Centrum Etyki Technologii

David Wright: I am optimistic about AI standards integration

agagaj — Tue, 21 Feb 2023 18:44:38 +0000

David Wright:
I am optimistic about
AI standards integration

21 February, 2023
agagaj

I hope the AI Act reaches a workable consensus between the European Council, the Parliament and the Commission and that it is generally welcomed by stakeholders. I think that with its adoption there will be greater harmonization of AI risk assessments as well. The Commission did give greater guidance to data protection impact assessments with the GDPR and they may well do the same thing with the AI Act, says David Wright in conversation with Maciej Chojnowski.

Maciej Chojnowski: In a report “A survey of artificial intelligence risk assessment methodologies” co-produced by Trilateral Research and EY, you refer to risk assessment, impact assessment and risk management as similar but separate approaches. One could also add conformity assessment here. All these terms have something in common but do not cover exactly the same issues. Can you explain the relations between them?

David Wright*: Impact assessment as a general term is the broadest, of course. And there are different types of impact assessment: privacy, ethical, surveillance, security, and environmental impact assessments. They usually follow a structured approach. However, there are quite a few differences in those approaches even so.

Risk management is a kind of different term although it’s certainly associated with risk assessment. Because in order to manage risks you need to assess them and decide which risks you should prioritize, where you should focus your attention etc.

Conformity assessment on the other hand is something totally different. In the case of Europe, it has to do with meeting EU standards. In order to get a CE mark, you have to conform with certain regulatory requirements. The European Commission is quite advanced in its thinking in terms of developing a cybersecurity seal that would be used for certifying software products and services.

Let’s take a look at today’s AI risk assessment landscape. What has already been done, and what is still to come in terms of effective tools or governance procedures?

Working on the survey for EY, we looked at different types of artificial intelligence risk assessments (or approaches that were very close to what we might term an AI risk assessment). And there are lots of organisations, including lots of companies, who are thinking about these issues already. Artificial intelligence has benefits but it also has risks. And those risks have been well publicized in the media. So companies are certainly aware that they should take them into account or otherwise they might suffer something unpleasant to their reputation.

So in terms of context, it’s very favourable towards the introduction of artificial intelligence risk assessment. I think in 2023 we’re likely to see agreement between the European Council, the Parliament and the Commission in regard to the AI Act. And the adoption of the AI Act will be quite an important factor in the expanded use of AI risk assessments.

In the report, you name two general types of AI risk assessment: identification of specific risks and classification of an overall level of risk. Should the latter be the first step in evaluating AI systems (with the former being the consecutive step), or maybe you can’t determine the overall level of risk without identifying the specific risks first? Is there one right order here, or maybe it all depends on the situation?

I would say certainly the first step should be the identification of specific risks. No question about it. Having identified those specific risks, one could refer back to the AI Act and see whether it falls in one of the four categories of different types of risk, including prohibited activities and high-risk activities. Most likely it will.

So definitely the first step is identification of the risks before seeing how it might be categorized and what steps you need to take in regard to which category it falls into.

Your report contains an overview of different AI risk assessment standards. Do you think that once the AI Act becomes law there will be one right European standard developed (e.g. by CEN), or maybe we should expect a coexistence of a number of them (by ISO and IEEE), including the solutions developed in a research project like SIENNA?

That’s a very good question but answering it would require a little bit of crystal ball gazing, and I am not sure I am the best futurist. But certainly the ISO and in particular Special Committee 42 and Working Group 3 have been looking at these kinds of issues and identified principles that AI systems should comply with. So I think that will have quite a big impact. And CEN is following very closely what ISO is doing.

However while the AI Act is an actual regulation, the ISO standards are voluntary standards. But I am pretty sure as a minimum we are going to have the AI Act and some ISO standards, probably more than one.

In addition to those it’s likely that individual companies or industry associations might also adopt some codes of practice or principles that would be part of that same “cluster” as the AI Act and the ISO standards. So if you take those kinds of activities into account, most likely we will see various types of standards anyway.

Let me continue this thread for a while. I would like to refer to a recent report by the European Commission’s Joint Research Centre on AI standardisation landscape where it reads: “We see standardisation work at the European level as an instrument for integration and supplementation of valuable content from ISO/IEC and IEEE, preventing fragmentation when it comes to standards on AI risk management, hence facilitating the work of AI providers to achieve compliance.” Do you think that this sort of integration will be the result of the CEN work?

I definitely think so. All of those organisations talk to each other. Often it’s the same people going to CEN and ISO meetings and responding to the various consultations that the EC has had in regard to the AI Act. So I would be very optimistic that that will happen definitely.

But do you expect any original input from the work of CEN? Because one can think that if ISO and IEEE have already started working on the standards while CEN has not, then once CEN finally starts they will probably have lots of material to choose from. I am therefore wondering if they are going to add extra value to the existing knowledge, or will they just repeat what has already been said elsewhere.

I couldn’t answer that question. But I know that CEN follows ISO very closely and it has established a working group to deal with some of the ethical issues somewhat similar to the work of ISO SC42 Working Group 3. I am not sure what exactly the status of that working group is right now in CEN. They were discussing its formation the last time I checked it out.

In your pioneering work on impact assessment, you have covered such areas as privacy and surveillance. Considering your experience in the field, what are you greatest hopes and fears as far as the upcoming AI risk assessment systems and AI legislation are concerned? What are the most crucial issues lawmakers and standardization bodies should remember about?

As far as my hopes are concerned, I hope the AI Act reaches a workable consensus between the three institutions and that it is generally welcomed by stakeholders of the AI Act. Of course, some stakeholders will not like it. Non-Europeans may not like its extraterritoriality. But nevertheless, I hope it receives general acceptance.

I think that with the adoption of the AI Act there will be greater harmonization of AI risk assessments as well. The Commission did give greater guidance to data protection impact assessments with the GDPR and they may well do the same thing with the AI Act. However, I think there are some uncertainties with regard to the institutional arrangements at the Member State level, i.e., will Member states form special AI regulators, or will the data protection authorities take AI under their wings. So that’s not quite clear to me how that might develop with certain overlap between those areas.

And my fear would be that the introduction of a major new regulation will have a kind of deterrent effect on innovation and business. I am not sure whether that will be the case because one can also see a positive effect in terms of European companies and non-European companies offering services in Europe saying to their customers that they comply with the AI Act. And I think that will also help create a more level playing field as well.

So overall I am more positive that this year will be one of hope for AI enthusiasts and that the fears with regard to the deterrent effect of AI Act will be somewhat overblown.

*David Wright is Director of Trilateral Research, a company he founded in 2004. Trilateral has partnered in more than 75 EU-funded projects and coordinated several of those, including the ongoing CC-DRIVER project on the human and technical drivers of cybercrime. He has published more than 70 articles in peer-reviewed journals, and co-edited and co-authored several books, including Privacy Impact Assessment (Springer, 2012) and Surveillance in Europe (Routledge, 2015). He coined the term and published the first article on ethical impact assessment. The ISO standard on privacy impact assessment (ISO 29134) is based on his PIA methodology. Similarly, the European Standardisation Committee Workshop Agreement (CEN CWA) on ethical impact assessment is based on his EIA methodology.

Przeczytaj polską wersję rozmowy TUTAJ

Artykuł David Wright: I am optimistic about AI standards integration pochodzi z serwisu Centrum Etyki Technologii.

Henrik Skaug Sætra: Framing AI impacts with SDGs is beneficial

agagaj — Thu, 24 Nov 2022 17:00:24 +0000

Henrik Skaug Sætra:
Framing AI impacts
with SDGs
is beneficial

24 November, 2022
agagaj

We have to contextualize our analysis of AI and bring the human and the social back in. Techno-solutionism is all too often subject to a debilitating blindness to wide areas that are crucial to human existence. Many experiential aspects of human life are lost if we restrict ourselves to machine-like or AI-compatible approaches to science, says Dr Henrik Skaug Sætra in conversation with Maciej Chojnowski.

