‘AI agents’ could one day replace workers – 10/18/2023 – Tech

ChatGPT, a widely used chatbot, is designed to generate digital text, from poetry to academic papers and computer programs. But when AI (artificial intelligence) researchers at chip company Nvidia got access to the technology behind the chatbot, they realized it could do much more.

Within weeks, they taught her how to play Minecraft, one of the most popular video games in the world. Within the game’s digital universe, technology learned to swim, collect plants, hunt pigs, mine gold and build houses.

“She can enter the world of Minecraft, explore on her own, collect materials on her own, and get better and better at every skill,” said Linxi Fan, known as Jim, a senior researcher at Nvidia.

The project was an early sign that leading researchers in the field are transforming chatbots into a new type of autonomous system called an AI agent. These agents can do more than just talk. They can use apps, websites, and other online tools, including spreadsheets, calendars, travel websites, and more.

Over time, many researchers say AI agents could become much more sophisticated and replace office workers, automating almost any such job.

“This is a huge, potentially trillion-dollar commercial opportunity,” said Jeff Clune, a professor of computer science at the University of British Columbia (Canada), who had previously worked with the technology as a researcher at OpenAI, the startup that created ChatGPT. . “This has enormous potential for growth and enormous consequences for society.”

The Nvidia agent plays a video game. Similar agents can schedule meetings, edit files, analyze data, and build color bar charts. The idea is that these automated systems will eventually act as personal assistants capable of handling a wide range of tasks over the internet.

Today’s agents are limited and can’t exactly organize their lives. ChatGPT can search travel site Expedia for flights to New York, but you still need to book it yourself.

This technology, as researchers improve it, could make office workers and consumers more efficient. It could also change the nature of video games, providing a new wave of bots that gamers can play and talk to.

GPT-4, the technology that underpins ChatGPT, is what researchers call a large language model (LLM). It is an AI system that learns skills by analyzing huge amounts of data.

In recent months, technology has impressed hundreds of millions of people with the way it generates emails, writes speeches and improvises on almost any topic. But your most important skill may be your ability to develop computer programs.

The technology can instantly generate a program that draws a unicorn or makes digital snow fall on your notebook screen. Professional software developers can request code that they can incorporate into larger programs, including everything from social media applications to search engines.

But that’s just part of what this technology can do. It can also generate programming codes that connect to other applications and websites.

This is how Fan and other Nvidia researchers taught GPT-4 to play Minecraft. “The most important word here is code,” Fan said. “The code can take actions.”

People use apps and websites by pressing buttons, menus, and other graphic elements. AI agents use applications and websites by accessing their programming interfaces (APIs), the underlying software code that allows them to communicate with other online services.

If you ask an agent to upload a video to the internet, for example, he can generate code that calls an API offered by YouTube. “An API is just text used to communicate with a machine,” said Silen Naihin, a researcher who helps run an independent AI agent project, AutoGPT.

In theory, a chatbot can write code to access any API on the internet. But today’s chatbots are still not skilled enough to do more than simple tasks. And even if they were, allowing them to freely roam the internet would be a huge security risk. Therefore, companies are starting slowly.

A few months after OpenAI launched ChatGPT, the company quietly rolled out a way for the chatbot to do more than generate text. After installing several plugins — extensions that increase what the chatbot can do — you could ask it to search travel sites like Expedia for available flights, get a map of your hometown from Google Earth, or even transform a spreadsheet detailing your annual expenses in a bar graph.

Equipped with a plugin called “code interpreter” [interpretador de código], ChatGPT could not only write the program but also run it. This allowed the technology to instantly perform tasks it previously couldn’t, including editing spreadsheets and turning still images into videos. Google, Microsoft and other companies are exploring similar technologies.

“These are projects where we’re essentially imagining AIs working with other AIs for you,” said Ashley Llorens, director of Microsoft Research.

Independent projects like AutoGPT are trying to take this kind of idea even further. The proposal is to give the system goals such as “create a company” or “make money”. It will then look for ways to achieve this goal by asking itself questions and connecting to other internet services.

Today, this doesn’t work so well. Systems like AutoGPT tend to get stuck in endless loops. But researchers like Nvidia’s Linxi Fan are constantly improving this type of technology in an attempt to make it more useful and reliable.

Other researchers are building a new type of AI agent designed to use software tools. In the summer of 2022, Jeff Clune was among a team of OpenAI researchers who built an agent that could use computer software in the same way a person would — mouse click by mouse click, key by key.

Clune and his colleagues fed the system hours of online videos showing people playing Minecraft. By analyzing the way people used their mouse and keyboard to navigate Minecraft’s digital universe, the system learned to play the game on its own.

Other companies, including a startup called Adept, are building similar agents that use sites like Wikipedia, Redfin and Craigslist and work-oriented apps from companies like Salesforce.

Clune argues that this type of agent will eventually allow artificial intelligence to use a much wider range of software applications and websites. He said everyone would have access to a digital assistant that could potentially do almost anything on the internet. This could make life easier — but it could also replace countless jobs.

“If AI can do everything we can do, it doesn’t just replace boring tasks,” he said. “It replaces all tasks.”