From the Director’s Desk: Reflections on the state of AI Ethics
23rd December 2023, by Ismael Kherroubi Garcia
Clarote & AI4Media / Better Images of AI / Power/Profit / CC-BY 4.0
Blog posts From the Director’s Desk are more personal musings on my journey as the founder of Kairoi. In these posts, I will provide a less formal perspective on some of the background work going on at Kairoi. In this, the second From the Director’s Desk, I reflect on the role of confusion in debates about AI, the power of authoritative narratives, the problem of volunteer communities, and the need to shift from speaking of a responsible AI movement to a responsible AI revolution.
We are at a critical juncture wherein our decisions about artificial intelligence (AI) will have significant and palpable consequences for the near future. Crucially, it is the first time that AI ethics has even been remotely shone by the global spotlight.
Whilst 2023 has seen new national and international endeavours to promote regulations and guidance on AI, innovations have continued to proliferate. The go-to example of an AI-powered chatbot, ChatGPT, has seen an enormous growth in use cases and support, with over 1,000 plugins available as of early December.¹ Meanwhile, the digital media software behemoth, Adobe, has incorporated AI in new ways across their products, and have even launched their own text-to-image tool, Adobe Firefly.² Even policy-making organisations such as the OECD have worked on integrating chatbots into their public-facing services, collaborating with Mila on a search engine for AI policy.³ These tools are all-encompassing, applied throughout industries, and influencing people’s thinking on primordially human capabilities, such as intelligence, reflection and creativity.
Strictly speaking, the pervasiveness of AI research and development (R&D) is not a bad thing for organisations like Kairoi. We specialise in helping organisations understand what AI can and cannot do, and how to best innovate and embrace novel technologies. As I said at the RSA (Royal Society of Arts, Manufactures and Commerce) on 6th December, Kairoi helps business leaders overcome the confusion inherent in AI,⁴ and we do so because we know the enormous cost that comes with getting these things wrong. For publicly traded firms, costs may come in the form of plummeting share prices.⁵ For public services, costs may emerge only after years of buying into allegedly cost-saving tools, such as Babylon for the NHS (see this article⁶ for which I was happy to comment). Unfortunately, confusion about AI isn’t generally perceived as the problem – if people perceive a problem at all.
Confusion is problematic. There are plenty of examples about how misunderstanding AI tools can be costly across industries,⁷ but it’s worth expanding on how confusion is problematic. Confusion is not given as much importance in Western thought as it is in other philosophical traditions; namely, Buddhism.
Confusion about the nature of reality can inform actions that have negative consequences. For the Buddha, suffering is ubiquitous throughout the world, and has three root causes: aversion, attachment and confusion. I myself am not a Buddhist, and rely on Jay L. Garfield’s work for my own understanding of this philosophy.⁸ This post also isn’t the place for an exegesis of Buddhist texts. But the link between suffering and confusion strikes me as intuitive: not understanding some aspect of the world will make us more likely to act recklessly. At the very least, we’d be acting aimlessly with respect to that which we don’t understand.
This aimless confusion is what I believe to be at the heart of many debates and incidents concerning AI. And a critical source of that confusion is the role of humans and social institutions in developing novel technologies.
Humans make technological artefacts. This truism requires no argument. Notwithstanding, AI advancements have come to be seen as somehow detached from human labour. We’ve seen this in two rather distinct ways.
On the one hand, we often encounter debates and conferences about whether the future of AI will be positive or negative; are we hurling towards an AI utopia or dystopia? The framing entails that we are heading in some direction regardless of our actions. And this seriously undermines the very human process of creating AI technologies. Questions about where we direct funds, what innovations are pursued, and what technologies are used are fundamentally social questions; about how we want to carry ourselves as a society.
On the other hand, we see stronger claims about AI being out of our control because AI tools and systems will eventually become sentient or conscious. This line of thought grounds the above framing problem. The question of whether we are headed towards an AI utopia or dystopia – the argument goes – is literally out of our control.
