When we hear the words “Artificial Intelligence” (or AI), most of the time our imagination rests on the Hollywood stereotype of Terminator. Ma the AI that is changing our lives is far more subtle, pervasive and dangerous. Mostly invisible, it silently penetrates into the very fabric of our society, into the way we do business, create and nurture friendships; it permanently influences our perception of the world, pushing us towards certain purchases, movies, songs or romantic relationships. In many cases, it reinforces and perpetuates historically racist prejudices, or on issues of gender and sexual orientation.
The quintessence of Artificial Intelligence is the ability to “learn from experience”, an idea introduced by mathematician Alan Turing in 1947. An AI system is designed to perform a specific task, but is not instructed on how to achieve ‘goal assigned by the person who planned it: learn from examples supplied by us, whether generated by human hand or by the machine itself.
This approach has proved hugely successful, especially in the field of image recognition. Today, we have automatic learning systems that recognize any object, face, handwriting or vocal instruction, and recommendation systems that guide our daily decisions. Proponents of AI claim to offer a tool that rises above human fallacy, inconsistency, ignorance, laziness and prejudice. A computer has no prejudices, does not favoritism, does not get bored and does not carry a grudge. But there are issues that cannot be ignored.
Much of the data on which AI algorithms base their learning is sourced from the Internet, which means that the resulting systems they inherit any prejudice contained therein, such as racist or gendered ones. When such disparities are incorporated into the system, the apparently “objective” algorithm will only reinforce and even multiply them over time. What is often presented as a system above human fallibility thus becomes a way to automate and mask – among the billions of carefully optimized neural weights – that same fallibility, often in ways more difficult to detect as purely statistical, but that cause real harm to real people.
A further question is: what goal should be given to the algorithm? From a mathematical point of view, the different definitions of what is “right” are incompatible: we can be right towards individuals, groups, categories, but not for everyone at the same time. Even establishing the loss function (which mathematically defines the goal of the system) is a task that requires a moral choice, and not a merely technical one.
In the words of Artificial Intelligence pioneer Joseph Weizenbaum, an MIT computer scientist who created a precursor to the current chatbot in 1966 and who later became an AI critic:
The relevant issues are neither technological nor mathematical; they are ethical. […] What emerges as elementary insight is that since we have no way to make computers wise, we should not assign them tasks that require the use of wisdom.
A third core of issues that emerged with the spread of Artificial Intelligence concerns power, responsibility and transparency. Online search results and social media feeds, tailored to our individual interests and profiles, create a bubble that polarizes opinion – with no democratic or institutional scrutiny, as the Cambridge Analytica scandal demonstrated in 2018. extremes are typically privileged by feed algorithms to maximize users’ viewing time, the true currency of the digital age: our eyes on the screen. It would be foolhardy to leave the decisions on issues like the democratic process and the mental well-being of our children to the web mega-companies, but that is in fact what is happening. We are accepting, often without realizing it, that we live in a moderately dystopian world.
Just as the splitting of the atom has given us clean energy and immense destructive power at the same time, Turing’s thinking machines have the potential to help us create a more equal and prosperous society, but also to amplify the current inequalities in terms of well-being and power, creating a world in which, like sleepwalkers, we trample on human freedom and action. A world of automatons, where children learn to swipe before they start walking, and are able to recognize hundreds of icons on the screen but not a single tree.
While Artificial Intelligence revolutionizes our society, rapidly redefining the essential parameters of being human, we find ourselves in a historical juncture where we groped beyond the material limits of the exploitation of the planet. After the rainforest, for the most part lost, after the oceans, overwhelmed by plastic, after the poles that melt from summer to summer, after the night skies crowded with artificial satellites, The Wise Man it is turning its extractive intentions towards the very essence of humanity. The decisions we make today will determine whether Turing’s legacy is a gift for humanity – as they have not seen since the days when Prometheus stole fire from the gods – or a irresistible poisoned fruit.
Roberto Trottaprofessor of Astrostatistics affiliated with Imperial College London and SISSA in Trieste, is among the speakers of Biennale Technology: he will participate in a dialogue with the philosopher Viola Schiaffonati, entitled “From the Big Bang to Artificial Intelligence”, on Saturday 12 November at 11 am, in Aula 6 of the Polytechnic of Turin.