Antifragile AI through ethical innovation
“Now is the time to turn our attention to ethically sustainable AI, which reduces uncertainty and supports innovation,” writes technology ethicist Salla Westerstrand.
In this article for Tekoälyfoorumi, Westerstrand examines how AI turns into value through sustainable business, even when markets are uncertain and innovation feels like it is stalling. With this advice, anyone can improve their own organisation’s chances of succeeding with AI.
AI appears to be here to stay
Almost every organisation is now looking for AI’s promised benefits and wondering how to realise them fully in their own organisation. AI has been shown to be useful in many professions, such as medicine, law, and logistics, and its potential to make work more efficient is discussed in almost every workplace.
The first experiments, however, have shown many organisations that the real benefits of AI do not materialise merely by encouraging employees to use ChatGPT. Many are wrestling with uncertainty and waiting to see whether growing investment will start to produce results over time.
AI is not the only disruption at hand, though. In his recent book (Disrupt with Impact, KoganPage, 2024), Roger Spitz describes the present as one of continuous disruption, in which many simultaneous upheavals sustain uncertainty, complexity, and a volatile market climate. AI has thus spread rapidly at a moment when we are also grappling with markets recovering from a pandemic, an uncertain political climate, and an environmental crisis. Many are looking to AI perhaps precisely for help surviving crises — maybe even flourishing in the middle of them.
In this equation, AI does not turn into value by itself. Now is the time to turn our attention to ethically sustainable AI, which reduces uncertainty and supports innovation.
Welcome, then, to the journey towards antifragile AI.
Fragile AI
AI is often marketed as a superpower suited to any task whatsoever, performing at least as well as a human. Generative AI in particular is anthropomorphised and mystified, which creates expectations of a miracle device with superhuman powers.
AI works very differently from a human, though, and there is no supernatural mystique to it. When an AI model generates text, it calculates the most likely next word in the sentence regardless of whether the content is correct — the latter cannot be calculated mathematically. It performs well when the expected output follows the same pattern as the previous comparable outputs the model was trained on. That is why, for example, generative AI tools for lawyers already produce good results.
Non-generative machine learning models are likewise based on selected training data, so that high accuracy can be achieved in predictive models for regularly recurring linear phenomena.
Because AI’s outputs are based on data fed to it and always curated by someone, the structural biases, errors, and value assumptions present in that data come along too. Research shows AI does not merely repeat but amplifies people’s own biases, and those who typically suffer most are people already in a weaker position. Unlike human biases, AI’s biases are also systematic and spread widely. When a tool like ChatGPT is used worldwide, its biases reach every group of users alike.
Bias is inherent to AI, since as a technology it is based on exploiting bias. Sometimes bias is not harmful — we generally do not want AI generating a picture of a blue orange when we just want an ordinary orange one. Because oranges are usually orange in AI’s data, the orange it generates is fortunately likely to be orange too.
If, on the other hand, we want AI to generate images of doctors and nurses, biases in the training data mean doctors are usually depicted as men and nurses as women. When AI algorithms learn these structures, the biases are reflected in AI-assisted decisions in recruitment, performance assessment, courts, or other situations that significantly affect people’s lives.
AI developers have already managed to correct the most glaring harmful biases. As Josh Simons points out in his book Algorithms for the People (2023, Princeton University Press), technical solutions that remove bias succeed on only one of several possible fairness metrics at a time. Because bias is not a technical but a social problem, removing it entirely through technology is impossible. Fixing it would demonstrably require proactively tackling structural bias.
Beyond bias, developing and maintaining AI is heavy on the environment and on an organisation’s resources. It brings societal challenges such as the spread of misinformation, harmful electoral interference, and manipulation, which threaten democracy, among other things. The big AI houses’ profits are often made at the expense of cheap labour in a weaker position in countries that ultimately gain little from AI development themselves.
All of these challenges make AI an inherently fragile technology, unsuited as a tool for solving every problem. Likewise, a technically mature organisation can be very fragile in the middle of an AI upheaval if its ability to identify AI’s ethical problem areas and weaknesses is limited and the opportunities for creating positive value are not seen on any scale.
Fragility shows up as uncertainty and a reluctance to experiment. It shows up as blanket bans on AI and individuals’ secret experiments in insecure environments, which at worst lead to trade secrets and personal data leaking into the wrong hands. As a result we pass over value potential that could be realised with good AI governance and ethical innovation.
Fragility can also show up as careless use of AI, as a result of which risks materialise that damage the organisation’s reputation, business, or end customers — risks that could have been avoided with skilled anticipation and strong ethical expertise.
If we want to create genuine value with AI that keeps giving even after the hype, we have to invest in the opposite of fragility: antifragility.
Antifragile AI
In his 2012 book Antifragile, Nassim Nicholas Taleb writes about how disruption demands antifragility of us. It is the level beyond resilience, one that does not merely help you survive an upheaval but turns it into a force that strengthens competitiveness.
