Many questions arise about artificial intelligence, particularly about ChatGPT, and their impact on work and employment. “We are going to be in a world where the amount of content that will be produced will explode. And these algorithms are not very good at sorting content“, says Friday June 14 Thierry Rayna, pprofessor of the École Polytechnique, researcher at the CNRS i3 laboratory and economist by training. For this researcher, AI is not necessarily synonymous with job destruction, on the contrary, “AI will change our lives because we will work more and more“.
The researcher points to the multiplication of content, deep fakes and the need to verify this content: “We are going to work more and more, since we are going to have more and more content, more and more news, probably, a lot of fake news, we have seen deep fakes, a lot more things to filter“.
franceinfo: What do we talk about when we talk about artificial intelligence? There is generative AI, there is discriminative AI. What is the difference ?
Thierry Rayna : I think that’s where the problem comes from. That is to say that AI has become a real catch-all and there are AIs, but of all kinds. There are some that are based on rules, there are some that are based on small objects that we will make intelligent collectively. And then there are AIs that we call based on machine learning, and there, whether it is generative AI or discrimination AI, they work in the same way. Roughly speaking, we will give information, an algorithm and try to make it actually guess the missing information. So, for example, when we train a ChatGPT, we take text, we remove words and we ask the machine to learn the words. So traditional AI, what we’re going to try to do is take large data sets and try to make sense of them. So what is the trend? What is this text about? Generative AI, in fact, the idea was to reverse that and say since ultimately, AI is not necessarily that good at sorting through data, we are going to make it create additional data, so we will create text, we will create images.
What new uses can emerge when you are a company with generative AI?
Most of the generative AI use cases we see today are recycled traditional AI use cases. That is to say, we always try to take data and try to understand what is in the data, whereas the real use cases, generative AI, are when we are going to create things. And that is not to say that there are no use cases for companies, but it is not at all obvious that it is necessarily very interesting for them, to the extent that in fact, the Companies, very often, already have a lot of data, they have access to a lot of things and what they’re trying to do is make sense of that data. So the idea that we can create an image or that we can create text works, but it has its limits. The problem is that what we hear relatively little about is that there is a real asymmetry in fact, in this AI based on machine learning. These AI algorithms are very good at creating content, but very bad at selecting content.
“Even for something as basic as a chatbot, creating text for the chatbot is good, but the chatbot is generally not very good at understanding the problem.”
Thierry Raynaon franceinfo
And so that’s why there are a lot of use cases that we’re seeing right now, that we’ll have to remember, in ten years we’ll be laughing a lot in fact.
We hear a lot about AI going to destroy jobs, AI going to replace my job. How should we look at this?
There are many studies that have been done. The problem is that most of these studies only looked at the creativity aspect. That is to say, roughly speaking, we are going to take someone’s work and say this person writes so many emails per day and therefore, thanks to generative AI, instead of spending ten minutes to create an email, he presses a button email is written. So that means they save time. And that means that instead of having ten people like this person, we take half of them. But what we fail to understand is that in fact, yes, for writing emails, it’s very good, but for sorting emails, it’s very bad. And so, we are going to be in a world where the quantity of content that will be produced will explode. And these algorithms are not very good at sorting content.
So there will always be humans?
Even more humans will be needed. Because actually, what’s happening is that right now, everyone’s getting too many emails, but actually we’re not getting as many as tomorrow. Because writing an email takes time, but already today, you press a button, the text is generated. But that doesn’t change the fact that humans are still needed to sort emails because an algorithm is simply capable of identifying the average email. But the urgent email from your boss for example, that will typically go by the wayside.
We often repeat that these new developments are a break with algorithms like that of OpenAI, but sometimes, we discover the moon, the GAFAMs themselves had algorithms. What is changing?
This is very interesting because it’s happened for just about every other digital technology. Indeed, contrary to what we hear in what OpenAI did with ChatGPT, technologically there is nothing new, in terms of data and quantity of data, there is nothing new. We are really on a fairly continuous progression, the famous Moore’s law, therefore completely predictable. And even generative AI is not something new. The only thing that is really new in fact is that this company, which did not have the best algorithms, which actually took an old version of its algorithm, made it available to everyone unlimitedly. . As always, as was the case with the web, with MP3, what actually makes the breakthrough is that instead of companies trying to guess what people can do with these technologies, people take hold of these technologies and start doing absolutely anything, like writing code with ChatGPT or making collaborative encyclopedias. Basically, these are ideas that honestly seem completely stupid and yet work. So that’s where the breakup comes from. It is not that we have made technological progress, it is in fact the openness and massive access to these technologies which means that everyone can use them to do what they want, including things that don’t make sense for companies, which ultimately create uses and disruption.
For example, what does artificial intelligence change for a factory?
Generative AI, absolutely nothing, since a factory, what we want is to control processes, to have exact answers, whereas here, we will try to add data. So maybe that could change, for example if the idea was that we needed in the factory to personalize a product, that could make it possible to personalize the product more quickly, change the colors, or make images , etc. But once again, we must not deceive ourselves, these algorithms work on an average, they are statistics, it is the average. So in fact, the only innovation we can have is the average. So if you want medium customizations, it works very well.
“If we want a real work of art, we’re going to need people who are well-wired and capable of doing the work.”
Thierry Raynaon franceinfo
There will always be someone behind a job, AI will not disrupt our lives, but perhaps adapt it?
What is quite clear is that AI will disrupt our lives because we will work more and more. We are going to work more and more, since we are going to have more and more content, more and more news, probably, a lot of fake news, we have seen deep fakes, a lot more things to filter. And we’re also going to have a whole bunch of opportunities. Everyone can improvise as a musician, everyone can improvise as an artist, everyone can improvise as a writer. Again, you won’t become a fantastic musician or an amazing artist, you will be an average artist, but that’s not bad. It’s clear that we’re going to work a lot more. On the other hand, what is also certain is that this generative AI allows almost everyone to become average. This means that there is a lot of work that companies previously did alone, that now individuals will be able to do.
