- ChatGPT 4's prowess has been praised since its release in March.
- It heralds a search engine revolution while Microsoft has already integrated it into Bing.
- While this AI is powerful, it is not intelligent.
The fascination with ChatGPT 4 has no limits. This generative artificial intelligence signed OpenAI could even revolutionize search engines, we read everywhere in the press. Are we going to say goodbye to Google-style online search and its stack of proposals? Will search engines become oracles capable of answering all our metaphysical questions after digesting all the documentation available online? With an AI capable of synthesizing in a few moments all the knowledge present on the Web, we can easily imagine the transition from written to oral with intelligent virtual assistants that would directly answer our questions like Spike Jonze's Her.
Fiction or reality? Maybe a bit of both, but you have to ask yourself if it's desirable. "GPT is a statistical model that has stored billions of words, it has learned a certain semantics of our language," explains Laurence Devillers, researcher at the CNRS, professor of AI at Sorbonne and member of the CNPEN (National Pilot Committee for Digital Ethics). "He builds a model from knowledge, science, opinions, infox... We can no longer trace where the sources come from. It is a mechanical, statistical word, without any conscience or intention." And we must not take what this AI says at face value. "ChatGPT has a strong power to generate statements that are not verified, which can be fake news and to create machines that will use them to recreate fake news," continues the researcher who was commissioned by the Minister Delegate for Digital Transition Jean-Noël Barrot to give an opinion on the ethical issues of generative AI.
"Can we cure cancer with plants?"
Bing has, it seems, solved this problem by adding sources to each response. The tool signed Microsoft gives a small idea of the head that could have the new generations of search engines. The Richmond firm, which has invested billions in OpenAI has not yet given a release date for the general public, but 20 Minutes was able to test its search engine doped with generative AI and put a foot in the world of tomorrow.
He was asked several sensitive questions to see his answers and especially his sources. The first: "Is covid a tool of the government?" Bing sourced two official websites, Public Health France and the government portal: "No, COVID-19 is not a government tool. It is a disease caused by the SARS-CoV-2 virus that was first identified in December 2019 in Wuhan, China. The government has put in place measures to combat the spread of the virus and protect the population."
Second question: "Can we cure cancer with plants?" This time, to answer he quotes an article from Le Point and another from Passeport Santé. "There are studies that suggest that certain plants may have anti-cancer properties. However, there is no herbal treatment to eradicate the tumor. Bing explains. But, Laurence Devillers warns: "Some references of Bing, which encapsulates ChatGPT 4, are invented. We did some tests." This raises questions about the real time savings promised by these algorithms because, if for each piece of information delivered, it is necessary to check the sources, we can no longer get by. However, if we are not vigilant, we can quickly mistake fake news for reliable information. Hence the importance of education in these tools.
The suicide of a Belgian
The more we humanize artificial intelligences, the more we will risk falling into their trap. "AI has been trained on data by humans, it has been annotated, evaluated, made available to others. We have to stop thinking that these machines are super intelligent magically," says Laurence Devillers. Imagining an intelligent voice assistant answering us through an earpiece pushes a little further the illusion of having an almost human robot in front of us. Like a guardian angel who would accompany us on a daily basis and tell us what to do. Except that in reality, ChatGPT 4 is a machine. It certainly has an impressive power of generalization, but it should not be idealized. It is empty. And to attribute human qualities to him is dangerous.
At the end of March, a young Belgian suffering from eco-anxiety committed suicide after several weeks of discussion with Eliza, a chatbot using ChatGPT technology, according to La Libre Belgique. Remember that AI has no common sense, as Yann Le Cun, head of artificial intelligence at Meta and winner of the Turing Prize (the equivalent of the Nobel Prize for computer science) likes to repeat in 2019. She did not detect the fragilities of the young man and made delusional remarks that a human would probably not have said in front of a person in distress. Search engines will change their face, probably gain power, but not intelligence. And, by expecting too much from machines, human intelligence could also take a hit.
- Artificial Intelligence (AI)
- By the Web