Nearly half of councils in England, Wales and Scotland have used or are using computer algorithms to help make decisions about benefit claims, who gets social housing and other issues, despite concerns about their reliability.
A Guardian freedom of information investigation has established that 100 out of 229 councils have used or are using automated decision-making programmes, many without consulting at all with the public on their use.
This is despite one council admitting that results from one algorithm showed it was only 26% accurate in some instances. The company behind it said it was because people often entered information wrongly.
Another council dropped an artificial intelligence tool to process new benefit claims, saying they were not satisfied with its reliability.
A range of private companies are selling machine-learning packages to local authorities that are under pressure to save money. The systems are being deployed to provide automated guidance on benefit claims, help decide who gets social housing, and allocate school places among a range of other uses.
Concerns have been raised about the arbitrary nature of these programmes, which inform important decisions about people’s lives, and their scope for making mistakes.
Martha Dark, the director of the digital rights group Foxglove, said: “It is very worrying to see so many councils putting in place algorithmic systems without any public consultation. These systems are clearly being used to make huge decisions about people’s lives.”
The use of artificial intelligence, or automated decision-making, has come into sharp focus after an algorithm used by the exam regulator, Ofqual, downgraded almost 40% of the A-level grades assessed by teachers. It culminated in a government U-turn and the system being scrapped.
One of the most used algorithms among councils is that for risk-based verification, a process in which claims for housing and council tax are automatically processed to determine the likelihood of fraud. Those that are considered higher risk are slowed down and people are asked to provide more evidence.
The programme is still being used by South Ayrshire council. When asked about the accuracy of this programme, the council said it had used the company Northgate Public Services, and the programme was found to be 26% accurate at finding low-risk claims, 36% accurate for medium-risk claims and had a 40% accuracy rate for high-risk ones. The service cost £41,500.
A spokesperson for South Ayrshire council said: “All applications are manually processed by staff – we do not use robotics to verify any application or evidence submitted. The type of risk associated determines the type or level of evidence required to verify the claim and so improves efficiency in claims processing.”
Northgate Public Services said the accuracy rates quoted were not related to the effectiveness but rather claimants often inputting incorrect data.
Wigan council uses an algorithm to allocate social housing. It says answers are submitted and a points-based system then decides which category to place people into. The council uses Northgate housing software to sort candidates into groups. The company stressed it was not involved in banding and assessment criteria, which is instead put in place by councils.
Wigan said applicants and council tenants are placed in either group A, B or C when they are able to demonstrate that they meet the appropriate criteria in either their permanent or temporary home. Group A includes young people leaving local authority care, those in severe hardship and those leaving the UK armed forces. Group B includes those living in poor housing conditions. Group C includes those living outside the borough with no local connection, among other criteria.
Nicolas Kayser-Bril, from Algorithm Watch, a non-profit group looking at algorithmic decision-making processes, said that while the council should be commended for its transparency, “there seems to be ample room for arbitrariness”.
“There will never be a way to objectively define who should be first in line for affordable housing. Transparent algorithms can help, but so can professional caseworkers who take the time to discuss with the applicants. Revealingly, none of the criteria in the list concerns the wishes of the applicants. Instead, they attempt to define the ‘good pauper’, much like in previous centuries.”
Joanne Willmott, Wigan council’s assistant director for provider management and market development, said: “Applications are assessed and prioritised based on answers submitted by customers around their own individual circumstances on our online housing application portal and then assessed in line with the allocation policy. A final check of the application is manually carried out.”
Flintshire council, which says it used the company Civica at a cost of £961.40 a year to process new claims for the period 2014 to 2018, said the service was terminated, as the council was not satisfied with its reliability.
A spokesperson for Civica said: “While the authority did not continue with this approach which was based on a third-party algorithm, the company continues to provide the core revenues and benefits software for Flintshire county council which is a longstanding customer.”
A Local Government Association spokesperson said: “Councils have been trying to improve how they use data in recent years, and predictive analytics is just one example of this. Good use of data can be hugely beneficial in helping councils make services more targeted and effective … But it is important to note that data is only ever used to inform decisions and not to make decisions for councils.”
A Hong Kong pro-democracy activist and former politician said that he will move to Britain in exile.