Artificial Intelligence is getting smarter, the robots in charge of hiring, a legally-binding treaty to remove plastics from the environment finds its legs.

This edition brings you links to the most interesting fun panels on AI, in Davos. Also the full-interview with Steve MacFeely, on why we need a UN data governance agency.

Scroll all the way down.

Become a paid subscriber.

Data Governance Needs its Own Agency

Interview with Steve MacFeely, Director of Data and Analytics at the World Health Organisation and Chair of the Chief Statisticians Group of the United Nations System

We spoke about the upcoming UN Statistics Commission meeting that begins on the 23rd of February, in New York. The heads of all international organizations and national statistics offices will discuss a wide range of issues, and this year health is at the top of the agenda, with a full-day dedicated to it.

We also talked about the right fora to discuss data governance, trade in services, and the necessary institutional framework to address AI risks and benefits.

Then there is the Summit of the Future in September where the UN Statistics Commission will have a role as data governance is central to the current conversation.

Interview has been lightly edited for length and clarity.

UN Data Governance in the Spotlight

Maya Plentz - Tell us what is in store for 2024?

Steve MacFeely - In February we begin the big international calendar with the UN Statistics Commission. That begins on the 23rd of February in New York.

There you'll have all of the international organizations, but you'll also have the heads of all the national statistical offices discussing a wide range of issues, but including health this year.

Health is on the agenda, and perhaps also of great interest to WHO, and to me personally, on the 23rd there'll be a full day dedicated to international data governance which is something I've been working on as well as the Chair of the Chief Statisticians Group at the UN.

This is an area that's really attracting a lot of attention, in fact I'll be in Davos speaking on the same issue so you can see around the world there's growing interest around data governance because that's really impacting AI, because data are the fuel that's really driving AI and there's growing concerns about the governance.

I mean we've been using AI in different forms for years because basically any model you could classify as artificial intelligence So one of the big issues, like I mentioned, is data. The quality of the data feeding into it is critically important. Not only your own data for what you're doing, but the data that was used to build the model in the first place. In other words, the training data.

One of the concerns we always have is if the training data have introduced biases into the model, then those biases are hard-coded in thereafter. Whether it's gender, whether it's ethnicity, whether it's some sort of bias regarding developmental status, there could be many.

It's not that this is new.

Almost every dataset has biases, but you need to be aware of what those biases are so that you can try and correct for them, or at least be aware of them when you're doing the analyses.

One of the big challenges with large language models is that the data are so vast, it's very hard to understand what data is actually been training the model. And that makes it more challenging to understand what the biases are. And then that, as a statistician, makes it more difficult to correct for or mitigate against those biases.

Maya Plentz - And also the question of queries, right? Making the right queries to this data. If you have this really vast body which you don't know where the data came from, how was it trained, from where this data was gathered, how can you possibly do a good query too. Good queries would come from understanding what body of knowledge these models were trained on, right?

Steve MacFeely - Well, that appears to be one of the challenges too, is that depending on the query that you ask, you could ask what appear to be similar queries, or at least to you as a human appear to be reasonably similar queries, and yet you get quite different answers.

And that's one of the things that we have to understand more is how the questions that we're asking of AI are actually being interpreted by the model.

And why are they giving sometimes quite similar answers, but other times what appear to be quite different answers?

And that's a worry, you know, like if I don't ask exactly the right question, then I get this crazy answer.

I mean, it's a bit like it's almost going back to Hitchhiker's Guide to the Galaxy.

You know, now we know the answer to life, the universe and everything.

The problem is we don't know the question.

It's almost a challenge that we're facing now.

We're getting answers, but maybe we haven't asked the right question.

Maya Plentz - So what else is going to happen in Davos next week? Because the world converges there and data governance, as you said, is becoming very, very important within international organizations. It always has been, but now the conversation has transpired around, from International organizations to the private sector, data governance is the talk of the town.

Steve MacFeely - It sure is. So last year, in the middle of last year, the Chief Executives Board of the UN published a paper called International Data Governance: Pathways to Progress, which set out a view from the UN Secretariat on the importance of data governance.

And then also kind of what steps could be taken to try and improve the situation.

But that's a position paper.

Now we need countries to decide if this is something that they want to pursue.

It's not an easy topic because, across the world, there's at least three different views or perspectives around data governance.

Here in Europe, you have very much the kind of human rights centered approach, protecting the rights of the individual. In Asia, we see much more a nationalistic view, which is more about the state perspective and the individuals are sacrificed towards that.

And then in the US, we really see the private sector kind of laissez-faire approach.

And it's not clear how to reconcile these views.

But if we're talking about trade in services, if we're talking about… almost anything we do now involves the sharing of information.

So how do we share information around the world across those different perspectives that still allows us to protect against misinformation, disinformation, and also protects the individual rights of people to make sure that their rights aren't being abused as we do that.

So data governance is huge because the entire future of global commerce, human rights, national security, they all hang on the balance around data governance.

Maya Plentz - Yes, that's a very good point you make there. What kind of system there is within the United Nations to discuss this? You say that country members need to get together and now balance these three possibilities of looking at data governance. Which one is going to prevail or should they prevail? Or they should find a perfect equilibrium, which is sort of the topic in some ways. But what would be the forum for this to happen, actually? Is there a forum already? Would we need to create another forum?

Steve MacFeely - Well, you've hit the nail on the head, Maya. This is the big question.

Where do we discuss this issue? I t certainly already has surfaced in the discussions around the Global Digital Compact and in fact if you look at the policy brief for that you'll see in section F there's a large section dedicated to data governance but we would argue as data people that actually data are so important it needs to be discussed in and of itself —- it's not just a subset of digital it's not a subset of something else —- it's actually a policy issue in and of itself so one of the questions is, are data now sufficiently important that they should be part of the Summit of the Future that will be coming this autumn just after the General Assembly?

Does it need an institution of its own?

Does it belong somewhere else?

So like this is one of the other questions around data governance is where does it belong?

Who's going to take responsibility for it?

Is it a subset of a different conversation or is it a conversation on its own?

And that's an open question.

So you have the issue itself and then who's going to take ownership of the question?

Maya Plentz - Yes, that is a big question, but it's a question that perhaps needs its own institutional framework, to speak just about data governance itself, because otherwise it can easily get hijacked by different interests.

Steve MacFeely - Yeah, possibly. I mean, or simply, I've been drawing analogies wit