Transcript of Michael Oflaherty’s keynote speech delivered at the 15th Europol Data Protection Experts Network (EDEN) conference “Data Protection & Cybersecurity in Law Enforcement”.
Deputy Executive Director, Commissioner Gafà, dear Colleagues and Friends,
As you heard, it is my second EDEN. I am very happy to come back.
I seize every opportunity to accept a police-related invitation, because it is my firm conviction that those who work for human rights, and those who serve us in the police, are or should be partners in the pursuit of societies that cherish and honour human dignity.
As we seek to deepen that partnership, we do so in the extraordinary times of artificial intelligence. There is no doubt that AI can transform policing for the better, to deliver safer and fairer communities.
We have seen this since the very beginning of computing. How each stage in the extraordinary story has brought about a positive impact.
- Basic computing and all the efficiencies it introduced.
- Single-purpose AI, which we were able to apply in so many settings to benefit.
- General-purpose AI in more recent years, again transforming the way we work across multiple contexts.
- And right now, the phase of generative artificial intelligence. A phase in which astonishingly we are moving from the use of AI for analysis, to the use of AI for creation.
And then, looking into the perhaps near future, we have the vast benefit that beckons with the application of quantum computing, at all its speed and capacity, with edge computing I think particularly well suited to police related functions. And then looking even further again, perhaps into the world of science fiction, the jury is still out. There is the extraordinary potential impact which can be directed for the good of AGI.
Now I have spoken a lot of good and of value, and I firmly and truly believe in that. But of course, just as we have seen an evolution in the value of AI in the context of policing and related security work, we have also seen the extent to which it brought risk.
There has been an evolution in our understanding of risk. Inevitably, in a technology that is driven by data, the very early understanding of risk had to do with privacy. Still the primary context for the work of so many of you. But beyond the risks associated with the human rights of privacy, we very quickly saw the extent to which AI can make mistakes. It's highly prone to error.
That propensity to error has, in the recent past, converted into something even more sinister, the phenomenon of “hallucinations”. And with error and with bias and with hallucinations, we see how a range of human rights challenges of AI well beyond privacy to impact just about every aspect of our lives.
We also see how specific applications of AI threw up diverse human rights concerns and considerations. How surveillance, for example, raised issues of civil liberties. How robotics has put into question even the safety of human life itself.
Another set of risks which the audience will not discuss at this conference, but I think needs to be recalled always in any assessment of risk, is the impact of AI for our planet in terms of its need for energy and for water.
And the risks of AI are exacerbated by four features.
The first is that it is now well recognised that the primary driver of AI technology and its development is not quality. It is speed. It is efficiency. There is nothing inherently wrong with that. But when efficiency and speed come before quality, that then raises obvious issues of risk, including as regards human rights.
Secondly, and still to a very large extent, AI technology and its development is driven by the private sector. A mention was made earlier of the development of tools directly by Europol and by police services, and I accept that. But at some point in the chain of the development of technology, the private sector will always play a large role. And again, nothing problematic in that per se, but when we recall that the primary driver of the private sector is profit rather than quality and human rights compliance, this must cause us to pause.
Third of my four considerations to keep in mind as we assess risk, has to do with something I have seen myself over the years. That is the mutual illiteracy of the various communities that need to speak to each other. The tech types, the engineers, do not understand the human rights types like me. People like me, to be honest, do not always understand the tech types. We do not seem to have a common vocabulary to have serious conversations.
Fourth and finally - and it is obvious right now with regard to an assessment of risk - is the increasingly clearly perceived differences in understanding of rights and values on opposite sides of the Atlantic.
Given all of this risk, I think it is an amazing achievement that we have been able to come up with so much wise regulation in Europe. We should treasure the GDPR. We should treasure the EU AI Act. We should treasure the recently finalised Council of Europe Framework Convention on Artificial Intelligence. These are good, global, good-practise instruments. And we achieved them despite all the odds.
Let me say clearly today the European regulatory approach is fit for purpose, and the talk of dismantling it, of massively simplifying it or even of not applying it, must be resisted firmly.
