AI: removing jobs or enabling change?

Meet your strategic enabler

The area I’m particularly keen to explore concerns its strategic application. This will increasingly become the case as AI saves us more and more time by handling mundane administrative tasks and freeing people up to do more value-added tasks.

One of them is related to contract management. Some public sector contracts run to about 700 pages, and if you could ask AI to pull out all the obligations, that could save you loads of time. And again, at the Request for Quotation stage, you could ask it to condense the tech key notes.

Another application might be if you are going into a negotiation. You tell it who the supplier is, what the problem is and then ask the AI how it might approach it. You can even ask what its top tips might be. Taking this even further, you could invite it to be the supplier and share your opening gambit, asking it what the response might be, using it as a sort of online coach.

Coming to terms with the limitations

There are ways that AI certainly isn’t a panacea, and there are more risks than this approach might suggest. Certainly, the fear of it replacing our jobs might be misplaced.

For instance, when we run tenders in the public sector, we need to be very careful about paying attention to the procurement regulations. That means we need to make sure we’re tendering on an open basis in most instances, making sure our evaluations are objective and we’re not talking to suppliers outside controlled means.

AI in evaluation has potential utility, but there is still much to test before it can be deployed. A fair and transparent evaluation is a key tenet of public sector procurement, with suppliers able to challenge contract award decisions based on a non-compliant evaluation. AI usage would therefore need to be fully transparent and objective, which is a challenge with current models.

The risks of early adoption

While there would be admin tasks within that which AI could be helpful with, it also means that if there is something we get wrong in the process, we could be in even more trouble. It’s important to note that it isn’t AI that’s the risk here, so much as how we use it. The same applies to information security.

"It’s important to note that it isn’t AI that’s the risk here, so much as how we use it"

It would be possible for an employee to go on the internet using the public version of ChatGPT or Copilot and upload a government document. At that point, the owners of the AI platform would have the document and could use it to train their model or, worse yet, it could be leaked.

Our solution to this has been to get an enterprise licence for Copilot. This means that data cannot go out of the department and that it cannot be used to train the publicly accessible Copilot model. This broadly mitigates the wider data security risk and in addition, we’ve put a lot of training, guidance and guardrails in place. The department has worked hard to ensure those internal structures exist.

This feels as though it’s comparable to the advent of the internet. It changed the way people work and now it’s second nature. We know not to go on certain websites and not to click on certain links. As it was with the internet, it will soon be with AI.

The other risk that we can’t avoid is with sustainability. At a departmental level, we have an IT sustainability team who are now quite busy. Part of the training they deliver is to use it only where it adds value. This means that you shouldn’t ask it to show you certain regulations because you can search for that online.

While we want to make some of the infrastructure, like the water or the processors more sustainable, the cat's out of the bag when it comes to AI.  We're never going to get to a place where people decide to stop using it or restrict it because of sustainability concerns. I think it'll be more about working with it and making usage itself more sustainable, than replacing it.

Finding a future-proof platform

The final risk, and perhaps the main one, is about the platforms we use. As I said, we are now using Microsoft Copilot, which has its own implications in terms of potential lock-in. It is important that we remain open to what is a very fast-moving market over the coming years.

There are some huge risks that come with being overly reliant on any particular AI tool and AI in general. What if we get used to using it and there’s a decision that we have become over-dependent on it? Would that mean we can't do our jobs, or that we can't do them to the same degree? Could we just plug in another tool and use that instead, or would it have implications we might not like?

Some organisations have rolled out AI on various platforms and then changed their minds because they've run into issues of the kind I’ve mentioned here. There are no signs of that happening at the DHSC, but all these things are concerns that will have to be navigated if AI is going to realise its potential.