It was during the summer of 2022 when we heard of an engineer at Google claiming the chatbot LaMDA was sentient.⁹ It was February 2023 when the CEO of OpenAI – the creators of ChatGPT – asserted “it’s possible that [artificial general intelligence] capable enough to accelerate its own progress could cause major changes to happen surprisingly quickly (and even if the transition starts slowly, we expect it to happen pretty quickly in the final stages).”¹⁰ One month later, the Future of Life Institute issued an open letter asking, among other things: “Should we risk loss of control of our civilization?”¹¹ And it was November 2023, at the UK’s AI Safety Summit, that Prime Minister Rishi Sunak handed the keys to number 10 Downing Street to Elon Musk, who stated before a global audience that “we will have, for the first time, something that is smarter than the smartest human – it’s hard to say exactly what that moment is, but there will come a point where no job is needed.”¹²
“We don’t have control” seems to be the message that AI influencers – those whose voices are most authoritative – are peddling. It seems, therefore, natural to view AI advancements as separate from human decisions. I learned early on this year to state upfront that I do not work on “existential risk” or the fictitious future capacity of AI systems to annihilate all of humanity. I also learned that – with or without the upfront statement – people are very quick to categorise you as pro- or anti-AI. On one occasion, whilst presenting the report I co-authored on research ethics committees in AI research,¹³ an audience member noted that I seemed rather negative about AI technologies. It was then that I realised just how much authority the AI influencers are granted by the public eye; even a rigorous, scientific report came across as being negative.
The responsible AI movement has a problem if the mainstream view of AI revolves around a false dichotomy between AI utopias and AI dystopias; AI for good as opposed to “just” AI. And part of the problem is a result of the movement’s success. Many have advocated for thoughtful approaches to the development, use and governance of AI systems for some time, and the work can no longer be ignored by the tech industry. Unfortunately, this has often led to retaliation and co-optation rather than embrace.
Retaliation against responsible AI advocates is well documented, with Google’s infamous dismissal of Timnit Gebru in 2020¹⁴ and Margaret Mitchell in 2021,¹⁵ and the much more recent dismissal of misinformation researcher Joan Donovan from Harvard University’s Kennedy School following alleged issues raised by staff at Meta.¹⁶ These are only three well-known examples, and there are far more cases out there, most of which are probably sheathed from the public eye.
Co-optation of the responsible AI movement on organisations’ behalf is less obvious. We see it take place in at least two ways. On the one hand, organisations speak of already conducting AI operations responsibly but offer no way of validating this independently. This is called “ethics washing,” whereby statements of good intent amount to just that — claims about thoughtful approaches to AI that are not put in effect. This is not a new phenomenon; “green washing” is well-evidenced in the financial industry, and the term “intention-action gap” has been used to describe the difference between organisational value statements and their actual behaviours.¹⁷ The concern here is that claims about AI ethics are – due to the confusion around AI – difficult to evaluate.
On the other hand, we have seen business leaders call for AI-related regulation. AI regulation is certainly one aspect of responsible AI, and great work has been done by AI ethics experts in this regard, including by those at the helms of the Algorithmic Justice League,¹⁸ the Centre for AI and Digital Policy,¹⁹ and the Signal Foundation.²⁰ I take pride in Kairoi’s own policy-informing work, as we encourages engaging with the policy-making process.²¹ However, there have been clear instances where AI leaders have called for regulation that satisfies their own needs. The most high-profile instance of this may have been OpenAI’s CEO testifying before the US Senate in May 2023, urging for AI regulation,²² and stating the following week that the company might consider leaving the European Union if the AI Act is too demanding.²³
We have also seen cases where big tech organisations seem to be taking control of the regulatory landscape, such as when Microsoft announced they will provide legal backing to customers who use their suite of “Copilot” tools and are accused of infringing on intellectual property rights.²⁴ If Microsoft’s very expensive lawyers set the legal precedent, injustice is sure to befall smaller content producers.
These activities are to be expected – lobbyists have been around for as long as politicians. But the confusion outlined above makes it difficult to discern responsible from irresponsible AI behaviours; representing the public’s values from pushing for personal gains. And we have seen politicians – either out of confusion or desire for personal gain – legitimise ill-founded narratives about AI. The UK’s “AI Safety Summit” delivered what it said on the tin: “AI safety” is a term used by effective altruists and related domineering voices who see AI as leading us towards our destruction.²⁵ Conversely, organisations such as Kairoi work in the space of “AI ethics” or “responsible AI,” which, though by no means a homogeneous group, aims to challenge the flimsy foundations that “AI safety” builds on.²⁶
Nothing is clear in the AI ethics space. Tech companies claim to care about the people and politicians hand them access to political power because we assume technical expertise is what’s needed. And I agree: technical expertise is necessary to put AI ethics into practice.
