Pues un ensayo estupendo que analiza todo ese ecosistema de la IA que no se ve: desde los costes medioambientales y humanos a la escurridiza cuestión de los sesgos inherentes y la dificultad de una IA ética. Sin duda, uno de los ensayos del 2023, de esos imprescindibles para entender el mundo en que habitamos.
Atlas of AI manages to dig deep into the systems and cost of Artificial Intelligence without ever overcomplicating the ideas for a general reader. Using contemporary feminist philosophy, Crawford compares extraction of minerals to extraction of data to extraction of labour, and concludes that a revised understanding of technology is needed.
One of the main arguments, which is very well developed throughout, places AI research by big tech companies in line with much eugenic and colonial thought systems, highlighting how they are embedding outdated and bigoted ideas in the underlying bias of supposedly neutral systems. Similarly, the colonial patterns of extractive human labour that are used to train such systems, and that provide the materials needed to operate them, are overlooked by most companies who develop or sell these systems.
A couple of small complaints: the last couple of chapters become a little journalistic and US-centric, and while Crawford hits …
Atlas of AI manages to dig deep into the systems and cost of Artificial Intelligence without ever overcomplicating the ideas for a general reader. Using contemporary feminist philosophy, Crawford compares extraction of minerals to extraction of data to extraction of labour, and concludes that a revised understanding of technology is needed.
One of the main arguments, which is very well developed throughout, places AI research by big tech companies in line with much eugenic and colonial thought systems, highlighting how they are embedding outdated and bigoted ideas in the underlying bias of supposedly neutral systems. Similarly, the colonial patterns of extractive human labour that are used to train such systems, and that provide the materials needed to operate them, are overlooked by most companies who develop or sell these systems.
A couple of small complaints: the last couple of chapters become a little journalistic and US-centric, and while Crawford hits out at many big tech companies, she seems reluctant to criticise her own employer, Microsoft (this may be contractual, but is pronounced). But these are minor issues in a book that covers more ground than most, presenting complex and long-considered research.
