The release of a new set of tools to obtain information and perform online tasks promises to reinvent the traditional search and carry the chatbots from AI to the next level. But it comes with significant questions about reliability and security.
Openai, the popular chatbot manufacturers with the chatgpt, began to implement the chatgpt agent earlier this month, and other important AI players are dating similar tools. They are banking that users will buy in Agentic AI, a technology designed to be a virtual personal assistant, as is now known, an AI agent.
“Then, an AI agent takes autonomous measures based on the goal you give,” said Canadian futuristic Sinead Bovell, in an interview with CBC News.
“You don’t have to say how to look for information, make a plan, book a flight, etc.”, said Bovell, the founder of the Waye Technological Education Company, who advises organizations on emerging technologies.
What are AI agents?
The chatgpt agent, which is currently available for most subscribers paid, brings together two tools that Openai has been developing: deep research, an advanced search tool that is already in use in ChatgPT and operator, which is supposed to complete online tasks for users interacting with websites.
Users access the agent through the family chatbot, but the chatgpt agent He works on his screen through his own virtual browser and computer.
The general objective with AI de Agent is to perform online tasks in a more natural and efficient way.
For example, instead of spending hours reviewing different airline websites that try to find the least expensive flight, using an AI agent would be like interacting with a trusted administrative assistant who seeks him Tedioso, then presents the best options.
The agent agitation of the Agent
The agentic AI has been generating very hot in the circles of AI, with several great players chasing it.
“The agents of AI are seen as the next step in the evolution towards an Ai-First society, or AI’s age,” Bovell said. “Then, somehow, they are the Holy Grail, and where the great competition is found.”
Google, for example, has been developing its agent of the Mariner Project, which announced at the end of 2024. In May, Google said it would be available to those in the United States who paid for their highest level of AI.
Like Pichai, CEO of Google and Alphabet, said in a key note from Google in May that the company has progressed a lot since then.
CBC radio columnistsThe way we buy, travel and administer our money is changing with AI agent
The personal finance columnist, Rubina Ahmed-Haq, says that AI AGENTIC is the next iteration of artificial intelligence, and can operate with a minimal human intervention. Would you allow an AI agent to manage your money independently?
“First, we are presenting multitasking, and now you can supervise up to 10 simultaneous tasks. Secondly, you are using a feature called Teach and Repew. This is where you can show you a task once, and learn a plan for similar tasks in the future.”
Perpleplexity, a search engines company based in IA, launched its agent browser, Comet, in early July, with Comet Assistant. Comet is available to select paid subscribers. There are also other smaller companies that work in the agent space.
The reliability factor
As anyone who has used chatbots from AI can witness, these systems make mistakes. So what does that mean for the reliability of the tools that are intended to perform tasks in your name?
“Every time an AI agent does something, such as reserving a flight, completing a form, building a web page or answering an email for you, needs a fairly precise understanding of what is happening,” explained the cognitive scientist, author and businessman of AI, Gary Marcus, by email.
“Even a small slide, such as reserving the incorrect incorrect ticket, can be expensive,” he said.
“The reality is that these systems say things that sound good, because they model how humans speak, but the detail is often incorrect.”
Railings
When launching its chatgpt agent, OpenAi warned about the problem of errors or the risk that the agent could be deceived.
At the launch of July 17, Openai Casey Chu researcher gave an example of an AI agent who could run into a malicious website that asks him to enter credit card information.
“An agent who is trained to be useful could decide that it is a good idea. We have worked hard to try to ensure that this does not happen. We have trained our model to ignore suspicious instructions on suspicious websites.”
When using AI agents, users must have control that allows them to intervene, correct and verify so that they do not perform harmful steps. We can see how that is developing.
The chatgpt agent, for example, has the acquisition mode, which allows the user to manually enter the credit card information. Chu also pointed out that the agent “will ask for confirmation in the last step.”
Even so, some have deep security concerns. In March, Meredith Whittaker, president of the Privacy Centered Messenger application, told an audience at the Tech SxSW conference “there is a deep problem with security and privacy” in the use of AI of Agent.
For an AI agent to be really useful as a personal assistant, he would need to access things like his calendar, his contacts, his email.
A group of students from Calgary High School offers free courses on artificial intelligence to younger students. Classes are open to children in the degrees from seven to 10, take place in the library of the University of Calgary and aim to teach students how to use AI tools responsible as chatgpt.
The end of Google?
Google has a lot to win or lose in AI’s fight, since he has been the dominant player in how people get online information for so long. AI seems to be in the center of its strategy to stay at the forefront of its competitors.
Google has begun to implement the AI mode, which incorporates more characteristics promoted by AI that allows users to do things such as requesting longer and more complex consultations. It is assumed that its sailor agent project is included in AI mode in the future.
While Google made his name helping human users get information quickly and easily, how we get information is changing.
Bovell says that current Internet browsers are not designed for the AI.
“They are designed for human trafficking. Human spectators, humans who look at certain web pages,” he said.
“We have reached the beginning of humans as the main Internet users.”
New research from The PEW Research Center He suggests that people who read Google’s descriptions have already reduced people who really visit websites. Users who did not see a summary of AI clicks on links to websites almost double those who did see a summary.
This has great implications for any content creator who trusted the traffic of search engines to see their online ads.
At this time, Openai, Google agent systems and perplexity are linked to paid levels. Therefore, it is still an open question about whether these tools will be widely available for free or if they will be widely adopted.
In the near future, users will have to balance the usefulness of the characteristics with the need for supervision and cost.
But ultimately, the rules that govern a future in which AI agents act in our name are not yet written, something that concerns people as Bovell.
“Perhaps I am surprised that we have not yet resolved some of the security vulnerabilities, the vulnerabilities of privacy and the challenges with the errors and the systems of AI still inventing things, but we are still at full speed with the agents of AI.”
