← Justice Matters
Justice Matters
An International AI BIll of Human Rights
00:34:54
On today's episode of Justice Matters, co-host Mathias Risse speaks with Yuval Shany, fellow at the Ethics in AI Institute at the University of Oxford. They discuss his recent white paper, “The Need for and Feasibility of an International AI Bill of Human Rights,” and the topics it touches on around AI’s profound impact on the understanding and implementation of rights. Other topics they discuss include: the impact of AI on society, opportunities and challenges the technology poses for human rights, why the need for a new International AI Bill of Human Rights, what the new bill would entail, the political liability of an international bill, the future of AI regulation, and the importance of integrating human rights principles into AI development and deployment. Yuval Shany is the Hersch Lauterpacht Chair in International Law and former Dean of the Law Faculty of the Hebrew University of Jerusalem. He was a member of the UN Human Rights Committee from 2013 to 2020, and served for...
Loading summary
Transcript28 lines
- [00:00]
A
Foreign. Welcome to Justice Matters. I am Matthias Russe, one of the co hosts of Justice Matters. Artificial intelligence is one of the biggest topics of our time. At enormous pace, AI is rapidly reshaping how we live, how we work, and how we are governed. AI offers opportunities to improve education, health, training, transport, and many other things. But even if we ignore dangers from a runaway AI that might take over all human systems that are designed in language, and even if we ignore the role that AI might play in great power competition, AI also puts human rights such as equal treatment and privacy under considerable pressure through practices that include both large scale surveillance and and individual profiling. The black box nature of many AI systems raises further concerns about transparency, accountability, and the ability of individuals to seek effective redress. I am delighted to be here today with Yuba Shani, who is the Hersch Lauterpracht Chair in International Law, the former Dean of the Law Faculty of the Hebrew University of Jerusalem, and has held numerous visiting positions in academia and has also served on a number of international human rights commissions. And while being an Accelerator Fellow at the Oxford Ethics and AI Institute, Professor Charney recently published a new white paper under the title of the need for and Feasibility of an International Bill of Rights, which is the subject of our conversation in this podcast. Welcome Professor Shani welcome you all to our podcast.
- [01:53]
B
Hi. Good day, Matthias. Thank you for having me.
- [01:57]
A
Yuval, can you can you tell us a bit about your background? What role do human rights play in your work and what kind of academic work have you done around them? And what kind of public policy work have you done in the human rights domain? Just to start the conversation?
- [02:11]
B
Yeah, well, like you said, I'm an international law professor at Hebrew University, where I've been teaching and researching for many years, including in areas related both to human rights law and also the laws of war, international humanitarian law. I've been involved early on with a number of Israeli NGOs and also worked closely with some international NGOs and with the International Committee of the red Cross. In 2013, I was appointed I was elected to serve on the Human Rights Committee, which is one of the principal expert bodies that is operating under the UN umbrella to monitor the manner in which states are complying with a very important treaty, the International Covenant on Civil and Political Rights, and I have served for eight years as a member of that body, including one year during which I chaired that body. In that context, I had the important task of trying to obtain a global picture of how human rights in a great variety of contexts are being implemented after stepping down from the, after two terms on the committee, I did reorient much of my research to deal with this interplay between human rights and new technology, which already, while sitting on the committee is something that I sensed is an area which is both extremely significant going forward, but also lacking in legal regulation, so to speak. And I've been in the last few years, I've been both working at Oxford, like you mentioned, with their Ethics in AI Institute, and also running a European Research Council project based in Jerusalem on the three generations of digital human rights.
- [04:05]
A
Okay, yeah, so it's like a lot of people in academia. So the importance of AI has impressed itself on us. And this is kind of a similar trajectory to mine actually. I wrote a paper about eight years ago on human rights and artificial intelligence and urgently needed agenda. More coming from a moral philosophy perspective. But I guess its relevance is simply everywhere right now. Right?
- [04:30]
B
Yeah, of course.
- [04:31]
A
So now let's talk about the various aspects of that. So AI obviously poses any number of opportunities and challenges for human beings and for the realization of human rights. Let's start with the positive side. Do you think there are ways in which artificial intelligence can actually help with specifically the realization of human rights?