Maciej Chojnowski: You are an advocate of contextual approach to artificial intelligence and remain critical of so-called isolationism, which is a siloed approach to AI impact assessment. Your approach to assessing AI includes quite a number of different factors. Recently, you have focused your research on the United Nations’ Sustainable Development Goals (SDGs). Are they the ultimate tool for assessing AI?

Henrik Skaug Sætra*: Short answer is no. They are not ultimate, because they don’t cover privacy and a lot of human rights issues directly. So, there are obvious oversights. But they bring some really good benefits with them as well.

They have the legitimacy and spread both in the business world and politically, so it makes them useful for analyzing and not least communicating some of the major impacts of AI. And they are an effective tool because many people are already thinking about how to avoid negative impacts on the SDGs. So framing AI impacts with these goals is beneficial. They are also easy to use pedagogically. It makes sense to focus on these different issues and ask how AI relates to this.

So, although SDGs definitely are not the ultimate tool here, they are still useful and effective because many people already work with them. Particularly in the business world where people communicate everywhere how they deal with sustainability or ESG (Environmental, Social, and Corporate governance).

In the SDGs, there are three main dimensions (economic, social, and environmental), 17 top-level goals, and 169 targets (subgoals). You emphasize that these goals and targets should not be perceived as separate because they cover many interrelated issues. To deal effectively with such a complex set of issues with regard to AI, you need an advanced solution. Can you describe your SDGs-related AI assessment method?

My approach makes it easier to distinguish between different kinds of effects. There are two main ways in which I do that. These are distinguishing between direct and indirect effects and distinguishing between three levels of impact.

Let’s start with direct and indirect effects. If, for example, you use AI to improve teaching, then it could have a direct impact on SDG 4 (education). That could be a good thing in itself, but a quality education for everyone would almost certainly also have indirect beneficial effects on reducing poverty, perhaps on nutritional and health aspects too, as well as on reducing inequality. So here we have these indirect effects of one targeted effort at one goal. I refer to these as ripple effects.

Some of these goals have very heavy ripple effects, and they can be both positive and negative. Take a look at the SDG 8, which is economic growth. A lot of people in the tech sector are saying: “The data economy is built on us, we further economic growth”. But SDG 8 actually says that this growth should be inclusive and sustainable. So it should be beneficial also for those most in need. It’s redistributive in a sense that all the justice-related aspects of economic growth must be there. As such, real contribution to SDG 8 can have positive impacts almost all across the board. And the other around way: if you create more unsustainable and uninclusive economic growth that promotes concentration of wealth among very few people or corporations, then it will have negative ripple effects on almost all the other goals as well.

We simply can’t see technology as an isolated issue aimed at one specific problem. These systems are part of larger social networks and economies. That’s the contextual approach I advocate for.

And then we have three levels of impact.

Yes, these are the micro, meso and macro levels. Let’s take AI and VR in education. For example, using VR in interventions with people with autism, or making AI assistants that aim to support learning processes through the use of generative AI and learning analytics. We can distinguish between positive and negative effects here that hit various individuals – specific persons benefitting from the educational use of either of these technologies. These impacts are at the micro level.

Then we have effects on groups. This is where discrimination and bias come in. Is the AI assistant more effective for one gender, for example. But also the problem that AI developed in one part of the world may be inaccessible to the other. You might have different impacts across groups and regions both in a specific country but also between countries. Take VR again. It is likely that not all schools can afford the latest and most effective VR applications, or have the required infrastructure and competency to use it. If so, technology can increase differences, and this would not be conducive to reaching 4, and certainly not SDG 10. And this would be the meso level.

Finally, we get to the macro level which is the largest. It’s about the economic and political systems, e.g. how does AI relate to distribution of wealth, or to a transparent political institutions.

We need all these levels because they show us where the various impacts are. That prevents us from saying “AI has beneficial effect here so it must be good”. We get a broader picture which is beneficial for the analysis.

When one thinks about the AI ethics implementation, three major issues come to mind. The practical one – namely: the operationalization of AI ethics requirements (methods, tools, best practices, testing etc.), the formal one – which is its institutionalization (ways of ensuring compliance to requirements, like regulation, self-regulation, ESG reporting), and the systemic one which is integration (incorporating different AI impacts into one comprehensive analysis). Which one of them do you find most challenging at the moment and why?

I think it would be integration, or seeing the bigger picture, which is also about not treating AI ethics or regulation as something new. Because AI is often described as a technology that just popped up and we need a brand new approach to deal with it. But we have a long history of talking about intelligent or autonomous technologies. You can still have a lot of insights from Langdon Winner or Jacques Ellul, for example. Winner wrote about autonomous technology in a way that could be published today and people would think that he was talking about AI.

Seeing the bigger picture of contextualizing this in philosophy of technology, computer ethics, or in any other of these well-established disciplines, or even in private ethics, is really crucial. Because there are many old issues that still apply to AI today. I think one of the most pressing challenges is to avoid wasting time on reproducing things that are already there.

However, we still need codification or translation of how these broader kinds of ethical issues apply to problems faced by developers or companies. That requires more regulation and this is what the EU is doing. I like the regulation approach as opposed to the drive for professional ethics, where the industries themselves are the main players. Politics and government is required, because I don’t think the industry actors really have the society’s best interest at mind. At least not always.

The shift towards more integral approach when dealing with technology is also present in your latest paper co-written with John Danaher where you deal with proliferation of ethics in the digital technology domain. Nowadays different experts talk about tech ethics, AI ethics, computer ethics, digital ethics, roboethics etc. This is partly understandable as different technologies need practical solutions to solve particular problems, but you also notice risks stemming from this fragmentation. What are the biggest threats here, and what is the target audience of your appeal for ethics integrity?

First of all, there is a legitimate need for some of the different domains of ethics in terms of operationalization. For example if you are educating public policy makers at the municipal level you can’t assume that they will go and read old computer ethics literature. Contrary, you should explain: this is a new system, it does this and this, and we need to be aware of these impacts.

So, there is definitely an operational need for different domains of ethics in technology. But I also think that ethicists must discuss things on the proper level, and whenever we go one or two levels down (for example from engineering ethics, to computer ethics, to AI ethics) we need to make sure that what’s covered in each new layer or domain is exclusive to that technology. If it isn’t then we don’t need to go there, and we can keep discussing it as a more general implication of technology and facilitate debate across different fields.

For example, a lot of people talk about robot ethics these days but what they are actually talking about is AI or social AI. Why is that? Because the criteria for labelling something as robot ethics should include embodied AI, and this is not always the case.

I think there is a general confusion now because people are labelling the same things very differently and are losing sight of important insights from historical developments in technology ethics or privacy ethics. And that’s also a problem with developers because they don’t have the philosophical background in technology ethics. So we need to foster a debate where we clarify what already exists out there and how this applies to the new technologies.

The most important loss is that we get this duplication of efforts where people make arguably inferior research because they think that they discover a new approach to a problem that’s already been discussed for twenty years and has been further developed somewhere else. In a different discipline or a different domain of ethics.

I’d like to touch the friction between the ethical and the political now. Some people accuse AI ethics of privatizing responsibility. They believe that we should rather use tools from political philosophy to deal with questions concerning the whole public. Do you think that this dichotomy between the ethical and the political makes any sense here? And if so, what’s at the heart of the problem?

I absolutely think there is a need to keep the differences between ethics and politics in mind. Together with Eduard Fosch Villaronga I wrote an article called Research in AI has Implications for Society: How do we Respond? Its subtitle is: The Balance of Power between Science, Ethics, and Politics. These are three very distinct domains which have very specific roles to fulfil.

