It's a bad year for the tech industry.
Not only has it lost value on the stock market and laid off thousands of workers, but it has broken most of its promises.
Neither clean energy nor autonomous cars nor metaverse, not to mention the disenchantment of crypto, web3 or 5G.
With one exception: generative models like ChatGPT have exceeded our expectations and fired our imaginations.
It's going to be the big year for artificial intelligence (AI).
OpenAI is the star.
It aims to raise more than $29 billion in the coming days, including $10 billion from Microsoft that will change the way we search the Web. However, the real innovation from LaMDA (Google), Galactica (Meta), Dall-E2, and ChatGPT (OpenAI) is the user interface.
Natural language is such a powerful interface that it oversells its potential to replace and destroy us.
If they do, it won't be because of the singularity.
ChatGPT's magic is in predicting with aplomb the most persuasive way to put one word before another.
It is the evolutionary and glorified descendant of the search engine
autocomplete
.
And even for that they need our help.
As Australian academic Kate Crawford puts it in her
Atlas of an Artificial Intelligence,
"they are not autonomous, rational, or able to discern anything without extensive and intensive training."
ChatGPT relies on the labor of hundreds of unskilled workers who are paid less than $2 an hour for exposure to the most disturbing content on the Net.
GPT-3 learned to master colloquial language by assimilating hundreds of billions of pieces of Internet content, including the kind of forums that don't always represent the best of the human race.
Preventing him from saying atrocities or repeating the propaganda of supremacists, anti-vaxxers, QAnon fanatics and other toxic groups that flood the Internet with disinformation campaigns requires a good purge.
A process that consists of searching and labeling by hand those contents that you do not want to repeat, including child sexual abuse, bestiality, murder, suicide, torture, self-mutilation or incest.
To do so, OpenAI subcontracts companies in Kenya, Uganda, or India that also work for Google, Meta, and Microsoft.
For a salary that ranges between 1.22 and 1.85 euros an hour, thousands of unskilled workers examine the darkest corners of human nature.
They are exposed for more than eight hours a day, in countries without labor rights that guarantee a minimum of training or psychological assistance.
It is the paradox of artificial intelligence: it is consuming more and more humans.
The model is called Potemkin or
fauxtomatic AI,
a term coined by essayist Astra Taylor to describe the illusion of automatism produced by thousands of hidden people, the secret goblins of the AI workshop.
They are the slaves of the 21st century, condemned to row in the dark of the galleys so that the ship moves as if by magic, promising us freedom.
Subscribe to continue reading
Read without limits
Keep reading
I'm already a subscriber