- [04:52]
B
Yes, I mean, definitely to the extent that AI is capable of promoting human well being across a variety of contexts and sectors, such as the health industry, the educational industry, the work safety, for instance, context in these regards, you could say not only that AI contributes to human well being and potentially to the enjoyment of human rights, such as the right to health, the right to safe labor conditions, the right to education. But at some point in time, we could actually maintain that it is indispensable for full implementation of such rights. So if states are required under international law, the Covenant on Economic, Social and Cultural Rights, for instance, to afford individuals the highest attainable level of health. And AI implies at this point in time, for instance, a superior level of diagnosis or even of surgery in some contexts, then actually there may be at some point in time, and maybe we are approaching that moment in time in which there would be actually a positive duty on states to render AI services accessible to individuals. But on top of the systems themselves being capable, capable of supporting human rights, I would say that they also can support the work of groups and individuals who are monitoring compliance with human rights in the sense that they could help us track violations in the same way, because they can provide us, for instance, early warnings about impending violations or identify violation trends, which is also very useful and very important in actually combating human rights violations. And they're also increasingly being recruited for the actual mundane work of human rights bodies, such as human rights courts that are often handling very large litigation volumes. And using AI is another measure to increase efficiency and to generate more outcomes, in this case, more decisions protective of human rights. I would definitely do not belong to the camp of those who see AI as a catastrophic, as a catastrophe for human rights. But of course, if not managed properly and carefully, it could also entail some very negative consequences.
- [07:23]
A
Yeah, and we, I mean, some of what you spoke about there, we have certainly seen quite a bit in recent years also in civil society, human rights organizations taking up human AI tools quite a bit for some of the purposes that you mentioned.
- [07:38]
B
Exactly, exactly.
- [07:40]
A
But let's turn to the challenges. So what do you think are the major challenges that come from AI? And maybe let me ask this question in two versions that may or may not amount to the same thing. So the first version is what are the major challenges that come from AI, period? And the second version is what are the major challenges that AI poses that can actually be addressed through a human rights framework?
- [08:06]
B
Yeah, well, it's a disruptive technology. It is disruptive in the sense that it could. When we are delegating to AI systems functions that were previously performed by human beings, we have to ensure that the safeguards that we have developed over the years, for instance in the area of safety or legality or responsibility, would migrate, together with the Delegating act, to the use of AI systems. So if, in the example that we gave before of a, of a hospital diagnosis, if we are now delegating the responsibility to perform the diagnosis to an AI system, but we are uncertain about the safety of that system, we are uncertain about whether the system complies with the existing laws that are in place, and we are uncertain as to what would happen if something goes wrong. Who is going to incur liability for this. Then we have in a way transitioned from a system where we might, in terms of performance, it may be an inferior system, but we had many considerable significant protections, legal protection and ethical protections. And we have to ensure that they migrate with the new technology going forward. Now, add to that that the technology itself is not only disruptive in terms of the replacement of a human with an algorithm, the algorithm itself comes with some significant baggage. One area is the famous black box baggage, where the episteme that is surrounding the use of these systems is a different from the episteme that is using, that is being used by human beings. And in many cases it is not widely accessible to individuals who are interacting with these systems. So we are creating a new form of disruption and epistemic disruption on which we have to accept that there would be significant human rights implications. Because if we do not know how the machine operates and we do not know what considerations and what data was actually used in order to generate a certain outcome, our ability to challenge the outcome and to challenge the operation of the machine and to evaluate it on the basis of human rights standards has been significantly curtailed. So the epistemic problem is a problem that really could block the gateway to effective human rights protection across the board. And. And finally, there is also this idea of what does it do to us as human beings, as a human society, when we are replacing an interaction between two human beings with a machine human interaction. And that too may have some human rights implications. So I think to the second part of your question that some of these disruptions can be already perhaps dealt with within existing human rights norms. For instance, insistence on safety. Safety and maybe insistence on legality and responsibility with regard to some other. The more I would say epistemic and ethical dimensions, such as the replacement of a human to human interaction with a human machine interaction, here I think we need perhaps to push the envelope a little bit and discuss whether we would need some new human rights frameworks in order to deal with it.
- [11:42]
A
Which actually brings us directly to what you are proposing in your white paper. So you are proposing an international AI Bill of Human rights. So maybe can you just. Let's take a step back here. What kind of thing is an international AI Bill of Human rights? What is the legal nature of that? Who would have to pass it? How would it fit in with the legal landscape that we already have in the international law domain? And do we really need a new document of whatever sort it would then be?
- [12:14]
B
Yeah. Well, first, I do encourage our listeners, if they are interested in this, to visit the website of the accelerator program at the Ethics in AI Institute at Oxford. And the whole document is there. And they can.