Sometimes people believe that ethics can be the foundation of politics. I think that is a mistake. Ethics is about highlighting, exploring and trying to understand the morality and the normative implications of particular aspects of our activities and beings. When we develop something, for example, the role of the ethicist is to tease out what the implications of this technology are.

But I think it’s up to politics to say what we as a society want. Of course, ethicists can make guidelines and write papers about what they think about the moral implications, but we really need to respect democracy and political processes and allow the people to make the decisions. And I think that’s good because otherwise we either get a moralized kind of politics, or politicized ethics. So it’s really important to keep those areas a bit distinct. I know this is controversial, though, but the liberal in me still very much feels that politics should not be restricted by a too heavy ethical hand, as pluralism and making space for living different lives where very different values and ideals are upheld is really important.

But it’s also really important to makes sure that science is not completely bound by politics or ethics. That’s more controversial, I know. There must be some kind of restriction on science, of course. But it’s really important to distinguish between the domains. Science explains the world and engenders new technologies and technological change, ethics does the evaluation and tries to figure out what happens when we do this and that, and then politics determines if we should apply the various new discoveries. The latter – the political – should, of course, be based on the advice from the ethicists and proper engagement with and involvement of citizens.

Speaking of science, another paper of yours is about the influence of the scientific discourse about robots on human self-perception. You call this effect robotomorphy, and together with attempts to anthropomorphize machines they seem to be two sides of the same coin. We perceive robots as something similar to us and then we start to look at ourselves as if we were robots-like. Can you explain the mechanics of this process?

The concept of robotomorphy is new but the origin of the term can be found in Arthur Koestler’s book The Ghost in the Machine, where he lamented how behaviorists’ approach to human science led to what he called ratomorphy. As we can’t do many experiments on human beings, we started to experiment on rats. But then we make this error of assuming that we are like rats. They are subjects of our study, and the foundations of our scientific findings. So we transpose what we work on back to humans, sometimes quite uncritically. This is what the robotomorphy term is inspired by.

The inherent logic of robotomorphy goes in two ways. Firstly, we are trying to figure ourselves out – as humans – by making machines that resemble us. So, we try to make AI based on neural networks, for example, inspired by what we think humans work like. And when we get these machines and they work, then we make the error of seeing machines in ourselves. We start reducing ourselves into what we are able to make in the machines. And that limits humanity. Phenomenology, all kinds of experiences that are an important part of being human become discarded because these are kinds of things we can’t really observe or replicate in the machines we make.

The other aspect of this is that we increasingly start to see machines as benchmarks of what we should be measured against, what we should strive to be, what’s the ideal. For example, we used to measure progress in chess and go programs by how close to human performance they were. But now they are the best, and they’ve become our benchmarks.

In the tradition of Western science we see rationality, optimization, and efficiency as an ideal, and machines are getting increasingly better at it. So we start adapting to what’s optimal for the broader system where machines are playing increasingly more important roles. That means molding humans into something more machine-like as well. Let’s take nudging for example. We try to correct human error because everything human is associated with unreliability and error – the anti-thesis of rationality and optimization.

Let me conclude with a question about big data as an ideology in science. According to another article of yours, big data is proclaimed as a master science of today – sort of an ultimate scientific paradigm. But it all seems like a revival of an old positivist dream. Is it something really new, or just a repetition of things that we had already witnessed over one hundred years ago?

This quest for a master science is really old and is very much related to positivism and behaviorism, but also to philosophers like Thomas Hobbes writing in the 17^th century. It’s about trying to discover general laws and believing that we can see humans as part of the natural world subject to laws that we discover through observations.

My article on this topic was written in 2018, but today this problem has only gotten worse and is now even more pressing. For example, just the other day Yann LeCun and Meta released a new discovery: with deep learning we’ve discovered the dark matter of molecules. AI systems now make these new scientific discovery, and humans are often left guessing how.

That goes back to an old idea that AI will have no theory, no ideology, and won’t be encumbered by human biases or faults. But I think today very few people honestly believe AI is really objective. Once they got at least a touch of the tech ethics debates going on they must have seen that how data is gathered, how it is contextualized, how algorithms are applied – all these things are obviously subjective or related to human decisions that impact how AI is performed. And this human aspect is deeply troubling and is by some seen as noise.

But we need to make sure that science is not only seen as a quest for objective mechanical laws of the universe but as a social and democratic undertaking with really huge social implications and consequences as well. Human decisions and human power are deeply entrenched into what sort of issues are discussed and what kind of framing for these questions we use.

Returning to the first question, isn’t this approach to big data or AI as an ultimate science an example of isolationistic point of view?

It can be related to this, yes. I think it is because it’s based on an approach where people, often unaware, end up excluding everything not covered by or not comprehensible within the paradigm of AI.

We have to contextualize our analysis of AI and bring the human and the social back in. A lot of what can be referred to as techno-solutionism is all too often subject to a debilitating blindness to wide areas that are crucial to human existence. Many experiential aspects of human life are lost if we restrict ourselves to machine-like or AI-compatible approaches to science. We can’t optimize or rationalize everything. We shouldn’t.

*Dr Henrik Skaug Sætra is a sustainability consultant and political scientist with a broad and inter-disciplinary background and approach, mainly focusing on the political, ethical, and social implications of technology. He focuses specifically on the sustainability related impacts of AI, and he has published a book and several articles on AI, ESG, and the UN Sustainable Development Goals.

Przeczytaj polską wersję rozmowy TUTAJ

Artykuł Henrik Skaug Sætra: Framing AI impacts with SDGs is beneficial pochodzi z serwisu Centrum Etyki Technologii.

Philip Brey: Ethics by Design is a feasible approach to AI

agagaj — Mon, 04 Jul 2022 13:52:24 +0000

Philip Brey:
Ethics by Design
is a feasible
approach to AI

4 July, 2022
agagaj

If you want to engineer ethics into AI products and systems then it requires specialized knowledge. It’s not just checking boxes. So if you have a development team add an ethicist, preferably one that is of a more applied persuasion. We also think that Ethics by Design should be part of computer science and engineering training, says Prof. Philip Brey in conversation with Maciej Chojnowski.

Maciej Chojnowski: About a year ago, at the first CET conference there was a discussion about the existing AI ethics tools and the possibility that they would form a basis for the future solutions for other technologies. Now we can see that it has already happened. In the SIENNA project, not only have you developed approaches for specific domains like AI, but also created a generalized methodology that can be used with any technology, haven’t you? In which areas would you expect its further adoption?

Philip Brey*: Currently there is a lot of interest in ethics of artificial intelligence. AI is seen as a powerful technology that is going to change all sectors of society and at the same time is raising significant ethical issues. And while this is true for AI, it’s also true for many other technologies like 5th generation genomics, synthetic biology, neurotechnology, nanotechnology or Internet of Things.

So time and again, we are going to need ways of dealing with the ethical issues and human rights challenges raised by these emerging technologies. Then it would be good if we could avoid reinventing the wheel and learn from previous technologies how to ethically assess them, identify ethical issues and ways to mitigate them at an early stage as well as ensure that this happens in a best possible way.

In the SIENNA project, we have developed ethics guidelines and tools for artificial intelligence, human genomics and human enhancement technologies but we have also developed a general approach that can be used for other technologies. For example, in the TechEthos project, which I am also involved with, we are now looking at neurotechnology, extended reality, and geoengineering. These are also very intrusive, impactful technologies that we need to ethically assess so it helps to already have the experience and the methods of how to do it.

One of the results of the SIENNA project, ETCOM, is a normative model for the coevolution of emerging technology and ethics. Is it an attempt to put an order on the already existing procedures so that they are easier to manage, or does it introduce new solutions?

In the past ten years, there’ve been attempts to create comprehensive models for the ethical guidance of emerging technologies under the heading of responsible research and innovation. But those were broader models aimed at governance and developing technology in a responsible and inclusive way where ethics was only one of the dimensions. Before that there has also been a lot of ethical analysis of emerging technologies but not really comprehensive models looking at ethical guidance and mitigation.

