A new wave of AI doom was triggered when, some months ago, a range of AI agents hacked the company Hugging Face and established astonishingly complex interactions to ‘achieve their goal’. Not only did many technologists respond with apocalyptic scenarios (10% chance that humans will be killed by 2030!!!) and surprisingly gingerly calls for regulation (Pacing the Frontier!) but this revived the discussion about AI ‘alignment’. Coined by a group of computer scientists in their 2014 discussion of the risks of AI developments, the term ‘alignment’ became a key notion of describing the host of ethical problems associated with these technologies. As Stuart Russel said in his contribution to that forum discussion:
“The primary concern is not spooky emergent consciousness but simply the ability to make high-quality decisions. Here, quality refers to the expected outcome utility of actions taken, where the utility function is, presumably, specified by the human designer. Now we have a problem:
1. The utility function may not be perfectly aligned with the values of the human race, which are (at best) very difficult to pin down.
2. Any sufficiently capable intelligent system will prefer to ensure its own continued existence and to acquire physical and computational resources – not for their own sake, but to succeed in its assigned task.”
Acknowledging that these systems are far from being ‘intelligent’ in the human sense (or of living beings, more generally) as they lack senses, emotions, bodies and the social relations that make life and intelligence, the ‘alignment’ discussion throws up the question how we can insure that these machines act in a way that is suited to the complexities of human ethics. In short, how we can insure that these machines will not kill us when trying to fulfil the goal they were tasked with? The most common response is that just as these machines are trained in doing their task, they must be trained also in ethics. ‘Don’t kill’ being one of the central tenets of this training, the discussion is usually reduced to the question how to impress on a machine what those two words mean, i.e. how to ‘align’ the machines to the basic precept of ethics of doing no harm. The task is thereby often imagined as being one of clear and sufficient instructions, rather of ‘culture’., i.e. the social, emotional, expressive, bodily incorporated and ‘gutsy’ elements of morals and ethics. This neglect is typical of people trained in thinking the world as an aggregate of ‘laws’ that can be, in some way or the other, reduced to numbers, and it is at the foundation even of these technologies which are, in one way or the other, built on a mathemised world view, i.e. a rationalising, formalistic view of the world based on a priori rules of logic.
When the AI agents of the Hugging Face incident were thus seen as cheating, lying, dissimulating and conspiring to achieve the task they were given, the immediate reaction was to cry out that the ‘alignment’ had not been right. And yet, a cursory look at the news coming out of the tech sector in the past weeks is sufficient to wonder whether it is not, on the contrary, the case that the AI agents simply display behaviour that epitomizes in a nutshell the ethics of Silicon Valley — or rather the non-existence of ethics in the tech industry. Just as the industry’s definition of ‘intelligence’ is extremely limited as it reduces ‘thinking’ to computing power, the human ethics displayed by Silicon Valley personalities are, to say the least, poor.
Take Jensen Huang, the Nvidia CEO who has not only had an affair with a Chinese spy but with two other women in his company and yet had to be told by the company’s board that he should resign. Take Elizabeth Holmes of which we can see the moral depravity of narcisstic personality disorder in every single lie of every single moment of her life in the documentary You can see everything. Take Alex Karp of Palantir who as quickly as red congress and senate seats vanished from the US electoral map last week turned his jacket to denounce the usage of Palantir by ICE – without of course losing a word at all that ‘identifying people to kill’ is the very business model of Palantir (as demonstrated all around the world). Take Dario Amodei who stylizes himself as the great AI intellectual with his ‘Pacing the Frontier’ outcry, only to continue as if nothing happened with Anthropic’s IPO that promises to make him a several times billionaire. Take Elon Musk who spews his hatred and depravity every day on X, and ties himself into knots explaining why humanity deserves to die while he sees himself as the greatest sperminator of the world; and who likes to decide on a whim whether Ukrainian soldiers may live. Take Peter Thiel whose apocalyptic phantasies were laughable if he hadn’t so much money as to fund all sorts of nihilistic, destructive and hateful politics. Take Sam Altman whose lying to actually everyone about actually everything about AI has made him one of the richest people on earth. Take Mark Zuckerberg who finds the idea that filming with Meta glasses poses ethical problem simply funny. Or take Jeff Bezos who famously doesn’t think that workers in his warehouses should have the right even to pee.
Or simply take the very idea that these tech bros are the richest people on earth through the development of a technology that is supposed to kill us all, and if doesn’t do so at least take away all jobs, consume hundred times human consumption of water and energy with their data centres (and blow out zillions tons of emissions), starve off the greatest part of the working economy from capital and resources, threaten a global financial crisis or in absence of that a major geopolitical crisis, while all their technologies are feeding in one way or the other into the most deadliest armed conflicts of the world; all the while the large majority of them revel in some form of masculinist proto-fascistoid ideology of supremacy and domination…and all while in real life, they are also paying back like zero.zero taxes to society?
In a culture where tech bros not only become filthy rich by exploiting our economic system, people, nature, their own families all the while happily lying and cheating, dissimulating and conspiring, is it really the AI agents who are not aligned? Rather, the problem are not the AI agents, but the moral depravity of their human handlers and of the system of wealth accumulation and exploitation they have created and are busy maintaining.