Artificial intelligence has returned to the headlines in recent months
with the public curiosity that ChatGPT, DALL-E or Midjournery havearoused. Between wonder, fears, calls for regulations and capitalistprofits, what is the reality of progress around AI? And what are theissues behind these debates? ---- The history of scientific productionaround artificial intelligence (AI) has its roots in the middle of the20th century. Until very recently, the successes in the field of AIknown to the general public remained limited to specific "entertainment"tasks : chess at the end of the 1990s, the game of Go in 2016 or pokerin 2017. .In the fall of 2022, OpenAI's ChatGPT came to shake up the news byproviding the general public with seemingly coherent text answersprovided to questions (in the form of prompts), containing parameterssuch as the shape of the expected answer ; a summary or a dissertationfor example. The stunning result in a field as complex as languageraises legitimate questions about the purposes of using this tool, whichcan self-improve, and the transformations that its multiple uses implyin the world.ChatGPT is a so-called "generative" AI, which can, thanks to itsalgorithms, produce content, enrich it by specifying it and making itmore complex in an "autonomous" way. Its construction uses unsupervisedlearning which is based on statistical models where during the machinelearning phase, the submitted data is not labeled and the result is notknown. Simple example of this method : submit photos of animals to an AIwith the aim of having a classification by species. The algorithm willfind links between different photos and will be able to classify themcorrectly by group of similar animals if the model is well constructedand the data sufficient.The various improved versions of ChatGPT released in recent months arepushing through their performance to sound the alarm within severalorganizations, professions and scientists around the potential damageand questions imposed by such a tool very quickly : replacement of thehuman, including in so-called intellectual professions ; what futureplace for education systems in the acquisition of knowledge?; how tofight against the production of false and invented content fromscratch?, etc.AI is scary, above all it raises many questions such as: What futureplace for education systems in the acquisition of knowledge?; how tofight against the production of false and invented content from scratch?At the dawn of a new world?However, the overall pattern in the debates and the answers given tothese questions remain very mediocre in the face of the real challengesposed by these AIs. On the one hand, the diagram in question draws onthe caricature and the imaginary of the fears maintained by sciencefiction. Looking at the front pages of the newspapers, there is no needto be convinced that the image of a Terminator or Matrix is much moremeaningful to the collective unconscious than the methodological nuancesevoking the reality of things, or the possibility of having a holdabove. To project the debate so far and in a vague way, sits thedispossession of a subject that will impact our lives, to leave it inthe hands of capitalists, politicians and experts who believe theirhallucinations.On the other, enthusiastic and messianic responses that marvel at theimminent arrival of a super intelligence that will usher in a new era[1]. In The Myth of Singularity [2], J.-G. Ganascia demonstrates thatthis fantasy maintained around the singularity, this moment when a superintelligence will far exceed what human intelligence can conceive, is afiction that feeds a very lucrative fundraiser, very real that one.One of the twin words of AI in recent months is that of regulation. Wecan cite the call [3]"to pause" and to regulate AI, signed by scientistsand businessmen, some of whom, like Elon Musk, do not bother withnonsense. States, for their part, say that innovation should not beprevented and that it is up to companies to manage ethical issues.Result : meaningless regulations.The tyranny of convenienceIn a world where the race for AI is of strategic importance for states,binding legislation is not on the agenda [4], giving a green light tothe politics of the established fact which makes it possible to makemassive profits. The convenience [5]of a product to the general publicis one of the most important aspects to debate about generative AI : theease of use, the fields of application and the quality of the response,will anchor these tools in human practices in the next few years, do wereally want it and for what result?Simple rejection would be an announced failure of the collective action.Just think of the positions of twenty years ago in militant circlesvis-à-vis mobile phones and their current place in daily use, despite amuch higher level of intrusion into private lives. . The list ofpossible damages can be made without waiting to see them happen. Workersin Kenya have taken industrial action and legal action against thedigital operators [6]who exploit them [7]to reinforce the "relevance" ofAI (and this calls into question the purely "unsupervised" nature oflearning).They and they say that the unsustainability of the images that they haveto swallow until they are sickened every day seriously attacks theirhealth, recalling that without their work, the announcements with greatfanfare by villainous CEOs are nothing. The actions of these workers[8]show one of the paths to follow to put sand buckets in the cogs ofthese AIs, in that it makes it possible to regain the power to actcollectively, on an announced dispossession.The challenge is to counter in the field of struggles and ideas, thisuniformity which denies the complexity of the living and the drying upof thought that it implies. Because the real world is not just anobstacle to be overcome or a commodity to be made profitable.Marouane Taharouri (UCL Naoned)To validate[1]"Yann Le Cun, director at Meta : "The very idea of wanting to slowdown research on AI is akin to a new obscurantism"", Le Monde, April 28,2023.[2]"Ganascia, The myth of singularity", Alternative libertarian, June 2020.[3]"Pause giant AI experiments: an open letter", futureoflife.org[4]"Macron on AI at the Vivatech trade fair: "Regulating withoutinnovating would be like trimming hedges that we don't have"",Liberation, June 14, 2023.[5]"The tyranny of convenience", Lemonde.fr, March 28, 2023.[6]"In Kenya, three complaints against Meta lift the veil on "the darkside of social networks"", Le Monde, June 15, 2023.[7]"OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPTLess Toxic", Time.com January 18, 2023.[8]"In Kenya, artificial intelligence subcontractors create the firstAfrican union of content moderators", Le Monde, May 2, 2023https://www.unioncommunistelibertaire.org/?Chat-GPT-Intelligence-artificielle-absurdite-humaine_________________________________________A - I N F O S N E W S S E R V I C EBy, For, and About AnarchistsSend news reports to A-infos-en mailing listA-infos-en@ainfos.ca
Geen opmerkingen:
Een reactie posten