Your AI assistant can now use your computer (and anything on it) • FRANCE 24 English
News & Politics
Introduction
In our daily lives, computers and phones are indispensable tools. Most of us likely rely on them for half or more of our daily tasks. The next major advancement in technology promises to revolutionize this interaction through the development of AI agents powered by large language models (LLMs). These hyper-capable AI systems aim to perform a variety of tasks that we typically accomplish on our computers.
Recently, the AI company Anthropic emerged as a leader in this space, providing a public demonstration of what it calls "computer use." This new functionality allows users to simply tell the software what to do, and it executes the tasks using its powerful Sonet 3.5 model. One notable example involves constructing a net worth calculator in Excel or LibreOffice—where the AI not only inputs data but also forms the necessary calculations. Furthermore, the AI showcased its creative side by painting a horse in Microsoft Paint, emphasizing its capability to generate content through direct action, rather than relying on diffusion technology.
How It Works
The underlying technology is quite sophisticated. Large language models are adept at processing text and images. By granting the AI access to the user's cursor and typing capabilities, it can navigate through tasks as a human would. It accomplishes this by taking multiple screenshots to reassess its surroundings at different stages of the task. For instance, if given the instruction to visit a specific website, the AI agent will click through links, take screenshots to evaluate its progress, and continue until it completes the assigned task.
In a recent demonstration, Anthropic's AI agent was tasked with finding a good location to watch the sunrise. It efficiently assessed the necessary logistics and updated a calendar invite, showcasing the potential of AI agents to streamline various everyday activities.
Limitations and Concerns
Despite the promising capabilities, several limitations and concerns accompany this technology. One immediate issue is the speed of execution. Demonstrations have shown that tasks can take longer than expected, necessitating time-consuming processes that may frustrate users. Another aspect is cost; the simplest operations can range from $ 3 to significantly higher rates for complex tasks, potentially making it a more costly alternative compared to human labor.
Furthermore, privacy and security issues arise with AI agents having access to personal computers. Users are advised to utilize virtual environments to mitigate these risks, but full functionality typically requires access to personal information, raising significant privacy concerns. Security risks were highlighted shortly after the release of this feature, when a malicious website was able to trick the AI agent into downloading harmful software.
Competitors in the Market
In the competitive landscape, Microsoft has made significant strides with plans to introduce ten different AI agents for its Enterprise clients. These agents will cater to various tasks, alongside providing clients the opportunity to create tailored agents. Similarly, Salesforce recently announced a suite of customer service-focused chatbots to assist in specific tasks.
Anthropic's recent developments represent a watershed moment, providing a proof of concept that many are eager to explore. As AI agents become more integrated into our daily lives, they hold the potential to reshape how we use technology.
Keywords
- AI agents
- Large language models (LLMs)
- Anthropic
- Computer use
- Privacy concerns
- Security risks
- Microsoft
- Salesforce
FAQ
What is an AI agent?
An AI agent is a sophisticated software application that utilizes large language models to perform various tasks on a computer, replicating human computer use.
How does the AI agent work?
An AI agent operates by gaining access to the user's cursor and typing capabilities, navigating through tasks while utilizing screenshots to assess its progress.
What are some limitations of AI agents?
The main limitations include speed, cost, and potential privacy and security risks due to the need for access to personal information and accounts.
Who else is developing AI agents?
Microsoft and Salesforce are two notable competitors working on AI agents for enterprise solutions and customer service, respectively.
Are there privacy and security concerns with AI agents?
Yes, granting AI agents access to personal computers raises significant privacy and security issues, as they could unintentionally download malware or expose sensitive data.