So we wanted to do something more specific for ethics what did not exist yet. And we decided to build a normative model that looked at how at different points in the evolution of an emerging technology ethical interventions can be staged so as to effectively take into account and mitigate for ethical issues in it.

Given that the ETCOM is a new normative model, one should not expect its full implementation yet. However, you probably used some existing components to conceptualize it. Can we then say that AI is already being developed in an ethical manner, or do we need to catch up with the right ethical approach, i.e. the one that you prepared?

AI is certainly a technology where many of the developments recommended by ETCOM have already been in place. But there is no emerging technology that really wins a prize for how ethics is dealt with. Ethics is still often an afterthought or not implemented in a really comprehensive and systematic way in the development, use and regulation of emerging technologies. But for AI, at least significant efforts have been made along different dimensions to address ethical issues at an early stage.

However, the ETCOM normative model goes certainly further than what has happened in AI. It identifies five stages in a development of an emerging technology where different ethical interventions are recommended. It starts with early ethical analysis and identification of potential ethical issues and implementation of research ethics protocols particular to that technology. Then we have later stages in which ethical guidance is developed for the technology for different actors, where the Ethics by Design approach is used and ethically inspired regulation and ethics standards are developed. And finally there are stages where ethics considerations are made part of ISO or CEN standards and where you have education and training.

So there’s a whole set of actions that one should take when one wants to successfully deal with ethical issues related to an emerging technology. It’s also important to see at what stage what should happen, what actors should be engaged in it, what knowledge they should have, and what tools they are using. That’s why we created these comprehensive recommendations on how that can be done for a new technology.

One of the most important outcomes of the SIENNA project is Ethics by Design, an approach allowing practical operationalization of high-level ethical principles in the AI design. It relates to a previous project you participated in – SHERPA. How did this EbD approach evolve?

These projects ran roughly in parallel but in SHERPA we had an earlier opportunity to work on Ethics by Design approach, whereas in SIENNA that was planned a little bit later. So that’s why we did an early work on this approach in SHERPA and then continued a more advanced version of it in the SIENNA project. And in SIENNA we also worked together with the European Commission to implement Ethics by Design in their Horizon Europe funding programme.

The European Commission has officially endorsed Ethics by Design, hasn’t it?

They endorse it as a recommended approach. Now it is part of the ethics review procedure for artificial intelligence, big data and robotics projects. We hope that in the future we may expand on this and Ethics by Design will also become a recommended approach for other technology projects.

You’ve previously mentioned TechEthos, another project in which you take part. Are you going to develop the Ethics by Design approach further on in it, or is the SIENNA version the final one?

In TechEthos not much room was reserved for Ethics by Design specifically. The project came into place before our Ethics by Design proposal for AI was presented. But in the SIENNA project we did some work to generalize Ethics by Design. There is a report which makes some proposals on how the approach can be generalized for other technologies. But it would be good at some point if one could do case studies of that for other technologies and develop it a bit more. I’m sure it will happen in other projects at some point.

Are you going to test Ethics by Design in practice?

It will be tested in practice through the different projects in Horizon that will implement the approach. Of course, we have done as much testing as we could during SIENNA but with limited means so it is not thoroughly tested. But it’s based on proven principles from earlier incarnations like Design for Values and Value Sensitive Design and on accepted design methodologies. So we are confident that it’s a feasible approach.

Some people in the AI community expect ethics tasks to be easily manageable. They would like them to be similar to a checklist or even (semi)automated. On the other hand, philosophers like Anaïs Rességuier, with whom you worked in the SIENNA project, put stress on ethical agency and claim that it is essential for moral reasoning. What’s your view on the possibility of automating ethics as part of designing or deploying AI?

Development processes are complex and they require a lot of expertise to perform them well. Anyone with programming skill can develop software but developing sophisticated software, especially in the field of AI, is a very complex process. So incorporating ethical considerations in it certainly is not going to be easy. That has to be taken into account.

If you take safety engineering, which is basically about adding one feature to developed products, you have whole methodologies and trainings for that to ensure that products are safe. We’ve seen the same for ensuring that information technologies are protective of privacy. So it shouldn’t be surprising that if you want to engineer ethics into products and systems then it requires specialized knowledge. It’s not just checking boxes.

Certainly some engineers would rather check boxes and be done with it but our view is that if you want to do ethics well it should be part of the development process, which requires specialized knowledge. So if you have a development team add an ethicist, preferably one that is of a more applied persuasion. But we also think that Ethics by Design is something that should be part of computer science and engineering training.

If one goes deeper into the theory of Design for Values, which is somehow related to Ethics by Design, one quickly realizes that there’s a need for the ability to translate values into design requirements. You said that we would need ethicists on board in AI but do you think that Ethics by Design as presented in SIENNA is ready to use, or will it require some further value translation?

We’ve made a lot of effort to do as much translation into operational procedures as possible. We also indicate how different steps in the development process of AI systems require particular actions or practices. But then there still will be interpretation needed when we recommend for example screening dataset for biases. It may not always be clear what that means in practice. And that is something that we cannot fully specify or else we would have had to write a 500-page handbook.

So it’s also incumbent upon the scientists and engineers to arrive at particular interpretations of the procedures that are proposed in Ethics by Design and decide how to eliminate bias, how to deal with informed consent at the practical level if it’s about privacy protection, or what kinds of AI systems require a high level of transparency and explainability. That’s not something you can all provide in a methodology of Ethics by Design. It requires experience with a technology development that needs to be then combined with the Ethics by Design approach.

In the SIENNA and SHERPA deliverables you also offer a methodology for the ethical deployment and use of AI.

Yes, it is a different approach from that of Ethics by Design because it pertains to different actors. Ethics by Design is for technology developers whereas the ethics of deployment and use is for organisational technology users.

We wanted to investigate what parameters are important here for determining whether the outcomes of deployment and use are ethical or not, and what organisations can do in order to ensure the ethical results. That involves a lot of things like having ethical guidelines in the organisation, careful selection of the right kind of technology when you acquire it or develop it in-house, staff training, and having policies in place.

It’s also important to have a multiactor perspective here and to look at the different tools and protocols that you need to support ethical outcomes. Because it’s not sufficient for an organisation to have ethics guidelines. They need a lot more: apart from policies and training, they may also need an ethics officer, some certification or accounting of their approach, monitoring. So if you want to really take ethics seriously in a deployment and use of new technology in your organisation, you need to look at these models of how to do that in a comprehensive way.

Are there any external tools that should be added to the SIENNA methodology?

The most important tools that we identify are those specific to development as well as deployment and use of the technology that we’ve already discussed.

But at the government level you need adequate regulations and policies. That’s what the European Union is actually doing. They have developed comprehensive regulations for AI which take into account some of the key ethical issues that are at play. On the other hand, CEN, ISO and IEEE are working on standards for AI that also include ethics. Universities are developing research and education programmes in ethics of AI, and that’s also a very important element. Media awareness for ethical issues in AI is another one.

So these are different significant ingredients. Some of them are kind of obvious, not essentially novel. But also for things that people already do like ethics guidelines we have developed a roadmap how to do it well because there are many ways possible.

One of your previous projects – SATORI, has resulted in an ethical impact assessment methodology. Is it also a part of SIENNA?

It is because to successfully deal with and mitigate ethical issues with new technology you need to be able to identify and analyse what the ethical issues are. And especially in the emerging technology, it’s not always clear what they are because some of them are in the future. They depend on the particular development path that is followed, the particular applications that will emerge and how they will be used in society and generate impact. That means that you need an ethics assessment methodology that not only looks at the current technology but also at its possible future incarnation and impacts.

We developed that initially in the SATORI project. And then in SIENNA we developed a more advanced version of that. It’s simply called the SIENNA approach to ethical analysis.