- [12:29]
A
And it's easy to find. It's really easy to find.
- [12:31]
B
It's easy to find. Thank you. And they can evaluate for themselves whether this is justified or not. I think that the thinking on my end goes in this manner. I mean, we do have some human rights treaties, but those human rights treaties are mostly based on the Universal Declaration of human rights from 1948, which of course was drafted in a very different era, both in terms of the technology that is present. So AI was at that time a very. Maybe just a pipe dream. Just the initial in the 1950s, we have started actually seeing publications in that regard, but operative systems were very, very far in the distance. So the treaties that we currently have are built for an analog world in which humans are still pretty much running the show. So here is one area that we already have, a gap that might need to be filled. There is also, within the existing human rights framework, I would also say there are also two problems when we are trying to adapt it to an AI based world. One, it's very much state centered. The existing human rights framework is really about the relationships of the government and the governed. It's about state obligations. In this AI world, the power is often placed in the hands of private actors, the large tech companies that are developing and launching AI products. And in addition, the existing human rights framework that we have is very much victim oriented. It is designed so as to allow individuals who have suffered specific human rights violations to claim their grievance in a human rights committee like the one I served on, or a regional court, sometimes domestic court. Whereas here, a lot of the harms that we are concerned about, I would say are societal harms, like a disruption of our informational ecosystem, which are very hard to enforce through individualized perspective. They need actually a different framework of analysis in order to effectively protect human well being. So we have a system of human rights which we should certainly build upon upon. But to put it bluntly, it is not fit for purpose for the kind of changes that we are confronting. Hence, the tradition within human rights over the years has been that once such gaps had been identified and a need to provide more detailed guidelines emerges, then new instruments are concluded. So sometimes they begin as declarations, as the Universal Declaration of Human Rights itself started, or the very important business and human rights declaration that was adopted at the UN in 2011, which identified already this transition of power from states to businesses. And on the basis of these declarations, sometimes you see also treaties coming later into play, such as the Convention on the Rights of Persons with Disability, which is a relatively recent convention. When the drafters of the original treaties in the 1940s and 50s and 60s were negotiating the rights of persons with disabilities were not at the forefront of their sensibilities. But in the early 21st century, of course, the world had changed. And my thinking is that in this area we should try to proceed in a similar manner. Develop a non binding statement, what is sometimes called soft law, that articulates the rights that we aspire to develop, and offer some general guidance for those who wish to follow these standards going forward, but hopefully later. On if consensus emerges, we can then try to render these norms more concrete and enforceable through their internalization in domestic law. And I should say, already many states are regulating AI. So. So the idea that AI meets regulation and that the problems that we have discussed need a legislative response. This is already in the table across many jurisdictions, including in the United States itself at the state level. So far. What we are missing, however, is placing these different regulatory instruments within a broader human rights framework that would connect them to the values that underpin the human rights movement. Dignity, equality, liberty, solidarity, etc.
- [17:22]
A
Let's get one level more specific. So your proposed bill, the. You know, the. The AI Bill of Human Rights includes a list of seven fairly broadly defined rights. Can you walk us through two or three of them so that our listeners get a sense of what this would amount to more specifically?
