Experts explain what the next level of AI tools, 'AI agents,' are
In recent years, many people think of chatbots like ChatGPT, which can generate natural conversations and sentences, and image generation AI, which can generate advanced images and illustrations simply by inputting text, as a field of AI that has been rapidly developing in recent years. Meanwhile, major technology companies such as OpenAI and Google have announced plans for 'AI agents' as a further wave of AI development. Brian O'Neill, a computer scientist at Quinnipiac University in the United States, explains what AI agents are, which hold the key to future AI.
What is an AI agent? A computer scientist explains the next wave of artificial intelligence tools
In November 2024, OpenAI announced 'Operator,' an 'autonomous AI agent' that performs multi-step tasks on behalf of users. OpenAI CEO Sam Altman said, 'The next big breakthrough is agents,' and it is considered a new major turning point for AI. According to reports, 'Operator' is expected to be released as a research preview in January 2025.
OpenAI plans to release an AI agent called 'Operator' that will operate PCs on behalf of humans in January 2025 - GIGAZINE
Google is also focusing on developing 'AI agents.' Google has revealed that it is developing a feature that allows users to ask AI 'what to do in the game' in real time using the AI model 'Gemini 2.0' announced on December 11, 2024. Google DeepMind CEO Demis Hassabis and CTO Korey Kavukcuoglu said, 'Using Gemini 2.0, this AI agent can infer about the game based solely on the actions on the screen and suggest what to do next. Not only will the AI agent act as a virtual gaming companion, but it will also be able to connect with a wealth of gaming knowledge on the web by using Google Search.'
Google announces it is testing a Gemini 2.0-based AI agent that can teach rules and strategies for games such as 'Clash of Clans' - GIGAZINE
A simple AI agent that is already familiar to us is the function that suggests replies to messages received in Google Mail. Other areas where AI agents are active include services that suggest flight and hotel reservations based on answers to questions such as destination and dates.
O'Neill defined an AI agent as 'a technical tool that can learn a lot about a particular environment and solve problems or perform specific tasks within that environment with just a few simple prompts from humans.' According to O'Neill, a robot vacuum cleaner that learns the shape of a floor or carpet and acts based on that information can be said to be a precursor to an AI agent. However, a robot vacuum cleaner is a 'goal-based' agent with a single goal of cleaning the floor, and its sole purpose is to achieve its goal by any means necessary, so it is a simple decision-making process.
On the other hand, today's AI agents are built on a 'utility-based' basis. Utility-based agents care more about how to achieve a goal than just achieving it. They can make complex decisions about which approach has risks and benefits, and which of multiple conflicting goals is more important, allowing them to solve tasks in a way that suits the user's preferences.
While basic chatbots and robotic vacuum cleaners are also types of AI agents, O'Neill noted that tech companies will increasingly refer to AI agents on a utility-based basis going forward. Unlike chatbots, which can recognize input words and provide simple responses, AI agents will need to be able to provide significantly more advanced responses and take action 'on behalf of the people and businesses that use them.'
If AI agents that can act on behalf of humans and companies, rather than simply continuing to perform specific tasks, develop, there will be concerns that they will take away human jobs. O'Neill points out that 'whether or not AI agents will erode human jobs depends on whether technology companies can prove that AI agents have the ability to overcome new challenges and unexpected obstacles that are not part of the tasks they are assigned to.' In addition, if they are to be entrusted with tasks other than specific tasks, it is also important to allow AI agents to access sensitive data.
Project Mariner , announced by Google on December 11, 2024, will allow AI to understand information on the browser and automatically operate Chrome. For example, if you want to buy a new PC, the AI will search for and suggest recommended PCs and peripherals that match it, but the AI agent cannot make the final purchase or agree to the site's terms of use. As Google has designed it, O'Neill says that by leaving the right to make the final decision to the user even in areas that are left to the AI agent, risks and AI bias can be reduced.
Related Posts:
