Two years ago, OpenAI started the chatbot craze with the release of ChatGPT. Now he hopes to spark interest in the new wave of AI technology.
On Thursday, OpenAI unveiled a tool called Operator that can go online and perform tasks autonomously, such as grocery shopping or restaurant reservations.
“It can navigate websites and take actions on websites, just like you and I do,” OpenAI head of product and engineering Yash Kumar said in an interview.
Artificial intelligence researchers call this type of technology AI agents. While chatbots can answer questions, write poems and draw pictures, agents can use other software on the Internet.
During a briefing with The New York Times, Mr. Kumar showed how the system could make San Francisco restaurant reservations through the OpenTable website and buy a grocery list through InstaCart. Operator looks and behaves like ChatGPT and other chatbots. The user types the request in a small window. Then the system gives its best answer.
A user can see when the tool opens a web browser and visits specific sites. The operator can make mistakes. But in some cases, it can correct these errors. During a demonstration for The Times, the system mistakenly assumed Mr. Kumar was in Iowa before it could correctly locate a restaurant in San Francisco.
The operator is not completely autonomous. Sometimes, the user needs to correct his mistakes and provide additional requests and suggestions. For sites like OpenTable and Instacart, users must provide their private usernames and passwords. But OpenAI said it does not store this private information.
However, the Company receives data that shows how the system interacts with users and accesses the Sites on their behalf. It can use this data to train future versions of the operator.
Starting Thursday, the operator will be available to anyone who subscribes to ChatGPT Pro, a $200-a-month service that provides access to all of the company’s latest tools, OpenAI said. It plans to roll out the tool through other paid services and eventually roll it into a free version of ChatGPT.
(The New York Times has filed a lawsuit against OpenAI and its partner Microsoft, alleging copyright infringement of news content about AI systems. OpenAI and Microsoft have denied the claims. )
In recent months, other well-known companies, including Google and Anthropic, have unveiled similar tools. Many of these tools, however, are not yet widely available.
The operator is based on the same technology that underpins ChatGPT. This technology is what AI researchers call a neural network — a mathematical system that can learn skills by analyzing large amounts of data.
Newer versions of this technology learn from a wider range of data, including text, images and sounds. In this case, the operator learned from images how people use spreadsheets, shopping sites and other online services. After identifying patterns in this data, the new system can use similar services on behalf of computer users.
Mr. Kumar acknowledged that like ChatGPT and other chatbots, Operator is still an experimental technology. But he said it would continue to improve in the coming months.
“It’s not the strongest thing in the world,” he said. “But it’s much better than this kind of technology.”
