This isn’t any of them: this is a production context with real data of real users and the privacy and security of these at stake. This is one of those rare cases where they’re true, in fact.

But even in scientific or recreational contexts, it’s still bad by itself. First of all, it deskills you. Second of all, it still sums to the enviromental and energetic effects of AI. Third of all, “AI” is NOT reliable at all, as it’s inherently chaotic, stochastic and indeterminist. Fourth of all, “AI” is tainted and blood-stained chain of origin, encompassing the datasets used for training and the methods by which they are acquired, the materials and processes employed to manufacture and maintain the hardware and software involved, as well as the procedures used to categorize and clean the training data (that’s underpaid exploitative labor done by people of the Global South). Sixth of all, it perpetuates and amplifies biases by presenting itself as an objective, impartial, and detached entity to be trusted blindly—a dynamic that makes fake news more destructive than ever. Seventh, it is built upon the theft of works by millions of artists (of all kinds—not just painters and illustrators), scientists, philosophers, thinkers, and countless others, all without their consent, proper attribution, or fair compensation, thereby amplifying the phenomenon of enclosure. Eighth, it facilitates license laundering. Ninth, it has inextricable ties to both fascism and authoritarian tankie movements, as well as to war, numerous massacres, unjust imprisonments, and the intelligence and psychological or information warfare operations of governments worldwide.

Elaboration and sources are here.

Edit: Forgot to add that the “AI” that everyone talks about is not the actual AI that scientists use. They don’t use ChatGPT, StableDiffusion or anything like that to do research. They use purpose-specific machine learning models, created by them for their particular context and use case.