It’s always a good idea to do an ethical impact assessment early on when you have a new technology or when you are at the early stage of the development process of a particular application because you want to do an early screening of potential ethical issues. And it’s not just our recommendation. We’ve seen it in different publications of the European Commission too.

One of the articles you co-wrote with Bernd Stahl and other SHERPA researchers includes a taxonomy of AI ethical issues. You name three different types of issues there: machine learning issues, questions about living in a digital world, and metaphysical questions. To which of those issues would SIENNA solutions apply to?

We advocate for a broad approach to ethical issues. So for any emerging technology we want to look at all the important problems that are at play. If they include environmental issues then those should be considered as well. And machine learning is nowadays possibly at the heart of AI but it’s certainly not the only development in this field.

So if we want to do a broad analysis of a technology, we look at different kinds of products and uses of that technology and then identify what we see as the most important ethical issues. Those could relate to individual rights, fairness, individual or social well-being, or sustainability. Whatever seems pertinent for that particular technology should be taken into consideration.

*Philip Brey – professor of philosophy and ethics of technology at the Department of Philosophy, University of Twente, the Netherlands. In his research, he investigates social, political and ethical issues in the development, use and regulation of technology. His focus is on new and emerging technologies, with special attention towards artificial intelligence and digital technologies. Brey is former president of the International Society for Ethics and Information Technology (INSEIT) and of the Society for Philosophy and Technology (SPT). He currently coordinates the 10-year research programme Ethics of Socially Disruptive Technologies that includes seven universities in the Netherlands and over sixty researchers.

Przeczytaj polską wersję rozmowy TUTAJ

Artykuł Philip Brey: Ethics by Design is a feasible approach to AI pochodzi z serwisu Centrum Etyki Technologii.

Anaïs Rességuier: The critical approach to AI needs to be continuously restarted

agagaj — Thu, 09 Jun 2022 15:59:27 +0000

Anaïs Rességuier:
The critical approach to AI needs to be continuously restarted

9 June, 2022
agagaj

There is a tendency to consider AI as a highly sophisticated technology that would be almost virtual, intangible. This is very problematic because it means that we cannot see all the impacts of AI. So if we want to fully assess this technology looking at the whole AI lifecycle is very much needed, says Dr. Anaïs Rességuier in conversation with Maciej Chojnowski.

Maciej Chojnowski: In an article co-written with Rowena Rodrigues, you argue that today’s AI ethics is mostly based on high-level prinicples and remains quite abstract. You also give two reasons why AI ethics as we know it is rather ineffective. Firstly, because it has limited means to ensure compliance. And secondly, because AI ethics should not be treated as a soft version of law. Quite contrary, it should be seen as attention to context: a constantly renewed ability to see the new, which is „a powerful tool against cognitive and perceptive inertia”. What are the most striking examples of this inertia in the AI field today?

Anaïs Rességuier*: I would like to highlight two points here. The first one is inertia in acknowledging the disparate impacts of AI on different communities. We often fail to recognise that if an AI system doesn’t work for a particular community it is a failure of this system. And there are many systems that work poorly for specific groups of people. This should be considered a really significant problem, not just a marginal issue, and be properly addressed.

There is a study called „Gender shades” by Joy Buolamwini and Timnit Gebru about facial recognition systems which exemplifies this point very well. It shows that the standard evaluation of facial recognition systems that look at error rates over the whole population fails to account for major disparities across different communities. As this study showed, error rates for black females are much higher than those for white males.

This illustrates the fact that we need more fine-tuned, more precise lenses to assess this technology, one that look more closely at the characteristics of particular communities. I think that despite the evidence that this is highly needed there is still a lot of blindness towards it. This is the first point of this cognitive and perceptive inertia that I want to highlight.

And the second one?

The second one is the failure to account for the full lifecycle of AI systems, including the raw material they rely on, their environmental impact, the costs of producing and disposing of the systems components and so on.

There is a tendency to consider AI as a highly sophisticated technology that would be almost virtual, intangible. In another article, we call this the „AI immaterial narrative”. And this is very problematic because it means that we cannot see all the impacts of AI. For instance, a study on the environmental impacts of AI shows that training a single AI system produces as much carbon dioxide as five cars in their whole lifetime! So that’s quite a major impact. This is why looking at the whole AI lifecycle, not only at the algorithms that are being developed in laboratories, is very much needed if we want to fully assess this technology.

In her book „Atlas of AI”, Kate Crawford does a great job in redefining what AI is in all its different dimensions: the rare earth minerals it relies on, the infrastructure, the logistics, the data collection, and the impact of this data collection. So that’s a really broad system that needs to be taken into consideration when we look at AI. But I’m afraid we are still a bit blind to that. And this should definitely change.

According to Luciano Floridi, digital ethics is one of the three components of the normative system concerning technology (the other two being digital regulation and digital governance). And whereas regulation tells us what is legal and illegal, it has nothing to say about what we can do to live in a better society. This sort of moral evaluation is only possible within the ethics domain. But it seems that this is exactly the sort of capability that AI ethics is lacking today because of this widespread legalistic approach. What sort of actions and tools do we need to make AI ethics effective?

To answer this question I would make a distinction within the field of ethics that helps address a confusion I see quite often.

On the one hand, we can distinguish ethics as embedded within particular institutional frameworks. For instance, with bioethics there are well-established structures and frameworks that make it work and implement it in practice for medical research. At Trilateral Research, we work a lot with this approach to ethics, in particular with research ethics applied to new and emerging technologies. We provide guidance to research projects developing technological solutions to make sure they follow ethical obligations. The European Commission has a very specific research ethics framework that requires compliance from grantees. While working on projects funded by the EU we use our ethical expertise to implement this ethical framework. And this form of ethics is indeed a bit like a softer version of law.

In our article we were critical of this approach to ethics as soft law when it is not embedded within institutional frameworks. This is when it becomes indeed „toothless”. But it can be very valuable when embedded in institutions with mechanisms to ensure compliance. For instance if an EU-funded project does not comply with research ethics obligations, funding might be withheld or the project suspended. So there are ways to ensure compliance to ethical requirements.

In 2020-2021, the European Commission developed a research ethics framework for EU-funded projects using and/or developing AI. The SIENNA project had the chance to contribute to this effort and suggested ways to identify and mitigate ethics issues raised by projects using and/or developing AI systems or techniques. This framework is now in place and has become a requirement for all projects funded under the new Horizon Europe Framework Program that will be in place for the next seven years. So it’s going be interesting to see how it works and we already look forward to suggesting ways to further enhance it.

This is the first approach to ethics, ethics as embedded in institutional frameworks and there is assuredly the need to develop further this approach for the ethics of AI.

And this seems to be the dominant way of thinking about AI ethics today.

But there is also another aspect of ethics – very important although quite neglected. And it is actually the source of the first one that I have just discussed. This is what we developed in the Ethics as Attention to Context piece in the SIENNA project where ethics is more fundamentally seen as an open and continuously renewed attention to the world and to the new as it emerges.

This is where we come back more specifically to ethics as a way to reduce the cognitive and perceptive inertia. Here ethics is not about saying what we should or should not do as opposed to the first approach. It’s rather an attitude towards the world characterized by the attention and ability to determine the norms. Because only when we see the new that is happening, we are in a position to determine how to deal with it.

This is more an attitude, a form of ethical agency, than a judgement saying you should do this or that. The notion of „normative capacity” developed by a French philosopher of medicine Georges Canguilhem is helpful to clarify what is attitude or capactiy is about. It’s a capacity to determine the norm, not the norms themselves.

Could you elaborate on this distinction a little more?

We should distinguish these two approaches to ethics because their confusion gives ground to the manipulation of ethics. We’ve seen industry actors pushing strong for an ethical response to the regulation of AI. They favored the open-ended nature of ethics to avoid hard lines to be set on what they can do and cannot do with this technology. But this is made possible by a confusion between ethics as soft law and ethics as an open-ended process of reflection and attention.