Now what we have to do is deliver on the regulations, deliver on the oversight. And in that regard, I propose four duties and one, what I am terming, a necessity, that I would like to very quickly run by you this morning.
The first duty obviously is that we have to stay committed to the regulatory oversight and the regulatory environment. Yes, we can look that aspects need to be improved and strengthened. Yes, we can look at the issue of speed. We can look at so many of these things and find technical fixes for them. But as I said, we must hold true to strong oversight in the interest of human dignity and human well-being.
It does occur to me that we would not dream of a society where we had no rules of the road applicable on our streets. And yet there are voices out there right now somehow suggesting to us that we should have an AI with no digital rules of the road, a truly frightening prospect.
Second, side by side with standing true to regulation, we have to effectively implement it. And this does require us to pay attention to a number of issues. The first of these is that we have to take very seriously the regulatory sandboxes and the lifecycle testing of technology. Each of them plays a critical role and is vital to ensuring and delivering safety. And we have to make sure that we never lose sight of human oversight of high-risk technology. There is a temptation to trust the machine. There can be dangerous anthropomorphism. Machines can be tuned to deliver astonishing results, but they must never be allowed to operate in an entirely autonomous manner, at least when it comes to the high-risk applications.
The third of my four duties has to do with the need to empower strong oversight bodies for our regulations. We have crafted instruments which envisage national capacities to deliver the frontline of oversight. Now we have to make sure that they are given the resources on the one hand and the skill sets on the other in order to do their job. The one dimension of the skill sets that concerns me, given the nature of my role, is that of overseeing the breadth of human rights risk that comes with the application of the technology. As I said, it is about discrimination. It is about privacy. But it is about so much more than that as well.
The fourth and final of the duties, is one that is primordial, that is necessary to deliver on every aspect of our ambition, and that is we must demand, we must insist on the transparency of technology, the transparency of algorithms. We must get beyond the black box thinking, take the lid off the box, and be sufficiently able to see within the box in order for the regulatory function to be carried out. It's important to resist the smoke and mirrors the confusion and the magic talk of parts of industry that suggest that somehow that is beyond us.
By the way, keep in mind that delivering transparency is not just about effective oversight, it is also about trust. When our communities can see how the operation, the technology works, then there is so much more likely to be a high degree of trust invested in it by our communities.
Finally, my friends, I said I would mention one necessity.
I thought it was a bit much to call it a duty, and it is simply the necessity of partnerships. It is absolutely essential that we continue to invest in partnerships in order to get the delivery of AI for human well-being to get it right. This meeting, EDEN, is an example of exactly such partnerships. So, therefore, it is necessary, it is essential that the tech world sits down with the security experts, that they in turn spend their time with the human rights specialists, that all of us listen to the ethicists.
There are three other partnerships that I believe we need to do a better job of. These follow on points I have made earlier.
The first of the three is the need to invest in a strong public-private sector partnership. Too often, the relationship of public sector-private sector is that of client and producer or of adversary to adversary. We need a much more constructive, positive, joined-up effort to identify problems and find solutions.
The second partnership that I believe is under-developed is that of the technology world and research. It is clear in this rapidly emerging technology that we have to stay very heavily invested in the academic and related research in order to identify what the problems are, never mind how to fix them, and it is in this context that I applaud the work of the Europol Innovation Lab. It is an extremely valuable resource, and such resources are needed right across the continent. They need to be replicated at the national level.
The last partnership I will mention is, again, one of which we can talk and where we sometimes only glibly speak of its importance, but it is underdeveloped, and that is the partnership of the technology world and civil society. Civil society plays an essential role in warning us, in alerting us to where we should go, where we must not go, the perils along our pathway, and indeed, very often, how we can avoid those perils. By not including civil society sufficiently in the room where the decisions get made and the talk happens, then we do so at our loss.
Let me close my own remarks by assuring you of my partnership. I will continue to accompany the police forces and the security services of our continent; I will continue to accompany the astonishing revolution of AI, with a full commitment to helping us, together, to get it right for human well-being.
Thank you.