Algorithms: Public Sector Decision-making

Lord Holmes of Richmond Excerpts

Wednesday 12th February 2020 Lords Chamber

[Lord Holmes of Richmond (Con)]

My Lords, I am glad of the opportunity to take part in this debate. I declare my interests as set out in the register and congratulate my friend, the noble Lord, Lord Clement-Jones, on securing the debate. The only difficulty in speaking at this stage is that we are rightly and rapidly running out of superlatives for him. I shall merely describe him as the lugubrious, fully committed, credible and convivial noble Lord, Lord Clement-Jones.

AI has such potential and it is absolutely right that it is held to a higher standard. In this country—somewhat oddly, I believe—we currently allow thousands of human driver-related deaths on our roads. It is right that any autonomous vehicle is held to a kill rate of zero. But what does this mean in the public sector, in areas such health, welfare and defence? As the noble Lord, Lord Clement-Jones, set out, over a third of our local authorities are already deploying AI. This is not something for the future. It is absolutely for the now. None of us can afford to be bystanders, no matter how innocent. Everybody has a stake, and everybody needs to have a say.

I believe the technology has such potential for the good, not least for the public good—but it is a potential, not an inevitability. This is why I was delighted to see the report by the Committee on Standards in Public Life published only two days ago, to which the noble Lord, Lord Stunell, referred. I support everything set out in that report, not least its reference to the three critical Nolan principles. I restrict my comments to what the report said about bias and discrimination. Echoing the words of the noble Lord, Lord Stunell, I agree that there is an important role for the Equality and Human Rights Commission, alongside the Alan Turing Institute and the CDEI, in getting to grips with how public bodies need to approach algorithmic intelligence.

When it comes to fairness, what do we mean—republican, democratic, libertarian or otherwise, equality of opportunity, equality of outcomes? On the technical conception of fairness there are at least 21 different definitions which computer scientists have come up with, as well as mathematical concepts within this world. What about individual, group or utility fairness and their trade-offs? If we end up with a merely utilitarian conclusion, that will be so desperately disappointing and so dangerous. I wish I could channel my inner noble Baroness, Lady O’Neill of Bengarve, who speaks far more eloquently on this than me.

The concepts and definitions are slippery but the consequences, as we have heard, are absolutely critical—in health, in education, in recruitment, in criminal profiling. We know how to make a success of this. It will come down to the recommendations of the committee’s report. It will come down to the recommendations—and not least the five principles—set out by the Artificial Intelligence Select Committee. Yes, mea culpa, I was a member of that committee, so excellently chaired, I say again, by the noble Lord, Lord Clement-Jones.

We need to consider the approach taken by the EHRC to reasonable adjustments for public bodies and the public sector equality duty; this is really about “CAGE”—"clear, applicable guidance: essential”. The prize is extraordinary. I shall give your Lordships just one example: in health, not even diagnostics but DNA is currently costing the NHS £1 billion. A simple algorithmic solution would mean £1 billion saved and therefore £1 billion that could go into care.

I am neither a bishop nor a boffin but I believe this: if we can harness all the positivity and all the potential of algorithms, of all the elements of the fourth industrial revolution, not only will we be able to make an incredible impact on the public good but I truly believe that we will be able to unite sceptics and evangelists behind ethical AI.