- [17:41]
B
Right. So sometimes some of the rights simply involve the transportation of a right that exists already, such as non discrimin to the realm of AI by adopting, by proposing the adoption of certain adjustments that would bridge the gap between the norms as they were originally conceived and the specific needs that arise in relation to AI systems. So, for instance, one of the rights in the bill was algorithmic bias and fairness. And here the concern is how do we actually ensure that AI systems are compliant with non discrimination provisions? And we already know that we have certain areas that we need to look closely at, for instance, training data. We know that to the extent that AI systems are based on non representative training data, this could be a problem because they will then generate outcomes that are also non representative. So we need to vet the data that is being used in training and also in post training by algorithmic systems for lack of bias. A related issue which could also be described as a problem of algorithmic fairness is the tendency of some users of AI systems to perpetuate existing inequalities. So if for instance, you are an employer and you are using an AI system to hire new employees and one of your system requirements is to try to identify employees who will succeed succeed in the firm on the basis of the characteristics of the existing workers in the firm that have succeeded, that have been promoted over the years, you may end up perpetuating certain demographies who have been in the past very successful in climbing the corporate ladder, while effectively discriminating against newcomers who do not have demographic newcomers who do not have this kind of track record. So that's effectively perpetuating the status quo. And this is something that could generate concerns about equality. Finally, I will just mention the Issue of proxy data. That's another specific application of non discrimination to AI systems. We have a good track record already in identifying what are some sometimes refer to as suspect classifications. We know that an employer or high education institution should not vet individuals on the basis of their skin color or gender or religion, et cetera. But what we are seeing with AI systems is that it's very easy for AI systems to circumvent these criteria, for instance, by looking at what appear to be neutral criteria such as zip codes or sometimes social media activity, in order to effectively generate the same discriminatory outcomes. So here, the adjustment of the existing framework to the new reality would require some tinkering, but not a revolution, I would say, not a revolutionary approach. The second type of right that I will illustrate is the right not to be subject to automated decision making, or the right to a human decision, which is an area with relation to which we currently do not have a clear framework within human rights law, simply because when human rights law was created, it was not seriously contemplated that important decisions such as whether to hire or fire someone, or whether a bank should be giving a loan or mortgage to a person, would be determined by an algorithm. And here the proposed right, which is also already mirrored by some regulatory instruments that exist in some parts of the world, would posit that when important decisions are taken with respect to an individual, decisions that affect one's legal rights or important interest, that individual may have a legal claim, a right, that is to insist that the decision would either be taken by a human being or at least would be reviewed by a human being as a sort of appeal. And the argument here that this is a way to protect rights, but also to preserve human dignity, because the sensation according to which you are being examined by an algorithmic machine that is reviewing you as essentially an assortment of data points, that is a sort of treatment that could be experienced by many people as fundamentally dehumanizing. And here, in order to preserve this sense of humanity and also expectations of empathy and solidarity, a right could be asserted. Though I should note that some people would actually prefer to be interviewed by an algorithm. But this is the nature of rights. It is your decision, it is your choice whether you want to be reviewed by an algorithm or whether you want to opt out. And I'm actually making in the white paper the case that a similar set of considerations could also apply not only to decisions by algorithms, but also to interactions with algorithms. And again, in certain specific contexts, such as having a right to be diagnosed by a human physician at a hospital or Having a right to be interviewed by a government, a human government official when asking for social benefits, etc.
- [23:48]
A
So that's this point about the right to a human decision here. I found, I find this quite intriguing. So my experience is discussing this with audiences is there's kind of a bipolar reaction to this where some people feel extremely strongly about this whole idea about keeping a human always in the loop and you know, having a possibility of a human appeal here. And other people say, you know, the whole point of this is to squeeze out human bias. Right. So we want to improve above human human bias. So that's, that's why we're interested. And is this, is this your experience also? It seems like the way you are passing this out reflects this kind of experience also.
- [24:22]
B
Yeah, definitely. I mean, again, it's always interesting to see the supermarket lines for the checkout that when you have an automated option you would see some, you could say maybe natural selection. I mean, some clients would be more comfortable working with the automatic machines and some clients would be more comfortable with the human checkout personnel. But when one is dealing with important interests, and at least at this point in time, where we do have a black box problem, where we are not clear that our rights are being adequately protected in the process, and where we may not have a clear line of liability to actually uphold if we believe that we were wronged, I think that here there should be a very strong case, at least in the interim, to allow this opt out option, namely the right to a human decision. We may end up in the future. And I think I've also written on this elsewhere, that part of the anxiety about non human automated decision making maybe has to do with our sociology being a generation that still had encountered human beings in critical juncture and maybe 20 years down the line, this wouldn't seem as something which is so anxiety generating as it does now when framed as something that entails a choice. And I should say all the rights that I present in my proposed bill are relative rights. Human rights generally tend to be relative. So unless the costs, the economic cost or the social cost or the efficiency costs are prohibitively high, I would maintain that we should try as far as possible to accommodate these sensibilities and to implement such rights.
- [26:25]
A
Let me touch on one last topic, namely the political situation in which this is all unfolding, since there have been also some dramatic changes there. So my own sense of the last couple of years, there was so between the release of, you know, the first really strong version of ChatGPT, basically in March 2023. Between that time and the reelection of Donald Trump in November 2024, there was a. There was a real window, a global window of opportunity for a shared approach of sorts, of quite possibly regulation. But this was a massive wake up call. There was a flood of summits and a lot of documents, white papers, institutional engagement. And people really just realized politicians and administrators who always tend to be quite behind when it comes to actually seeing where technology is going. Everybody was in their faces during that time and there was a very serious willingness also coming out of China and definitely out of the United States. But now there's been a complete sea change because of changes in the United States. So the United States has put out their own new AI approach. And it's, you know, basically drawing on a speech that J.D. vance gave, you know, last year. It's basically replacing everything in the domain of AI safety with what they call AI opportunity. It is basically seeing the whole world as one competitive field where the only interesting competitor to the United States is China. Nobody else really terribly matters. So it seems like the political scenery has changed away from any kind of global approach or in a way away from the kind of thing that you are proposing just at a time when it is most urgently needed. So how do you judge the political parameters here? And just kind of asking this bluntly, how politically realistic is this that your proposal will go anywhere in the near future?
