Luc Julia is the co-creator of Apple’s voice assistant Siri and one of the most influential engineers in the field of human-centered artificial intelligence.
He previously served as Chief Technology Officer at Samsung Electronics and held leading roles at HP, Apple, and several Silicon Valley startups.
Luc Julia is known for challenging the hype around AI and promoting a pragmatic, engineering-based understanding of what intelligent systems can and cannot do.
His bestselling book “There Is No Such Thing as Artificial Intelligence” argues for a clearer, more realistic view of computational systems and their limitations.
He has advised governments and international organizations on responsible innovation, data-driven policymaking, and the future of digital ecosystems.
As a renowned public speaker, he is celebrated for making complex technological concepts accessible and transparent.
Today, Luc Julia continues to develop applied AI solutions while advocating for a more human-aligned and accountable technological future.
Article Zero:
Our vision for the first issue of Article Zero is to combine expert voices like yours with real stories and practical perspectives on AI. And the main idea is to go beyond the hype and to build a place where readers can find both critical and inspiring views on AI.
You worked for many years at Apple on Siri as one of its main creators and co-founders. So, my first question would be, when you look back, what was the biggest challenge when creating Siri?
Luc Julia:
First of all, there are several steps, that were taken to create Siri.
It started in the 90s, in 1997, 1998. It was the creation of a voice assistant. It was a long time ago.
The speech recognition wasn’t very good back at the time, but what we did was basically give a persona to the voice assistant. 10 years later, the company Siri was created and another three years later, the company Siri was acquired by Apple.
Therefore, when the company was acquired, I came back to Siri to take the direction of the group. But one of the main challenges was that the technology wasn’t very good, as it was from the 90s, 80s.
At that time, it was still machine learning. And regarding the recognition, there were two steps in speech recognition. On the one hand, there was the actual speech recognition itself, and on the other hand there was the natural language understanding.
Consequently, you had to first understand the words and only then could you fathom their meaning. But in any case, the recognition system was very good as it is difficult to make sense of something, when you only have 80% of the words.
So, the main challenge was to have better speech recognition in order to have a better natural language understanding. Therefore, it was what we worked on at Stanford Research back in the 90s. Afterwards the company Siri worked on enabling it on smartphones.
The beginnings were made on an Android phone. But later on, after it was acquired by Apple, the goal was to put it on iOS.
When Siri went in the labs at Stanford Research it wasn’t made for the general public, thus there wasn’t any problem with scale. But for Siri to go public it needed to have servers.
Initially they only had 118,000 users. With this sum, you needed to have some server space, because all data of Siri is on the cloud. But of course, after moving it to Apple it became a lot more than 118,000 people.
It went up to 165 million users for which you need to have scale servers in order to implement and deploy Siri. That was one of the main challenges before and during the implementation of the product.
Consequently, there were issues both at the research and the production level. Ultimately, it proved successful. It was already quite stable at launch and now with over 500 million people using it it’s certain that the initial assessment remains well-justified.
Article Zero:
It leads me to the next question – in 2019 you published the book “AI Does Not Exist”. It became very well-known and really made many people in the world think differently about AI. Therefore, my question is: Can you explain the main topic for our readers with your voice?
Luc Julia:
Of course. The title in English is “There is no such thing as AI”, whereas in French it is called “AI doesn’t exist”.
The thesis of the book is that the version of AI, which people often dream about doesn’t exist. It is what I call the Hollywood AI.
This is science fiction. Unfortunately, a lot of people believe that this science fiction version is real when in reality it is not.
In reality we have to set the term AI as plural. Because it’s simply not only one AI, it’s multiple AIs. And at the end of the day, those AIs are only tools. They are tools that we control and build.
Neither do they have questions, nor do they create or do anything without us. Therefore, we need to be very careful when talking about all of these autonomous things, that AI is said to be able to do by themselves in the future.
That’s something I wanted to clarify to be sure that people stop dreaming about a version of AI that simply doesn’t exist.
AI is a toolbox with a lot of different tools. And each one of those tools, by the definition of a tool, is better than me.
Because otherwise it’s not a tool and it’s not useful. So, each tool for the very specific domain for which it was developed is better than me.
But as a whole, AI is not more intelligent than I am. As we don’t even know what it means to be intelligent, I want to try to develop something that is going to be more intelligent than I am.
I want tools that are going to be better than me in every single domain that I’m planning to use the tool for. This is what we – the AI community – have been building for the past 70 years.
