– in the House of Lords at 2:57 pm on 5 December 2023.
To ask His Majesty’s Government how the Department for Work and Pensions is using artificial intelligence and what governance process is in place for such use.
My Lords, DWP has used forms of AI for some time and we continue to investigate new opportunities. This includes looking at how generative AI can help us deliver high-quality services to improve customer experience and colleague efficiency. We are aware of the transformative benefits of AI, as well as the potential risks. We have created the AI Lighthouse programme to explore opportunities, and we have a framework ensuring that we work safely, ethically and transparently.
My Lords, this Question has become topical since I tabled it, since the Government have started to take powers to look into the bank account of every pensioner in the country. But that has made me even keener to understand exactly how DWP is using AI. Can the Minister tell the House whether it is used to select people for health reassessments, or to decide who to investigate or who to sanction? If so, what safeguards are in place to ensure that it is used transparently and fairly? How do we avoid it becoming a sort of digital version of stop and search?
I hope I can reassure the noble Baroness that we already have a proven track record in delivering technology in a responsible and well-governed way. We have extended our governance to include an AI steering board and an assurance and advisory group. DWP always ensures that appropriate safeguards are in place for the proportionate, ethical and legal use of data, with internal monitoring protocols adhered to. I further reassure her that the Cabinet Office’s Central Digital and Data Office has recognised our Lighthouse programme’s safe acceleration framework as an exemplar for AI development in government.
I can reassure my noble friend that it will. I shall give a bit of granular detail: a 2021 DSIT report highlighted the potential impact of AI on the UK labour market, and this of course includes DWP. Automation is forecast to increase, rising from an estimated 7% to 30%, but I can reassure my noble friend that, with the changes, there will be a net gain. We have an average of about 900,000 employees per quarter moving from one job to another, so I can reassure my noble friend that my department’s employees will reduce, but there will be opportunities for those in AI.
My Lords, as the Minister said, the Government are rolling out massively complex new systems, with significant risk to claimants because they have not got their original systems in order. We hear constant reports of backlogs at the Future Pension Centre, payments for national insurance credits being lost within the system and more and more historic pension errors coming to light when it comes to things such as home responsibilities protection. Can the Minister update the House on the steps to get those existing systems in order and on what learning exercises will be carried out to ensure that no such errors will be carried forward on the new and potentially more powerful systems that he has outlined?
We are certainly working very hard to look at and mitigate delays, and AI will over time be a game-changer for that. To manage and mitigate risk, we have produced a risk framework, in line with the Department for Science, Innovation and Technology. We are setting out AI governance and an approach to AI enablement which will be transformational.
My Lords, I shall pick up on what my noble friend said about digital stop and search, because there is growing concern about the potential for hidden bias in the use of algorithms to detect social security fraud. What steps has the DWP taken to prevent such bias, with potentially discriminatory outcomes?
The noble Baroness raises an important point. We are committed to building trust in our use of AI and are fully aware of the risks of the technology, as discussed at the UK AI safety summit. Where AI is used to assist its activities in the prevention and detection of fraud within UC applications, DWP always ensures appropriate safeguards, and bias is something we are very alive to. It will very much depend on the input of data and we have some risk profiles in place to ensure that we adopt best practice in that respect.
My Lords, given the appalling amount of fraud within the DWP, costing billions per year, surely we should welcome the fact that DWP is using AI and algorithms to target this problem. The key is presumably that, once AI has reached a conclusion, actual human beings should review the situation. Can the Minister tell the House whether the DWP has robust internal quality assessment procedures?
There are couple of questions there. We continue to explore the potential of AI in combating fraud. This includes the integrated risk and intelligence service, using AI to assist in identifying possible fraud in processing universal credit advances. To answer my noble friend’s question, importantly, DWP does not use AI to replace human judgment when considering the potential for incorrectness to either determine or deny payment to a claimant. The NAO and the ICO looked at this issue recently and found no areas of immediate concern.
My Lords, the issue that my noble friend raised about access to millions of people’s bank accounts came up at a very late stage—Report—of Commons dealings with the Data Protection and Digital Information Bill. Can the Minister outline why such contentious measures were introduced only after the line-by-line consideration of the Bill in the elected House? Why did the Government refuse the Opposition’s request that the legislation go back to Committee, as did the Online Safety Bill in the last Session? Can the Minister justify why this very contentious piece of legislation is being rushed through?
I will not be drawn into answering questions on that, but I can say that it is important that the scrutiny of the Bill is done in an effective way and, of course, this House is very good at doing that. As I have mentioned before, it is very important that there is trust in AI solutions; this must be a prevalent issue among all users of AI.
Will this AI enable people who are on social security to get a better deal and get off social security, so that we can tap into the skills and abilities of millions of people who are caught in the Bastille of poverty and social security?
Yes, and I can outline that a lot of very good work has been done so far. As I said earlier, the work still has to include human judgment, but AI is being used to assist with improving on repetitive processes for staff. We are introducing conversational platforms for triaging, which will lead directly to having a human face. The whole point is to speed up the process and include more human judgment in ensuring that more people get into work, and faster.
Has the Minister seen the Paradot website? If not, will he look at it? Is the department examining whether the buddy concept developed there could be used in the department that he represents and in other government departments, and what the consequences of using that in government service would be?
I am not aware of that, but I will most certainly look at it.
I will need to write to the noble Lord with those specific figures.
The Minister has said repeatedly that he wants the public to have trust in the use of AI in the system. Can he therefore tell us what proportion of cases where AI has been used are subsequently checked by a human? Will he publish the results of that analysis to show whether the AI decisions are the same as human decisions, or perhaps better, or worse?
I can give the noble Lord some reassurance on the processes that we have in place. AI is an evolving, iterative process and it is important to highlight the fact that we have a test- and-learn approach. We must proceed with extreme caution in what we are doing. Test-and-learn means that we need to get to a point where we are assured that this will work and that nobody will be affected detrimentally. Then we can accelerate our programmes.