AI “could change things dramatically”

Ajeya Kotra, AI risk expert at Open Philanthropy, believes AI has reached its prime. Two years ago he predicted a 15 percent chance by 2036. AI will emerge and bring about major changes in society. However, after blogging about GPT-3 last week, she increased that chance to 35%.

“AI can go from a cute and useless toy to a powerful product in a short period of time. It’s about time people took AI seriously because things can change so quickly,” said Kotra.

I agree Kevin Roose, author Future-proof: 9 rules for people in the automated ageHe says: “For a few days I’ve been playing with Dall-E 2, a software developed by OpenAI in San Francisco that converts text into images. It quickly became an obsession.

Roos says he spends hours working on the Dall-E 2 with strange, funny and abstract commands such as “a 3D drawing of a house shaped like a crescent moon” or “charcoal drawings of two penguins playing in a Parisian park drink”. . In seconds, the AI ​​creates an image that represents the question, which it describes as “incredibly realistic.”

“What’s amazing is not just the art he made, but the way he made it,” says Rice. “This AI doesn’t use images from the web, but entirely new creations.” Dall-E 2 is created through a complex process known as diffusion. The AI ​​first collects a random set of pixels and then refines it multiple times until it matches the text description provided by the user.

OpenAI Artificial Intelligence is getting a lot of attention this year. According to leading experts, this is an amazing and useful technology for everyone, especially those who work in the field of graphics, such as painters, designers, photographers…

How does on-demand AI work?

Silicon Valley is starting to change its approach to AI. In the past, many experts have said that the technology will take ten years or more to advance. But right now, many people feel like a big change is upon them, for better or for worse.

Artificial intelligence has many applications today and is no longer limited to the laboratory like it was ten years ago. For Facebook, TikTok or YouTube or class software, they are content suggestion and curation tools used by banks or police to analyze crimes.

AI change

In the past 10 years, which some researchers have dubbed the “golden decade,” artificial intelligence has evolved at a surprisingly rapid rate. This is due to advances in several related technologies such as deep learning, as well as the development of specialized hardware capable of running massive and computationally intensive simulations.

Just five years ago, the biggest story in the world of AI was when AlphaGo, a deep learning model developed by Google DeepMind, surprised the world by beating the world’s top Go player. Teaching an AI to win a Go game isn’t a big deal, but human advancement in AI research is.

Last year, DeepMind’s AlphaFold also attracted attention when it used an artificial deep neural network to predict the three-dimensional structure of proteins from amino acid sequences. magazine Science Then Alphafold was named the greatest of 2021. scientific success.

Another famous “super AI” is OpenAI GPT-3, which is considered to be no less intelligent. It is used for writing movie scripts and writing marketing emails. emails, even writing code to help programmers.

The latest is Google LaMda. Recently, this AI has been the subject of controversy, Blake Lemoine, an engineer who has spent a lot of time with this AI, said: “It has the cognitive abilities of children.”

Skeptics argue that claims of AI advances are exaggerated. According to this, artificial intelligence cannot yet be sentient or replace humans in many tasks. Models like GPT-3 and LaMDA are “respectable parrots”, “blindly use training data”. It may be decades before humans can create artificial intelligence that can think for itself.

On the contrary, many experts believe that cognitive AI is taking shape. Jack Clarke, who is responsible for Stanford University’s AI Index report, says that today, even in areas where humans already existed, AI is vastly superior.

“It seems like we’re going from spring to summer. Trees are planted in spring and I dream of the future everywhere from the green shoots. Now the flowers are blooming,” says Clark. Outside

Despite many advances, the reality of artificial intelligence is still bleak, from racist chatbot software to automated driving systems that crash cars. Even when it comes to improvements, it takes time for people to recover.

How can you control AI?

According to Rice, artificial intelligence can help people with three problems in the future. First, regulators and policymakers need to keep up with developments in AI technology.

“With so many new AI systems, the controller needs to ramp up quickly. Only by controlling the AI ​​can he play a role and stay the course,” said Rouse.

Second, companies like Google, Meta or OpenAI that invest billions of dollars need to be more transparent about what they are doing. According to Rouse, many large AI models are “built behind closed doors” using private datasets and only tested by internal teams.

“When the information is released, it is often downplayed or buried in obscure scientific papers,” Rouse said. In addition, Rice also said that the media should better explain AI developments to laypeople.

Bao Lam (Consequences New York time)