It, as always, started off with excitement of how great everything is going to be, the AGI – the artificial intelligence.
Only later we realized that our AGI is complicated. A usage-drop and window of AI follows. But that doesn’t have to mean anything.
We found some applications and use cases, where the AI that is being developed is useful in those different domains. It is what we are working on right now.
We are starting to understand that LLMs are not going to be able to do everything right as there will certainly be times in which they are going to be wrong. But regarding specific areas, domains and use cases they are most definitely going to be very useful. That is what the book is about.
It’s time to stop dreaming and start working, because those tools need us.
To further understand the matter, compare AI with a hammer. A hammer is an excellent tool for driving a nail into a wall, but it is far less effective when one is attempting to insert a screw or even dangerous when attempting to use it on one’s head.
That’s why regulation is much needed as it helps to prevent bad choices and their possible consequences.
Article Zero:
You briefly addressed the term “Hollywood AI”. Do you think the media is creating too much hype about AI? And how does it affect society?
Luc Julia:
Of course, the media is always pushing for something incredible. Consequently, if it appears on the front cover, it has to be attention-grabbing in order to captivate and keep the readers engaged.
But it is rather unfortunate, as it creates expectations which later on are invalidated as the truth gets revealed. This often times leads to people being disappointed.
As a result, it could lead to a winter of AI as people who are disappointed or afraid of using it could stop interacting and funding it which in turn leads to AI’s disappearance. Still, the current phenomenon is unlikely to disappear anytime soon, given the extensive attention it is receiving.
As discussed earlier, we found some use cases that are quite interesting. Nonetheless, it is important to avoid overemphasizing the subject.
Another particular reason for the media to put those kinds of titles on the front page is fear. One of the corporations that is afraid of AI are journalists as they fear getting replaced by AI.
But AI is not going to replace people, it is going to serve as a tool in order to help people achieve their tasks. As I said earlier, AI can be used in both good and bad ways. But when used for the good, incredible things can be achieved.
Luc Julia:
You know, I live in Silicon Valley, right? And in the Silicon Valley, there is a wording which says: The US invents, the Chinese copy, and the Europeans regulate.
And unfortunately, it’s quite true as we spend more time in Europe – especially in Brussels – doing the regulation.
But let’s take a look at China first. There regulation doesn’t really exist, whereas in the US, they wait until the issue happens and only afterwards it is decided what can be done to prevent it from happening again. In Europe the AI Act the regulations get done beforehand so that the issues can be prevented.
But the problem with that is that the innovation of some technologies gets stopped because of the “forbidden bag” we put it in. The regulation should happen at the application level, not the technology level.
In order to do that, you’d have to be very agile as there is constant reaction – of whether it is good or bad – to the new application necessary.
In my opinion, if China and the US are ahead, it is primarily because of regulation. Regarding the fundamental technology, they are not as advanced.
The number of Foreigners from Old Europe, who are in Silicon Valley and contribute to AI with their background are much more mathematical than the US. Consequently, Europe is better in mathematics than the rest of the world which is helpful, because AI consists entirely of mathematics.
There are more and more initiatives in France that push the creation of start-ups. This has led to over 1.000 start-ups that are using and researching AI.
The Chinese are very great as well. They used to copy like the Japanese and the Koreans in the 70s, but now they have learned a lot from it. They are all over the world, in universities you see Chinese people everywhere. They are very good with AI and don’t have any issues regarding its regulation.
Article Zero:
And perhaps one final question: based on your experience at Renault, what do you believe will most astonish us about automobiles over the next decade?
Luc Julia:
Over the next decade? That is a difficult question as when you look back at history there are some waves of AI. Therefore, there is always something new to discover.
In my opinion the general AI is here to stay on very specific domains. But we are definitely going to realize that the current AI, as it is, has a negative impact on the environment as it uses way too many resources such as electricity and water. In addition to this it produces CO2. Therefore, in Europe, as we are a bit more sensitive in regard of this matter the next generation of AI is going to be much more frugal.
The race of being bigger and bigger is going to be turned to being more frugal as it is important to save the planet.
Elon Musk wants us to go to Mars, but it’s going to take some time. I’m not sure if I want to live on Mars. The point is that technology won’t fix technology, right?
Consequently, we need to find some new technologies that are going to be better. Understanding what we did before so we can achieve the same result, but with less danger of damaging our surroundings.