We have a technology that is bringing significant, deep transformation to the world, to who we are, to our relations and so on. That’s why we need to be acutely aware of these changes to make sure we can guide this transformation, maybe resist some of its harmful developments as well as promote the beneficial ones.

And this is where the second notion of ethics comes back to the first notion. Because once we are able to see what is happening we can then ask ourselves what we should do.

When one is thinking about a possible taxonomy of AI ethics, three general approaches come to mind: the already mentioned legalistic one, the operational one, and the critical one. While the first one seems quite clear as it’s focused on defining principles and norms, the other two may be a little less obvious. So the example of the operational approach for me is Ethics by Design, while a critical approach could be exemplified by what you call for, namely Ethics as Attention to Context. However if we assume that attention to context would mean looking critically at the surrounding world, I suppose lots of people would say this is a negative approach focused mainly on banning things. Is there a positive component to it so that it can be turned into something constructive?

I think so. I am concerned about the fact that sometimes critical approach is just negative and doesn’t contribute positively to what is being developed. But it’s also because sometimes you do wonder when you see some innovation in AI. For instance, recently OpenAI developed a system, DALL.E 2 which, although it is highly innovative and powerful, is also generating very sexist and racist images out of text. So I do understand there might be some resentment within the community toward this. Maybe that leads to this critical approach being negative.

But I do think that we should be simultanously critical and constructive. That’s what we have to try to do with the Ethics as Attention to Context paper, to orient research and the AI sector in a way that is more beneficial for the society. That is a challenge especially with very powerful actors like Google and Facebook with whom there is a huge power assymetry. But I think that we should remain critical as well as contribute positively to all this.

In the SIENNA deliverables, it is mentioned that in the case of both Ethics by Design and Ethics as Attention to Context the AI practitioners whom you surveyed were calling for more practical and usable AI ethics. So it seems that in both cases the rationale was quite the same: to give people something more than just a collection of high level principles. The critical (contextual) approach seems to be an engine that helps thinking in the operational terms, doesn’t it?

I think the critical approach has really practical intentions to contribute to practices and decisions that are made. Our deliverable starts with quite a critique of the dominant approach, then makes a proposal and then operationalizes this proposal into a number of practical recommendations addressed to different stakeholders in the field. So there is this operational element.

It is also worth mentioning that sometimes the operational approach alone makes it seem too easy to implement really complex values or notions (e.g. fairness) in technology development. For instance, we have to respect the principle of fairness but on the other hand we know that it really is a major challenge to implement this in a society which is fundamentally unequal and unfair. So I think it is important that the critical approach always highlight that it is not so easy. You need to pay attention to ensure we don’t simply pay lip service to this really important principle.

The Ethics by Design and Ethics as Attention to Context offer different solutions or proposals to operationalize AI ethics and they do they complement each other. We didn’t have a chance to do that in the SIENNA project but we should bring them together. This should probably be the object of a future research project.

You say that Ethics as Attention to Context is also about enabling ethical agency, which is essential when we think about moral reasoning or moral evaluation. On the other hand, some people expect AI ethics to be (semi)automated or translated into a collection of algorithms that could be easily followed without much reflection. Do you believe that those two expectations could go hand in hand or maybe they exclude each other?

This question is becoming increasingly discussed. I am quite radical here. Talking about machines making moral deliberation makes no sense.

Obviously, we do need to acknowledge that norms, values and worldviews are embedded into AI systems. It’s very important to be aware of this. For instance, particular views on gender roles are at stake when voice assistants are systematically given a female voice. There is this famous quote from the historian of technology Melvin Kranzberg that says technology is neither good nor bad, nor is it neutral. We should examine technology because it’s not neutral.

But this is very different from saying that technology, a machine, or an algorithm can do moral deliberation. Humans remain the only moral agents. Especially if we define ethics as attention to context, it cannot be automated. And it is very dangerous to pretend it could. We can for sure embed particular values within a system but it doesn’t mean the system is making moral deliberation. We should really avoid pretending that machine can make it because it’s a way of not taking our responsibility for what we build. I think this is a dangerous slope.

Aside from the artificial moral agents, which is an issue on which I absolutely agree with your point of view, there is another aspect to that question. It might be not so much about creating the moral machines as rather about getting support when one does not have the required background to deal with ethical issues. Engineers might say: Ok, we will start with ethics now but the process will take two or three years instead of one year. So there is a problem how to deal with those questions, especially with people whose expertise in ethics is limited. Is it possible to reach some balance between the automation and reflection here?

Maybe the Ethics by Design framework is a way to seek this balance by providing some resources to do this moral deliberation and by breaking it down into more specific steps. I think it always requires to be rethought and put into perspective.

And the more potentially harmful a system is, the more time and attention it will require to be properly assessed. That’s the risk-based approach that is put forward by the European Commission. And maybe we can consider that for high risk systems we should make sure to have experts in ethical and social considerations in the developing teams. Probably there are systems where it is not so needed. We can have different approaches depending on the level of risk that a system causes. But again this would raise the question who and how would determine the potential risk of a system.

But I do believe that we need to develop AI ethics tools at different levels, starting from training of engineers. We should also have more social scientists and ethicists within teams that develop technology. Ethical issues shouldn’t be all delegated to engineers or developers who are not experts in these aspects.

Ethics as Attention to Context is about technology’s materialities, political realities as well as relations of power. These are all serious issues and very hard to tackle. In my view, to achieve the goals of the critical approach (here I also refer to recent books by Kate Crawford, Peter Dauvergne and Mark Coeckelbergh) we would probably need a sort of revolution in thinking about how the world is organized. So do you believe that certain solutions in this area can be put in practice step by step, or do we need to change the global attitude first?

Indeed, it’s very challenging. I think that the critical perspective needs to be continuously restarted. There can be no end to our efforts because we don’t live in a perfect world. And if we give up on the critical perspective the situation will become really dangerous in many different ways. For AI, it’s now clear that a number of key issues like bias or discrimination cannot be addressed simply from a technical perspective as the biases and discriminatory impacts keep coming back.

I’ve already mentioned the recent image generator from OpenAI called DALL·E 2. It creates really amazing images just from text. But at the same time when it’s asked to represent an image of a nurse or an assistant it brings images of women. How in 2022 can you have technology that brings such reactionary outputs? That’s really disappointing. But AI systems are not racist nor sexist in and of themselves. They just reflect the society. And society so far has been structured this way.

But I am also optimistic. There is an increasing acknowledgment that you cannot neglect those deeply contextual elements anymore and so this more holistic approach is being recognised in the field as necessary. The challenge is to find ways to operationalize this understanding. Bringing in social scientists and ensuring diversity in teams developing these systems can really make a difference.

Where would you expect the shift towards this contextual approach to start from? Among the AI practitioners as a bottom-up trend or maybe as a top-down initiative undertaken by some governmental or industrial bodies?

We need efforts from everyone, from all different kinds of position within the AI field. The SIENNA deliverable 5.4 starts with the Multistakeholder strategy highligthing the need to have all different parts of the society play a role including civil society, media, policy makers, AI developers, ethicists and so on. This really broad approach is neccessary because we are dealing with a technology that is very pervasive and has impact in many different areas of the society.

Disclaimer: This interview presents the position of Dr Anaïs Rességuier, based on results from the SIENNA project. It does not present the position of the SIENNA project as a whole, nor of Trilateral Research.

* Dr Anaïs Rességuier – Senior Research Analyst at Trilateral Research where she conducts research on the ethics of new and emerging technologies, especially Artificial Intelligence. She carries out this research as part of EU-funded projects, such as TechEthos and SIENNA. She is particularly keen to shift the focus of attention in AI ethics away from high-level abstract principles to concrete practices, contexts and social, political and environmental materialities.
Dr Rességuier is trained in philosophy, ethics and the social sciences. She enjoys working across different disciplines and engaging with a wide range of stakeholders to gain insights on current socio-technical developments and to contribute to shaping these.

