Siri Co-creator Debunks AI Misconceptions

Luc Julia has a long list of accomplishments on his resume. Among them are co-creator of Siri, CTO of Samsung Electronics, director at Apple and how chief scientific officer at Renault. He recently sat down with AI Business to debunk widely held beliefs about AI and explained why he believes fully autonomous vehicles - despite Elon Musk's claims - will never exist.

Julia is the best selling author of "There Is No Such Thing as Artificial Intelligence" (in French and English) and most recently, "On Va Droit Dans Le Mur?" or "Are We Headed for a Wall?"

Listen to the podcast or read the edited transcript below.

AI Business: I seem to sense a skeptical theme here in your book titles about AI. Let's take your most recent book, are we headed for disaster?

Luc Julia: The book is saying in the title that for the past 120 years or so, we are using more technology, but we are also creating more and more CO2, and we are destroying the planet, basically. … We need to realize we have a big impact in our use of technology.

(For example,) a selfie is about 180 Watts if you share it. If there are one billion selfies taken every day, it is basically 180 megawatts that are taken every day just for ego. (However), we should understand that technology is also good, potentially, to help us not to go into war and to save the planet, maybe. So that's the balance.

AI Business: Your first book had this very provocative title, “There's No Such Thing as Artificial Intelligence.” Have you changed your view?

Julia: The media view of AI … four or five years ago, was it was this magical thing. What I tried to make clear in this book was to say, this is not Hollywood. This is not science fiction. This is not Robocop. This is not Her (movie about a man falling in love with an operating system).

They actually don't exist. This is not what AI is.

… There are multiple AIs. … This is the first thing that we need to understand. If I had to give a definition of one AI, it would be it is a toolbox. Inside the AI, you have multiple APIs, and each of them will be very specialized in some tasks.

I always compare one AI to a hammer. I love the hammer, it is a very good tool. It is going to drive the nail in the wall. But the very same hammer, I can use it to hit your head. And obviously, this is not good. It is not what it is meant for. But I can use it like that. This is my decision.

And it is our decision to use those APIs. We can use them in the right way, and we can use them the wrong way. … So, AIs are powerful tools, but they are only tools that we build. And it is our fault if it (ends up being like a) Terminator.

AI Business: The AI celebrity du jour is ChatGPT. Can you explain to us in realistic terms how we should think about GPT and other large language models?

Julia: One of the issues with AI in general is that we should not have called it that. The word intelligence in artificial intelligence should not have been used (because it is not really intelligent). … But the good news about generative AI is that they called it (appropriately). It is called generative AI, not creative AI. And that is a very interesting point there because it means that the creativity still lies on my lap: The creativity is in the prompt. …

This is a language model that is going to interpret what I am going to say. So if I say, ‘write a paragraph about the fact that the Earth is flat,’ … it is going to generate a paragraph that is going to say that the Earth is flat. Of course, the Earth is not flat. But it can generate a very good article or a paragraph … because on the internet somewhere, there are people that believe that the earth is flat.

We have to understand that (ChatGPT) is about 175 billion parameters. It is a huge language model. It is basically (trained on) almost the full internet.

AI Business: How transformational is ChatGPT, really? Right now, the hype cycle is just over the top.

Julia: It is crazy - over 100 million people are using it or have used it at least once. It is a very interesting tool, because it is going to help you to generate text, images or whatever that you would not be able to create yourself very easily.
(But) It is very dangerous to use that in the search engine. Microsoft was very happy to push (integration with Bing). … But I think it is a mistake, because the accuracy is not good. And unfortunately, because of this push basically by Microsoft, ... Google (will be forced to do the same). (Google had this capability) way before open AI, but they actually did not use it in their search engine because they knew it was going to be a disaster if they do. …

Scientists now are saying that about 36% of the time ChatGPT is wrong, which is a lot. So if you use that in a search engine (and are wrong that percentage of the time), it is not going to fly.

(However,) one very smart way to use ChatGPT is in (a very narrow task like writing formulas in) Excel. … This is a very good helper, but in a very, very narrow and very, very pointed way. ... But in a search engine, it will be a big, big mistake, I think.

AI Business: ChatGPT can introduce errors in writing as well. Why is the search engine worse?

Julia: (When writing with ChatGPT,) you are putting more and more of your ideas in the essay, and you are actually creating. (In search engines), it is just following whatever data is coming from anywhere, including wrong facts.

AI Business: You have worked in top management in some of the largest companies in the world: Apple, Samsung, HP and now Renault. Can you share with us what your approach to AI is?

Julia: The approach for me is always the same. I'm very cautious, first of all, so I do not claim that AI is going to solve all the issues. In companies like Samsung or Renault today, (the first question is) where should we use AI and why is it going to be beneficial enough for the company, for the people? Where are the places where it could impact the most, not negatively, but positively?

So for instance, in the factories we have robots for years spitting out a lot of data, but the data is not being used, because we are not very comfortable with it. (But then take the data and) create a model that is going to (predict) when a specific robot is going to be out. You’re going to do what we call predictive maintenance. … This is what AI is about.

AI Business: Talking about manufacturing, you have said full autonomous driving, or Level 5, will never exist. Do you still hold that view?

Julia: There are different levels in autonomous driving, from 1 to 5. (Example of Level 1 is adaptive cruise control.) ... What I am claiming is Level 5 will never exist − this is full autonomy. … This part will never exist; it's just impossible. This is perfection, basically.

We should not lie about the technology, we should always say what is possible or not. Since 2018, when Elon Musk declared that autonomous driving, Level 5, was around the corner, it was a lie.

(But other levels of autonomous driving are already here,) helping drivers to brake better, see better or to do other things better. This is thanks to AI.

Today, there are one million deaths on the road every year all over the world. By going to Level 3 and maybe Level 4, we will eradicate about 90% of those accidents. We should not say that there will not be any accidents because it would be a lie. There still will be accidents. But it will be 900,000 people that we save every year, which is huge. …

This is why Level 5 cannot be because there will be always a (situation) somewhere that the car cannot handle. … We will need human intervention because there will be cases that the cars would never have seen before. And so they will have to have some help somehow. This is why Level 5 cannot exist, because Level 5 implies that it can handle every situation.

AI Business: Waymo and Cruise have been making announcements that they are deploying these robotaxis in different cities.

Julia: They are about Level 4, but they can be Level 5 (only) in very controlled environments.

AI Business: One of the things that you are known for is your role around the creation of Siri. Can you bring us back to those exciting days?

Julia: (In the mid-1990s at SRI International,) we had the idea to create this assistant that was going to help us navigate the internet. We (with Siri co-creator Adam Cheyer) had a very clear vision that the internet was going to be huge … and so we are going to need some help to navigate this internet.

We created a little guy on the side of the browser that was helping us basically to find stuff on the internet. And we were able to, naturally, tell him what we wanted. This was before Google. … It was basically our version of the search engine (that responds to) speech, because for us the most natural way to interrogate and to interact with this assistant was voice. … And then it evolved in the next 14 years to land on the iPhone.

I’m surprised how good digital assistants are today and the great strides that NLP has come since then.