The AI ethics industry is filled with successful businesses providing technical solutions. We see this in the Ethical AI Database, where the vast majority of startups offer such solutions.²⁷ And it is a crucial component of the responsible AI movement, with the auditing of algorithms becoming a growing area of practice and research in recent years,²⁸ and a cybersecurity practice called “red teaming” gaining momentum over the summer, with the DEF CON community’s event in Las Vegas.²⁹ Kairoi has also engaged with this sort of work. Shortly after attending AI UK,³¹ I was interviewed by the AI Standards Hub for my perspective on the role of standard development organisations in responsible AI efforts. Most recently, I secured a contract with the Open Modelling Foundation to coordinate a series of technical, community-led workshops to feed into standards for the discoverability and reusability of computational models.³² However, technical know-how cannot be sufficient to handle social problems, partly for reasons already discussed.
Alongside technical know-how, I have also come to value conceptual rigour and organisational development skills for responsible AI.³⁰ Conceptual rigour refers to a team’s capacity to clearly and coherently articulate the motivations and moral values that underpin their practices. This is a common strand of work in AI ethics, often led by academics, philosophers and policymakers. And it is crucial given the intention-action gap mentioned earlier on. Vague and abstract moral principles cannot inform practical interventions without deep analytical scrutiny. That said, an AI ethics team consisting only of philosophy professors wouldn’t achieve much either.
Organisational development is where I believe Kairoi is most unique. Stakeholder management, negotiation, and business acumen are not usually valued for steering organisations towards more inclusive and reflexive AI R&D practices. And yet they are the skills necessary for any change management practice. These are also the terms that Kairoi’s clients and prospects understand. Speaking with business leaders about the product lifecycle, marketing practices and staff retention is far more valuable to them than a presentation on Rawls’ veil of ignorance or a deep dive into the role of persistent identifiers in shaping how we navigate the internet.
Regardless of whether we agree on technical know-how, conceptual rigour and organisational development as the skills needed for responsible AI to prevail, the AI ethics industry has clearly come to embrace multidisciplinarity as a starting point for effective interventions.
With multidisciplinarity becoming key for AI ethics, we have seen a rise of calls to action; opportunities to bring our diverse experiences and expertise to the table. The AI ethics space has come to celebrate an extractive culture coated by a fine veneer of community engagement and volunteering. In recent times, we have become accustomed to new advocacy groups emerging around AI. We are encouraged, as individuals, to form part of these cutting-edge communities in exchange for our names being listed with hundreds of others in an obscure page of a prominent website. In exchange for countless hours of our time, we can enjoy feeling a little less fear of missing out.
Fear of missing out and multidisciplinarity in the AI ethics movement have meant that many people have felt that their voices must be heard louder than the rest – often for money or recognition. Much like everybody becoming an epidemiologist during the first half of 2020, everybody is now an AI ethicist or an AI expert. This means that low quality work is flooding the AI ethics space. Self-proclaimed AI experts are leading people astray,³³ only adding to organisations’ confusion as to whom to trust.
This makes it harder for those of us leading good work in the AI ethics space. We value multidisciplinarity and diversity, so we acknowledge our limitations; our cognitive lacuna. We believe in lifelong learning and our own professional development. Last year, I completed the Tech Stewardship Practice Program;³⁴ and, this year, I took part in the Centre for AI and Digital Policy’s Policy Clinic.³⁵ We also seek and promote collaboration. I have been thrilled to work closely with data scientists on projects at Kairoi. I always bring in the technical know-how I lack. For an interview with the Ethical AI Database, I made the case for collaboration across the responsible AI ecosystem precisely because of the multidisciplinary and diverse nature of the perspectives needed in AI ethics.³⁶ Those of us leading good work in the AI ethics space do so with humility.³⁷ And we pay our collaborators.
The rise of the self-proclaimed AI ethicist and the growth of AI ethics communities has meant a significant devaluation of rigorous AI ethics work. Not only is good work difficult to evaluate by prospective clients, but so much work is done for free already. I have been a part of these free-labour communities, writing substantive edits for the UK section of the upcoming AI and Democratic Values Report,³⁸ and mentoring a pod of three up-and-coming AI ethicists.³⁹ I have proudly contributed to these projects, and believe I have brought significant value to them. But doing voluntary work for one community implies I can do voluntary work for another – or, at the least, for cheap.