Achieving antifragility requires the capability to understand complex phenomena. When we adopt AI, antifragility shows up as an ability to observe AI’s wider ethical and societal effects, to experiment safely, and to take controlled risks. It builds a strong foundation on which it is easy for everyone to identify the opportunities AI brings for improving competitiveness.
It makes use of the richness of a complex environment rather than being paralysed by upheaval and surprising turns.
An antifragile organisation does not settle into the role of passive bystander but gets on with it and puts in place the AI governance structures needed to accelerate innovation. It acquires the expertise with which AI’s possibilities and limits are identified at an early stage. That way resources are directed at solutions that produce sustainable value.
An antifragile organisation sees itself and AI as part of a wider, planetary digital ecosystem, so its attention turns from making current activity more efficient to creating something new. Then the whole system grows together and feeds global development in an economically, ethically, socially, and environmentally sustainable way.
Antifragile AI is not, in other words, only technology — it is an organisation’s structures, expertise, design thinking, and management. Around antifragile AI work data scientists, ethicists, lawyers, designers, strategists, and engineers, whose collaboration produces both the technology and the governance structures.
As Spitz describes in the book mentioned above, antifragility lays the basis for anticipatory business that turns AI-driven disruption into competitiveness.
Ethical sustainability – the first step towards antifragility
Many factors strengthen antifragility, a number of which Taleb and Spitz set out in their own books. The key to all of it, though, is managing complexity and uncertainty, and the number one tool for that is found in ethics.
Ethics brings to mind for many the philosophers of antiquity asking each other hard questions in the marketplace. Modern applied ethics — such as AI ethics — is something else. It offers instruments for resolving real-world ethical challenges and value conflicts that hamper innovation and the possibility of using a technology fully.
Ethical innovation demonstrably helps us identify more and more varied value potential. Ethics also strengthens an organisation’s ability to identify the ethical effects of AI that stand in the way of value creation. It helps us ask the right questions, reduce uncertainty, and thereby strengthen both the capability and the appetite for innovation.
With ethics, in other words, we better understand what AI is worth using for and which use cases fall down on ethical unsustainability.
Being ethical does not mean you cannot experiment. On the contrary — we often understand a technology’s effects best when we get to experiment and see how the algorithms work in practice. When ethical consideration is part of a culture of experimentation, we build AI on an antifragile foundation.
The questions at hand can be complex, but so is AI. Because the problems are not always technical, the solutions to apply are often not found on a coder’s desk alone.
Antifragile organisations take on hard questions fearlessly, because handling them is familiar to everyone. The organisation has skilled facilitators of ethical discussion, and its ways of working leave appropriate room for ethical discussion at the right points in the process.
Ethical consideration is a skill and a capability that an organisation can and should grow. Without ethical consideration we will not achieve the antifragility that makes AI’s competitive benefits possible.
Start by identifying where you stand and consider:
How used are we to asking unclear questions about responsibilities, values, or societal effects?
Do we know at what point a discussion about ethics accelerates innovation and when it slows it down?
Have we already supplemented our expertise through recruitment or identified partners whose expertise we can draw on?
If the answer to any of these questions is no, there is still some way to go to antifragility.
That is good news. It means there is still enormous latent potential in your organisation waiting to be found.
So take a firm step towards antifragility and grow your organisation’s ethical expertise. Get to grips with AI’s ethical and societal dimensions, experiment safely and in a controlled way, and invest in ethical leadership. An antifragile AI organisation grows stronger amid upheaval and has a positive effect on people and the surrounding world too.
You may soon notice hesitant expressions turning into insight and curious determination, and the AI upheaval into an accelerator of competitiveness.
Links and sources
https://www.nature.com/articles/s41591-024-03434-4
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5162111
https://www.sciencedirect.com/science/article/pii/S2667096823000551
https://www.nature.com/articles/s41562-024-02077-2
https://arxiv.org/abs/2403.02726
https://link.springer.com/article/10.1007/s13347-024-00814-z
https://www.wiley.com/en-us/Why+AI+Undermines+Democracy+and+What+To+Do+About+It-p-9781509560943
https://www.bloomsbury.com/us/feeding-the-machine-9781639734979/
https://link.springer.com/article/10.1007/s12599-023-00837-4
About the writer
Salla Westerstrand (LinkedIn) is a technology ethicist for a new era, helping organisations build sustainable value amid the AI upheaval. Westerstrand makes ethical leadership a cornerstone of digital strategy and turns AI governance and ethical expertise into competitiveness. She harnesses design thinking and design methods to shaping sustainable AI, so that both the technology and its governance support your organisation’s strategic goals.
Westerstrand coaches leaders and specialists in overcoming ethical uncertainty and in the strategic governance of AI. She is CEO of Harmless, a board member of Sustinaires ry, and an AI Designer at Solita. She is on the verge of completing a doctorate in information systems science at the Turku School of Economics, where she researches the ethical and societal effects of AI.
Enjoyed this article? Subscribe to get notified about new Tekoälyfoorumi articles straight to your inbox. We’d also hugely appreciate it if you left us a review on Google. Thank you. 💙
