Google, OpenAI and Anthropic all make AI systems that run your PC, performing online tasks such as filling in forms, researching, and even shopping.

If you were hoping that AI was just a tech bubble, that would burst soon like 3D televisions (remember those? You'll need to keep hoping as major players in this industry are rushing ahead with new AI agents. Google, OpenAI and Anthropic have all been busy creating computer-using agents that will take control of your web browser to perform various tasks.

ChatGPT is a powerful tool that can help you with a variety of tasks. It can summarize a document, complete a form and even suggest products to buy. To do this, you will need to provide the AI system with all the relevant documents and links. The next step for artificial intelligence is to simply tell it what you need and it will do it all for you.

According to a report from The Information (via Reuters), Google is currently working on this project. The new AI system will be codenamed Project Jarvis and will work with Google's next generation Gemini LLM (large-language model). It is likely to directly integrate into the Chrome web browsing.

The system will literally take control of the browser and do all the things you ask it to via the interface. These AI tools are known as computing-using agents (CUAs), and they are the current darlings of the AI industry.

Earlier this year Reuters reported about OpenAI's 'Strawberry,' a project whose stated goal was to "enable the company's AI not only to generate answers to queries, but to plan ahead to navigate the Internet autonomously and reliably in order to perform what OpenAI calls deep research."

Anthropic announced last week that its latest LLM Claude 3.5 Sonnet can be used to operate a computer through its API. "[D]evelopers are able to direct Claude to work computers the same way humans do--by viewing a screen, moving a mouse, clicking buttons, or typing text." The company has produced a short video that shows a possible use of the system.

While I can see that they all have real benefits, I am concerned about a few issues. One is that generative AI doesn't work perfectly. Just look at Google's experimentation system when you search for anything.

Anthropic admits that the CUA can be "at times cumbersome, error-prone", and that it comes with several serious risks.

My biggest concern is who is ultimately liable for any mishaps that the CUA may cause. Imagine you asked it to do some online shopping, and you found out later that your bank account was emptied.

Would you be reimbursed by any AI companies? I doubt it, and there will likely be very specific clauses in their user agreement that say you are responsible for the entire cost.

If this is the case, and it most likely will be, then I don't think many tech-savvy users would agree to use a CUA. These people are few compared to the number of PC users who don't know what goes on behind the scenes, or the dangers associated with using them.

Some people, according to surveys, want strict regulations in place to prevent AIs from becoming super-powerful. As CUAs become more popular, I wonder if people will want to take the same action against them. My instinct says that the average PC user is going to see them as a useful tool, and will be oblivious of the risks they pose for cybersecurity and personal safety.