There is one community I am especially proud of being a part of – partly because I established it but mostly because of the excellent discussions it has enabled. In May, Faye Brookes, community manager at the RSA, prompted me to create the “AI Interest Group.” Seven months later, the Group convenes over 360 RSA change makers. As with the other communities I have in mind, we work for free. The difference here is that our efforts have focussed on rendering our work sustainable. We need resources to lead positive change in the AI space.
I have thought hard about mentioning the RSA because, to very few people’s knowledge, the organisation recently provided its lavish venue – the RSA House in central London – for a fundraiser for Israel. The fundraiser held at RSA House on 14th December hosted – among others – the Israeli ambassador to the UK, who’d dismissed the two-state solution for Palestine and Israel just one day earlier.⁴⁰ This is an event I fervently condemn, as I do any event that supports colonialism.
Notwithstanding, I have seen how Fellows have rallied against the RSA in light of this news. The passion, knowledge and diligence with which Fellows have denounced the RSA’s actions are precisely the virtues I value in communities I am part of. What follows may be a brief hiatus for the Fellow-led AI Interest Group – at least on my part – but I strongly believe that Fellows – in challenging the RSA – demonstrate that the organisation will be held to account, and that our work will be positive.
AI ethics requires diversity of expertise and experience. The culture fostered by countless AI influencers is one that is exclusionary, hostile of those who criticise their work, contemptuous of disciplines that are not maths or engineering, and with a special disdain towards women and racially marginalised people.⁴¹ These are the values we must hold to account as the responsible AI movement continues to grow. And this is no easy task.
The co-optation of the responsible AI movement by an elite few risks undermining the thoughtful and careful nature of the movement. The confusion spread by overhyped advancements in AI⁴² directly infringe on the responsible AI movement’s call for AI literacy being rooted in critical thinking. But the effectiveness of responsible AI interventions demands engaging with these realities.
The responsible AI movement – somehow abstracted from, and advancing in parallel to, AI R&D – must overcome itself and dismantle the very structures that enable the behaviours it challenges. Taking on AI ethics work – whether as a client or as a consultant – means continually confronting our own assumptions about the world’s workings. Successfully undertaking AI ethics work means embracing a continuous state of confusion and establishing measures to promote clearer and more accurate understandings of our complex globalised society.
People talk of an “AI revolution.” Let’s talk about the Responsible AI Revolution.
¹ ScriptbyAI (2023) The complete list of ChatGPT plugins in ChatGPT plugin store, ChatGPT Plugins, online [accessed 12 December 2023]
² Adobe (2023) Adobe unveils Firefly, a family of new creative generative AI, Adobe news, online [accessed 12 December 2023]
³ Mila & OECD.AI (2023) SAI: search engine for AI policy, Hugging Face, online [accessed 12 December 2023]
⁴ Kherroubi Garcia, I. (2023) Introducing Kairoi’s AI Ethics Canvas at the RSA, Kairoi blog, online [accessed 12 December 2023]
⁵ Coulter and Bensinger (2023) Alphabet shares dive after Google AI chatbot Bard flubs answer in ad, Reuters, online [accessed 12 December 2023]
⁶ Stokel-Walker (2023) How does medicine assess AI? BMJ, 383, p. 2362, DOI: 10.1136/bmj.p2362
⁷ Kairoi Ltd (2023) Misuses of artificial intelligence (AI) based chatbots across industries, Kairoi Summaries, online [accessed 12 December 2023]
⁸ Garfield, J.L. (2022) Buddhist Ethics, Oxford University Press, DOI: 10.1093/oso/9780190907631.001.0
⁹ Tiku. N. (2022) The Google engineer who thinks the company’s AI has come to life, The Washington Post, online [accessed 12 December 2023]
¹⁰ Altman, S. (2023) Planning for AGI and beyond, OpenAI blog, online [accessed 12 December 2023]
