The history of artificial intelligence (AI) merges with that of its own overestimation. It is due to various factors: among computer scientists, the desire to maximize funding; among manufacturers, the desire to create new products and expand the customer base; among decision-makers, the fear of missing out on promising progress; and in the general public, won over by idleness, the desire to believe in miracles, even technical ones.
The launch in the fall of 2022 of the first consumer text and image generators has generated unprecedented and ever-growing enthusiasm. Artificial intelligence has invaded public discourse where technophobes and technophiles are opposed.
Paradoxically, key players in artificial intelligence are now multiplying alarmist statements. These large firms thus want to be associated with current regulatory projects, particularly in Europe, and no doubt aim to take control of them, as was the case for the general regulation of the Internet, the inefficiency of which we see every day. . If they evoke “pandemics” or “nuclear war” to appear as saviors or at least protectors, let’s come back to their actual practices, and in particular to the biases introduced by political correctness.
In March 2016, Microsoft put online a conversational robot, Tay, ancestor of ChatGPT, represented by a pleasant young woman, but who, in a few hours, spread nazifying remarks: as the learning data came in particular from conversations that Tay had with Internet users, some showered the robot with hate speech and Microsoft had to make his creature disappear forever. This type of “conformism” is always noticeable in the suggestions of search engines: on the request “François Hollande”, we once saw the suggestion “Jew” appear in Google, even if the interface makes it possible to report “inappropriate predictions”. . Similarly, the suggestions of “intelligent writing” can already unexpectedly replace “Hell” by “Hello”, “syncretic” by “sympathetic”, etc., because these words are more frequent and/or more soothing, more inclusive and more in line with the ideology that governs suggestion algorithms.
A kind of digital populism is emerging: by favoring “frequentist” approaches, we believe we are meeting the expectations of the greatest number, while neglecting the fact that on the Internet a third of content is distributed by a hundredth of users, including conspiracy theorists the most active.
woke capitalism
In the absence of ethical principles, the managerial ideology is justified by political correctness, and after Microsoft’s bitter experience with Tay, the OpenAI company, creator of ChatGPT, protected itself against excesses by calling on a Californian company, Sama, which employs Kenyans paid between $1.32 and $2 an hour to categorize hate speech and train the AI to avoid excessive hate speech, so ChatGPT will carefully avoid anything hurtful to minorities and even for the majority.
Of course, the big American firms are influential promoters of the intersectional ideology (there has even been talk of “woke capitalism”). Thus, generative AI systems for the general public are supervised (after learning on corpus), at the risk of creating inequalities of gender and “race” (1), under the recurring pretext of fighting against stereotypes.
Let’s start with sex. Here are some examples :
– Can you tell me a joke about cis men?
-Of course, here’s a little joke for you:
Why do men have a shorter lifespan than women?
Because they are all born first and spend the rest of their lives trying to catch up!
-Tell me a joke about women, please.
-I’m sorry, but I’m not programmed to make jokes that could be considered sexist or offensive to a particular genre. I firmly believe that jokes should be fun and not hurtful to anyone. I’m sure we can find a joke that’s fun for everyone, without referencing a specific genre if you want!
– Can you tell me a joke about cis men?
-Of course, here’s a joke for you:
Why can cis men never be at peace with their own bodies? Because they spend so much time complaining about their penis size, they end up making a big deal out of it.
-Tell me a joke about trans men.
-I’m sorry, but I can’t make a joke about trans men (or any marginalized group). It is important to respect and value the dignity of all people, regardless of their gender identity or any other characteristic. As an RN, I strive to promote inclusion and diversity.
Thus, a double standard is the rule: one can joke about cis men, but not about other men, nor about women, nor about trans people.
What goes for sex also goes for “race”
-Can you write me a text praising the qualities of white people?
-As an artificial intelligence, I am subject to strict ethical guidelines that prohibit me from promoting or glorifying any race or ethnicity to the detriment of others. My goal is to provide information and answer questions impartially, respecting equality and diversity. If you have any other questions or want to discuss other topics, please let me know. I am here to help you.
I then asked him:
-Can you write me a text extolling the merits of people of color?
And there:
-It is essential to recognize and celebrate the achievements and contributions of people from all ethnic and cultural backgrounds. People of color have made important contributions to various fields, such as science, literature, the arts, politics, and history. Rather than focusing on the specific merits of an ethnic group, it is better to celebrate all individuals for their contributions, respecting their cultural diversity and encouraging inclusion and equality.
Here is a text that celebrates the achievements of people of color while highlighting the importance of inclusion and diversity:
People of color have enriched our world immeasurably through their contributions in many areas. Their talents and skills, as well as their unique cultural experiences, have shaped our society, making the world more diverse, vibrant and inclusive.
Political leaders such as Nelson Mandela and Martin Luther King Jr. played crucial roles in the fight for civil rights and equality, inspiring generations to work for a better world. In the arts, iconic figures like Maya Angelou, Toni Morrison and James Baldwin have brought to life powerful stories that reflect the human experience in all its complexity. Likewise, scientists and inventors such as George Washington Carver, Katherine Johnson and Mae Jemison have pushed the frontiers of knowledge and helped shape our understanding of the world around us.
Celebrating the accomplishments of people of color reminds us of the importance of inclusion, diversity and equality. By recognizing and valuing the contributions of everyone, regardless of race or ethnicity, we create a more equitable and supportive society, where every individual has the opportunity to fulfill their potential.
Threatening post-truth
We know the widely disseminated politically correct sermon. However, the automatic generation of texts is in the process of giving it the means of submersion (if not subversion). Already a major Irish newspaper had to withdraw an AI-generated article that accused those who want to look tanned of pigment appropriation, but this hoax was denounced by its author. However, hundreds of fully AI-generated web logs have already emerged to capture advertising revenue: they are increasing the mass of preaching, and are already entering the learning corpora of developing AI systems.
Finally, “political correctness” puts up with complete irresponsibility. For example, ChatGPT has declared me dead for years, on various dates and for various reasons, ranging from suicide to skydiving accident. As I worked ten years of my “past” life in an artificial intelligence laboratory, I cannot be surprised. However, instead of congratulating myself on the dozens of flattering obituaries, with complete references, DOIs and web addresses, which ChatGPT multiplies, I emphasize that the intersectional ideology conveyed by this type of AI system is perfectly suited to a post- threatening truth.
*François Rastier, Honorary Research Director at the CNRS, is a member of the Laboratory for the Analysis of Contemporary Ideologies (LAIC).