amazon scrapped an AI recruiting tool that showed bias against women. Because the algorithm was trained on Amazon’s 10-year hiring data – during which Amazon had predominantly hired men – the AI system taught itself that male candidates were preferable.
In 2019, a widely used healthcare algorithm was found to routinely choose healthier white patients over black patients with poorer health for access to extra care.
Big Data LDN’s Industrial Revolution 2020 Report surveyed 250 UK data leaders working in large enterprises across five sectors.
Broken down by sector, retail organisations had the worst ethnicity data bias at, with 56% of respondents conceding there was a problem.
Telecommunications and high-tech companies ranked highest for age data bias at 36%.
Some 82% of UK businesses said their data is disproportionately weighted to specific location demographics.
Meanwhile, insurance and financial services organisations had the most disproportionate weighting of data for specific gender demographics at 70%. Telecommunication and high-tech firms came second, at 64%.
Brexit and Covid threaten UK data innovation
Big Data LDN’s findings come at a precarious time for the UK’s data industry, with the pandemic, Brexit and a shifting regulatory landscape all creating uncertainty.
Some 66% of respondents said they needed more clarity on data sharing regulations post-Brexit.
“Data professionals are watching intently as data becomes the battleground for competing global nation states,” said Big Data LDN founder Bill Hammond.
“UK businesses are losing trust in governments, favouring major cloud players in their stead. Meanwhile regulation adapts to protect the privacy of citizens, raising questions about who will steer the UK’s Fourth Industrial Revolution.”
While companies have ramped up spending on technologies to facilitate remote working, from collaboration tools such as Zoom to data centres, spending on other digital transformation projects has taken a hit due to the pandemic’s economic fallout.
Despite the growing importance of data in business, the average spend on data-driven initiatives has decreased to £1.8m – a 22% year-over-year decrease.
During the pandemic, 76% of surveyed public sector organisations said they have scaled-down in-house data teams.
“As the UK’s data leaders map out the future in a data-driven world, skills play a vital role. Covid-19 has presented challenges to the way the UK data economy works,” added Hammond.
“With smaller teams having to do more with less, the question of the UK’s leadership in the Fourth Industrial Revolution is, for the first time, unclear.”