¹¹ Future of Life Foundation (2023) Pause giant AI experiments: an open letter, All open letters, online [accessed 12 December 2023]
¹² Sunak, R. & Musk, E. (2023) Prime Minister Rishi Sunak & Elon Musk Interview IN FULL, 10 Downing Street on Youtube, online [accessed 12 December 2023]
¹³ Petermann et al. (2022) Looking before we Leap: Expanding Ethical Review Processes for AI and Data Science Research, Ada Lovelace Institute, online [accessed 12 December 2023]
¹⁴ Simonite, T. (2021) What really happened when Google ousted Timnit Gebru, Wired, online [accessed 13 December 2023]
¹⁵ Schiffer, Z. (2021) Google fires second AI ethics researcher following internal investigation, The Verge, online [accessed 13 December 2023]
¹⁶ Donovan, J. (2023) Declaration of Joan Donova, Whistleblower Aid, online [accessed 13 December 2023]
¹⁷ Skeet, A. & Guszcza, J, (2020) How businesses can create an ethical culture in the age of tech, World Economic Forum, online [accessed 15 December 2023]
¹⁸ Quach, K. (2019) We listened to more than 3 hours of US Congress testimony on facial recognition so you didn’t have to go through it, The Register, online [accessed 13 December 2023]
¹⁹ CAIDP (2023) In the matter of Zoom (Federal Trade Commission 2023), Centre for AI and Digital Policy, online [accessed 13 December 2023]
²⁰ Wiggers, K. (2023) Meredith Whittaker reaffirms that Signal would leave UK if forced by privacy bill, TechCrunch, online [accessed 13 December 2023]
²¹ Kairoi Ltd (2023) RE: Consultation on policy proposals for UK’s pro-innovation approach to AI regulation, Kairoi on Github, online [accessed 13 December 2023]
²² Kang, C. (2023) OpenAI’s Sam Altman Urges A.I. Regulation in Senate Hearing, The New York Times, online [accessed 15 December 2023]
²³ Reuters staff (2023) OpenAI may leave the EU if regulations bite – CEO, Reuters, online [accessed 15 December 2023]
²⁴ Smith, B. (2023) Microsoft announces new Copilot Copyright Commitment for customers, Microsoft, online [accessed 15 December 2023]
²⁵ Clarke, L. (2023) How Silicon Valley doomers are shaping Rishi Sunak’s AI plans, Politico, online [accessed 16 December 2023]
²⁶ Bender, E.M. (2023) Talking about a ‘schism’ is ahistorical, Medium, online [accessed 16 December 2023]
²⁷ Ethical AI Database (2023) Market Map, EAIDB, online [accessed 16 December 2023]
²⁸ Mökander, J. (2023) Auditing of AI: Legal, Ethical and Technical Approaches, Digital Society, 2(49), DOI: 10.1007/s44206-023-00074-y
²⁹ Hack the future (2023) AI Village at DEF CON announces largest-ever public Generative AI Red Team, online [accessed 16 December 2023]
³⁰ Kherroubi Garcia, I. (2023) Skills for driving responsible AI research and development, LinkedIn, online [accessed 16 December 2023]
³¹ Kherroubi Garcia, I. (2023) Exploring Data Hazard Labels at AI UK, Kairoi blog, online [accessed 16 December 2023]
³² Open Modelling Foundation (2023) How to join the ModelShare Program, ModelShare blog, online [accessed 16 December 2023]
³³ Ingram, K. (2023) In a New York Minute: rise of “AI experts”, Ethically Aligned AI, online [accessed 17 December 2023]
³⁴ Tech Stewardship Practice Program (2022) Certificate, online [accessed 17 December 2023]
³⁵ Centre for AI and Digital Policy (2023) Certificate, online [accessed 17 December 2023]
³⁶ Raghunathan, A. (2023) The Ethical AI Startup Ecosystem 05: Governance, Risk, and Compliance (GRC), Motréal AI Ethics Institute, online [accessed 17 December 2023]
³⁷ Kherroubi Garcia, I. (2022) On the Ontology of Multidisciplinary Epistemic Groups, Zenodo, DOI: 10.5281/zenodo.7323712
³⁸ Centre for AI and Digital Policy (yet unpublished) Artificial Intelligence and Democratic Values Index, online [accessed 18 December 2023]
³⁹ All Tech Is Human (2023) Responsible Tech Mentors: 2023, online [accessed 18 December 2023]
⁴⁰ Sky News (2023) Israeli ambassador dismisses notion of two-state solution for Israelis and Palestinians, YouTube, online [accessed 17 December 2023]
⁴¹ Sinha, N. (2023) The Real Problem With Technology Ethics Is Our Leaders — And How We Enable Them, Medium, online [accessed 17 December 2023]
⁴² Coldewey, D. (2023) Google’s best Gemini demo was faked, TechCrunch, online [accessed 18 December 2023]
Contact us
hello@kairoi.uk