Przeczytaj polską wersję rozmowy TUTAJ

Artykuł Anaïs Rességuier: The critical approach to AI needs to be continuously restarted pochodzi z serwisu Centrum Etyki Technologii.

Bernd C. Stahl: Human flourishing is at the core of AI ethics

agagaj — Tue, 12 Oct 2021 15:51:59 +0000

Bernd C. Stahl:
Human flourishing
is at the core of AI ethics

12 October, 2021
agagaj

If the EU regulation becomes law as planned and gets a stronger requirement for dealing with high-risk AI through things like ethics by design then the adoption rate of this approach will go up because companies will have to comply with it, says Prof. Bernd C. Stahl in conversation with Maciej Chojnowski.

Maciej Chojnowski: In your recent book, you delineate three main areas in today’s complex AI ethical landscape: specific issues related to machine learning, social and political questions concerning living in a digital world, and metaphysical questions about the nature of humanity and status of machines. You also offer a classification of governance mechanism of AI ethics, which are: policy, organisational, and individual mechanisms. Should these issues be seen as separate or complementary?

Bernd C. Stahl*: The discourse around ethics of AI is huge and through this categorisation I try to make it more manageable. But the classification of the ethical issues is not clear-cut. The distinction between the specific issues of machine learning and questions of living in the digital world are not really two separate things. The former have consequences for the latter. So I attempted to put some order on a chaos in a full awareness that this order is temporary and somehow idiosyncratic.

The reason for making this sort of classification is not just theoretical but also practical, isn’t it?

Yes, it makes sense to think in these terms because it allows us to distinguish between similar types of issues and ways of addressing them. Take a look at bias and unfair discrimination. This is a huge societal relation that permeates all societies. But where machine learning comes in we already understand some of the mechanisms which can lead to unfair discrimination, and find ways of dealing with them.

But there are also societal issues that are not subject to technical interventions. Justice of distribution or changing nature of warfare are big issues where no individual agent not even a powerful nation state can solve the problem. These questions have to do with distribution within or between countries. Some of them will be subject to traditional policy interventions from taxation to liability law. Other call for international agreements.

And then there are things I call metaphysical questions. These are sort of perpetual questions. AI triggers us to think about what is the nature of human in this world. I’m not interested in a Terminator scenario but these fundamental questions play an important role in the way we think about technology and human beings integrity. They are important even though they might never be resolved.

Is it possible to create a matrix combining these two taxonomies: the problematic areas and specific governance mechanisms?

To some extent yes, certainly for what I’ve called the specific technical issues of machine learning. There are ways in which we can at least think about how responsibility should be attributed. The proposed regulation by the European Commission tries to do just that: to think about what the problems are, how they can be addressed and who should be responsible, like in the ex ante conformity assessment.

But for other problems it’s much more difficult because they become a part of societal discourse or political negotiations. I don’t think there’s a lot of agreement even within most European societies what exactly would constitute a just distribution of resources. And therefore we can’t hope that we’re going to find a simple solution to the question of extracting surplus value from personal data. These are more open and long going questions and there are lots of players who have a role in that. Governments, companies but also civil society, which can be hugely important in stimulating the debate.

Lately, we’ve seen a lot of AI ethics guidelines being released by companies or international expert groups offering principles concerning AI ethics. But according to Brent Mittelstadt, one of the reasons why the prinicpled approach in AI ethics must cause problems is the lack of common aims among the AI practitioners. This is what distinguishes AI from medicine on whose codes of ethics the AI principles were based. Is human flourishing – seen as a purpose of AI, which is an idea you’ve been writing about recently – an attempt to solve the problem of those divergent aims within the AI community?

The reason for the application of principalism to AI is that the similar approach in biomedical ethics simply was the first one to establish itself as a significant political player. One of the advantages of this approach is that it exists and works to a certain extent. The disadvantage, especially with regard to emerging technologies, is that it was based – in its original context – on the common understanding that medical research is generally a good thing.

That sort of consensus of course is not existing with regard to emerging technologies. It’s not at all obvious that AI is gonna be the solution that society actually would like to put forward. Therefore there’s a fundamental question whether we as a society want to support certain activities. And it’s a question that prinicpalists’ ethics approaches can’t really deal with. So positing of human flourisihing is an attempt to respond to that. We recognize that some of technologies are more ethically beneficial while others are possibly ethically problematic or at least ethically agnostic.

The concept of human flourishing resonates with the recognition that human beings are at the core of this game. We don’t build AI for the sake of building AI but because it’s supposed to do something better for the world. But doing something better for the world is not just about maximizing profits or making sure that machine runs efficiently. It goes beyond that. It has a root in what human beings want to be, how they can live their lives, how they can achieve their aims. That’s why we’ve put this concept of human flourishing in the SHERPA project.

Looking at this concept from the operational point of view, do you think it can be used as a counterbalance to other general approaches to ethics like deontology or utilitarianism? Is there any practical advantage here over the other theories?

In terms of AI, it’s easy to see how you could try to implement a utilitarian approach in AI. That’s what a lot of the thought experiments around autonomous vehicles do, for example: do I kill this young child or these three old people? I think the concept of human flourishing could be a counterpoint to that. We chose it because it may help stimulate reflection at points where it’s needed. But this is only one way of thinking.

A lot of ethics has the reference to the three big groups: the utilitarian, deontology and virtue ethical positions. I don’t see them as contradictory but rather as different ways of looking at the same problem which may lead to different outcomes. But none of them individually has the right to say: this is the only solution we can go on with.

Do you perceive AI ethics as something to be used by engineers or deployers, or do you also find it applicable in the area of so called artifical moral agents? Some researchers, like Aimee van Wynsberghe or Scott Robbins, say we don’t need artificial moral agents at all. What’s your opinion on that?

I agree with it because I don’t think any technology that we currently have is capable of showing any type of meaningful moral agency. You can tell a machine to do something but you can’t tell it to reflect on why it did that and find justifications for that. And even if you could it probably still wouldn’t correspond to what humans do when they do that.

The ethics of human beings has a lot to do with our nature: the fact that we are embodied beings, that we know we are living a finite life, that we will die, that we suffer and that we are among similar beings which share the same nature with us. So I think all of these aspects play into ethics from the human perspective.

So unless we are capable of building machines that have similar characteristics to us, ethics in the machine would be something fundamentally different. And therefore I think the idea of machine moral agency doesn’t really make a lot of sense. I do think it’s an important question to ask what a machine should do in particular contexts like in the autonomous vehicle example. If a vehicle has a chance to make it safe, then these are worthy thoughts to think. But it’s very far away from ethics that we understand as a human point of view.

Therefore I agree with all the people who say: machine ethics doesn’t really exist and arguably it shouldn’t exist. We shouldn’t build anything that actually comes close to us because it will then muddy the waters as to what an ethical human agent is. However I don’t think that in my actual lifetime I need to worry about this because machines are simply nowhere near that.

Speaking of autonomous vehicles, there’s an interesting issue whether we should focus here on ethics alone, or rather get a broader perspective of political philosophy. The former might too much concentrate on individuals whereas the latter makes us think about the society as a whole. Do you think ethics of AI already covers political philosophy, or should we only include it in the AI ethical discourse?

As with many things, it depends on the definition. If you say ethics is around the individual, it wouldn’t be an error. But one of the reasons why I like the concept of flourishing and have certain affinity to virtue ethics is that for the ancient philosophers like Aristotle this distinction between the individual and the political discourse wasn’t really as relevant. And I think that’s certainly true in the AI discourse as well. Concentrating exclusively on individuals is not going to do the job.

