Siri Co-Creator Talks AI's Future, Urges Sustainable Practices

AI has advanced but concerns such as its potential misuse, intellectual property violations and ever-increasing data demands need to be addressed, according to Luc Julia, one of the co-creators of Siri.

Speaking at the World AI Cannes Festival, Julia, who is also Renault's chief scientific officer, said that the rapid increase in data usage and energy consumption in the AI industry is impacting both sustainability and responsible use of AI.

“The more compute, the more data we are using, the more energy we are using. And this is maybe not the right trend. We need to think about that, we need to think about our planet.”

According to Julia, if we were to take the same amount of queries that go into GPT-4 and put that into Google, there is not enough energy in the world to process that amount of data.

“Those AIs, in addition to taking a lot of energy to create them, they're also using energy while you are using them at inference time, which is unacceptable. … We need to go more frugal; we don't have a choice.”

Improving accuracy

Julia’s issues with data extended to AI accuracy, arguing that hallucinations need to be addressed. He cited a study from the Hong Kong University of Science and Technology that found the GPT 3.5 version of ChatGPT had an accuracy score of 63.41%.

Related:Meta, Hugging Face, AMD: AI's Future Must Be Open Source

“It means that 36% of the time, it was wrong,” Julia said. “If I have an intern or somebody working for me that is wrong 36% of the time, they won't work for me anymore. I don’t trust them.”

Julia said that among the key ways to improve model accuracy was through fine-tuning, taking your own data and training the model further.

“When you do (finetuning), it's going to be much more specific to whatever domain you want to constrain it, but it's going to be something that is going to become much more accurate,” he added.

Julia said his team at Renault fine-tuned models on company data, with the resulting system achieving far higher levels of accuracy. However, he stressed that users still need to check the output.

He said more users need to be educated on hallucinations so they understand not all outputs are accurate. Julia showed off his attempts to get Bard, now known as Gemini, to write him a bio, only for the chatbot to be filled with misinformation.

“The best way to do it is to just do it ourselves. Listen to the people playing with those things and the people that are creating those things. They need to be open and they need to tell you the truth about the tools.”

Credit: AI Business

Generative is not creative

Julia was stern on his view of generative AI, calling it “not creative.”

Related:Meta’s Yann LeCun Wants to Ditch Generative AI

“I'm hearing a lot of people saying it is creative, creative, creative. Creativity is on my side. Creativity is on our side, the human side,” he said. “It can produce stuff; it can generate stuff that is going to help me to push my creativity.”

The Siri co-creator said that generative AI tools such as Stable Diffusion or Adobe Firefly would not replace creatives like designers, instead helping them achieve their artistic desires.