- [28:25]
B
Well, first, like you said, I mean, it is urgently needed. And I think the acknowledgment of this need is significantly growing. And you're right to point at the introduction and the very fast dissemination of generative AI as an inflection point. So it is, in a way, you could say, at some level, unfortunate that the current administration, US Administration has walked back the approach of the previous administration on that issue, because the previous administration did issue a soft document called the AI Bill of Rights, which was supposed to produce this sort of soft guidance to different regulatory bodies in the United States. Now, having said that, I do see states moving in the direction of regulation. This is true. I mean, we have seen only in 2025, we have seen new laws in Japan and in Korea and in Italy and in Taiwan. In 2024, the EU had adopted an AI Act. And we are also seeing inside the United States, we are seeing that different states are passing AI act mostly in the form of consumer protection instruments, but they all seem to be. This is already happening and it is happening everywhere. Like you said, China also has some regulations that protect AI. So I think that there is still A strong interest in many places around the world, including inside the United States, to introduce certain guardrails. And I think one of the key challenges for people who believe that we need regulation which will be adopted at somewhat at a more universal level, or at least invoke universal principle, is to in a way deal with the false dichotomy between innovation and regulation. So human rights are not, would not deprive AI systems or AI companies of opportunities. They would actually create new opportunities for these companies to demonstrate that the products that they are marketing are trustworthy because they are ultimately benefiting humans. They are promoting human well being and the way in which I am approaching the project by actually advocating for a soft law instrument which would be voluntarily adhered to or not adhered to by governments and companies and other relevant stakeholders. This is a way to deal with some of the concerns that you alluded to. So I don't think that this is something that is going to happen in a day. On the other hand, I do think that the companies themselves do have an interest in identifying standards that they are complying with. We have just seen a couple of weeks ago Anthropic publish a very comprehensive document on Claude, it's called the New Claude Constitution, that is making a great effort to show that their system is safe and that the system is ethical. And I think actually human rights would be the proper grounding of these interests in ways that would have currency around the world and would also enjoy the aura and political pull that human rights often have. But I do take your point and I realize that my project and similar project like mine are facing at this point in time, some headwind.
- [32:17]
A
Yeah, yeah. But, you know, the United States has midterms coming up later this year, and there's a distinctive possibility that, you know, that will lead to a change, domestic politics here that eventually will also have implications for AI, though it remains to be seen. So before wrapping up, may I invite you to speak maybe a few concluding words on the subject and its relevance and nature?
- [32:42]
B
Yeah. So I think this is a topic which is, I think, fascinating from an academic point of view, partly because it's such a complex terrain, partly because it does require us to deal with very complex issues in human rights law, but also because it forces us to think about the foundations of human rights law and the foundational values and to speak again about what does it mean to enjoy human dignity, what does it mean to be humanely treated, what does it mean to exercise, for instance, freedom of thought in an era where your thoughts could be in a way shaped and nudged and navigated with the use of technology. So I find this really as not only a voyage towards the future but actually also going back in time or read to the time in which human rights were created and introduced domestically and internationally. And that is I think part of the challenge of how do we reinvigorate human rights and how do we reassert our humanity and human rights in an age which is much more automated and mechanical.
- [34:02]
A
Thank you Yuval for joining us today on Justice Matters.
- [34:05]
B
Thank you.
- [34:13]
A
Thanks to our podcast team, Kyle Van Neuf, Mac Foley, Joer and Peter Kokoma. Justice Matters is a podcast production of the Carl Ryan center for Human Rights at the Harvard Kennedy School and the podcast co hosts are Amin Ta Awsum, Diego Garcia Blum, Maggie Gates and myself Matthias Risse. You can find more information about the podcast on our website, easy to find if you google the Carl Ryan center for Human Rights. Thank you for listening to Justice Matters.
- [34:39]
B
That.