From conception to implementation, AI isn’t neutral

AI doesn’t just pose the technological challenge of creating ever more sophisticated models to do our bidding. There are profound political choices, too, that shouldn’t be left to Silicon Valley.

In November 2023 OpenAI, the company best known for ChatGPT, its flagship large-language model, was the site of a bizarre boardroom battle. The board, led by Ilya Sutskever, a computer scientist and OpenAI co-founder, ousted CEO Sam Altman. Altman, who is also a computer scientist and co-founder of the company, eventually regained his position, but the episode revealed an internal divide between two ideologies that may appear opposed on the surface but turn out to be not that dissimilar: effective altruism and effective accelerationism. Advocates of the former at OpenAI tried – and failed – to remove the latter’s leaders, fearing they might bring about the end of humanity.

Effective altruism, which took off in the US in the 2000s, aims to maximise the use of resources for the common good. Its proponents believe their superior intellectual, financial and technical acumen makes them uniquely well equipped to prioritise and solve humanity’s biggest challenges, including risks such as pandemics, nuclear war and the emergence of artificial general intelligence (AGI), sometimes called ‘the singularity’. AGI, a form of artificial intelligence possessing consciousness and created using vast quantities of data, could either lead humanity to an era of universal prosperity or cause its extinction. But definitions of what actually counts as AGI are so vague that some consider it to have already been achieved, while others envisage it emerging some time in the next half-century.

Effective accelerationism is more radical; its proponents advocate unrestricted technological development to bring superhuman AGI into existence as soon as possible in order to take humanity to a higher evolutionary stage and thereby escape its current perils. To do so, they call for the removal of all regulatory and ethical guardrails and dismiss concerns about intellectual property and data privacy. All that matters is speed.

This unashamedly techno-libertarian stance favours (...)

Full article: 1 964 words.

Victor Chaix, Auguste Lehuger & Zako Sapey-Triomphe

Victor Chaix is a PhD candidate in digital humanities; Auguste Lehuger is an AI research engineer; Zako Sapey-Triomphe is an engineer. This article synthesises a note published in October on the X-Alternative thinktank’s website.

(1) Marcello Vitali-Rosati, Éloge du bug: Être libre à l’époque du numérique (In praise of the bug: freedom in the digital age), Zones, Paris, 2024.

(6) Victor Petit, ‘Vocabulaire d’Ars Industrialis’ (The vocabulary of Ars Industrialis), in Bernard Stiegler, Pharmacologie du Front national, Flammarion, Paris, 2013.

(8) Christophe Denis, ‘Esquisses philosophiques autour de la compréhension de phénomènes complexes avec des outils de prédiction basés sur de l’apprentissage machine’ (Philosophical sketches around the comprehension of complex phenomena with prediction tools based on machine learning), French Conference on Knowledge Extraction and Management, EXPLAIN’AI workshop, Blois, January 2022.

(9) aialchemy.media.mit.edu/.