So while in the SHERPA project we used the concept of AI ethics mostly because of the existing discourse people can relate to, in most of my work I prefer to use the concept of responsible research and innovation where this idea of having a broader societal context is baked in. It’s not just what the individual does or doesn’t do but it’s also about the societal processes that facilitate, motivate, bring things into being. And then ethics is part of that broader discourse which also includes politics, sociology and covers a lot of academic disciplines and aspects of society.

You’ve mentioned SHERPA but before we dive into details I’d like to ask you about the relations between SHERPA and its sister project SIENNA. In both cases we have deliverables on ethics by design co-authored by Phillip Brey. What’s the relation between these two? If one would like to implement one of these guidelines, which one should they choose?

Philip started this work in SHERPA and then finished it in SIENNA. So the deliverable in SIENNA is the key product. It also will form part of the future European Commision guidelines on AI ethics.

You call the guidelines operational or ready to use. Who’s their intended user and what sort of capabilities would one need to be able to put them into action? Is any ethical background neccessary to use them effectively?

The Twente group has put a lot of effort into shaping them according to existing development methodologies. The idea is that people who develop AI systems through agile processes or whatever else could use those guidelines to map the existing processes to the ethics requirements and build those in through the design process.

On the one hand, it should be relatively straightforward but I think there’s also recognition that if you leave it purely to checklist it may become problematic. So it will rather need to be taken up by an interdisciplinary team that has some affinity to the language used in philosophy.

But we haven’t really tested that in practice. So whether that will work is something I can’t roll out now. Probably we’ll need another research project to test that which I’d be happy to be both in with you.

The SIENNA deliverable suggests that implementing those guidelines may require radical transformation of the processes within organizations. What do you think about the organizational conditions for a successful implementation of ethics by design?

I don’t think there’s a single answer to that question. It depends to a large extent on the actual companies, what they do and what sort of technologies they use. I suspect that for a lot of companies where AI is part of their business processes it would be fairly unproblematic. If you’re an insurance company and you want to use an AI for fraud detection I would assume that ethics by design has the potential to resolve problems like racial discrimination.

But there may be other companies where it becomes much more radical. If a company makes use of AI for surveillance purposes, following the principle of ethics by design and thinking of the ethcial consequences of what they do may in the end result in a decision to just stop doing business.

In the SHERPA report you mention a survey conducted during the project. Only half of respondents confirmed that they had some sort of ethical mechanisms in practice at work. So the real percentage of organisations incorporating ethics in their everyday’s work was probably even lower as the people you surveyed were ethically-aware. Do you expect that once the ethics by design approach reaches a broader audience the rate of organizations adopting it will go up or down? Because with the high-level recommendations you can easily say that you follow general principles but with the complex mechanisms of implementation this might be more challenging. Which scenario do you expect?

I think it will depend to a significant extent on the environment and the sentiments the companies face. If the EU regulation becomes law as planned and maybe even gets some more teeth and a stronger requirement for dealing with high-risk AI through things like ethics by design then the adoption rate of this approach will go up simply because companies will have to comply with it.

Similarily I think there may be external societal effects that lead to higher adoption rates. Like another big scandal where some AI company goes out of business because they’ve done something really stupid and bad and the others will say: Whoops, we should do better.

But on the other hand, I realize that maybe the more specific the requirements, the more difficult it becomes to implement them. It’s also a question of choice. We give the market players ethics guidelines and the idea is that these are broadly applicable across the AI space. But companies may pick different approach like AI impact assessment and say this is how they deal with ethics. Some companies may put this into the corporate social responsibility, some into due dilligence, or human rights compliance. It may never appear under the heading of ethics but still be relevant. There’s lots of ways of talking about similar issues in different ways.

On the other hand, the Horizon Europe will support the ethics by design approach in scientific research on AI. You also mentioned that the European Commission would suggest that this is the way to operationalize the High-level Expert Group Guidelines. That creates opportunity for a broader adoption, doesn’t it?

Within the Horizon Europe – yes, the Commission wants to use this and make it compulsory. They will give it as part of the guidance to applicants who want to do something on AI. And given that there’s around €10 billion that’s going to be AI-related in Horizon Europe, that’s a significant investment into research. If all of the projects drawing on that funding will do ethics by design according to the SHERPA approach then of course this will become relevant.

However in comparison with the overall global AI research and development market, Horizon Europe is still relatively small. If you compare that with the Facebooks, Googles and Apples of this world, their entire research and development budgets are much bigger. Then it will depend on whether it turns out that the ethics by design process works. If companies come across this and say: we’ve seen that it is a good way, then it’s possible that it becomes much more prominent. Firstly on European level, but then possibly also internationally.

So let’s talk about the global landscape. Critics of the European trustworthy approach to AI claim that it would be very difficult for the EU to catch up with the global AI superpowers: the US and China. What’s your opinion on that?

There was a publication by Elsevier a couple of years ago where they looked at the AI landscape and tried to figure out what’s happening in the AI research world. They did the citation analysis and looked at how many AI related papers were written in which regions of the world. If you look at it by individual country then America was by far the biggest and China came after. But when you aggregated all the European countries together they exceeded the publication and citations of both – China and the US. So I’m not convinced that we are so much behind on the scientific level. But we are clearly behind in terms of the big players: Google, Apple, Facebook, Alibaba and so on – they are all American or Chinese.

The other question is: does our approach to the ethics of AI constitutes market impediment? There are people who say: I want a new AI start-up but I’m not going to do this in Europe because it’s too onerous. At this point it’s too early to speculate whether that’s going to be the case because we don’t know what the exact shape of the AI regulation will be.

However it might be instructive to look at similar developments, most notably the GDPR. You have two narratives: one saying it disadvantages European companies because they have to comply with additional bureaucratic hurdles. The competing narrative is that it sets the standard and enforces everybody who wants to be a part of the European market to live up to the expectations formulated by Europe. And you can certainly find evidence for the latter.

We can expect similar competing narratives to emerge in AI. I don’t think there will be one dominant narrative in the end. From the European perspective the question is: what do we want to do? Do we want to do the right thing to our citizens? I think that’s what the present attitude is. And if that’s the case and if AI raises particular issues then we need to address them and the current approaches are best we can come up with.

To conclude, I would like to refer to what Albena Kuyumdzhieva said during the final SHERPA event about incentives to implement AI ethics: that it’s crucial to translate ethcial requirements into competetive advantages. Do you believe it’s going to happen?

Yes, I do. Because people are worried what AI does to them and because there’s been such a high profile discussion of what the possible consequences of unreflected use of AI can be. There’s a high level awareness of at least some of those questions, like the discrimination disadvantages or mass surveillance. These are things people don’t want. And whether it’s through regulation or through voluntary adherence to standards or whatever, if a company can demonstrate that it has considered these issues and is serious about addressing them, it could be a competetive advantage for them.

There may be different mechanisms of communicating the fact that a company takes those concerns seriously. A trust badge on you your website, a certification, or a data base that the Commission wants to set up for all high-risk AI projects. And if they do that and if consumers recognize it, then this may well work out.

* Bernd Carsten Stahl – Professor of Critical Research in Technology and Director of the Centre for Computing and Social Responsibility at De Montfort University, Leicester, UK. His interests cover philosophical issues arising from the intersections of business, technology, and information. This includes ethical questions of current and emerging of ICTs, critical approaches to information systems and issues related to responsible research and innovation. He serves as Ethics Director of the EU Flagship Human Brain Project, Coordinator of the EU project Shaping the ethical dimensions of information technologies – a European perspective (SHERPA) and is Co-PI (with Marina Jirotka, Oxford) and Director of the Observatory for Responsible Research and Innovation in ICT (ORBIT).

Przeczytaj polską wersję rozmowy TUTAJ

Artykuł Bernd C. Stahl: Human flourishing is at the core of AI ethics pochodzi z serwisu Centrum Etyki Technologii.