Release a new toolset to get information and accomplish tasks Online promise online tasks, reinvent traditional searches, and take AI chatbots to the next level. However, there are important questions about reliability and security.
Openai, the manufacturer of popular ChatGpt AI-powered chatbots, began deploying ChatGpt agents earlier this month, with other AI players unveiling similar tools. They are banking businesses that buy into Agent AI, a technology designed to make users become virtual personal assistants, or as is currently known, designed to become AI agents.
“So AI agents are primarily autonomous behaviour based on the goal you give it,” said Canadian futurist Sinead Beauvel in an interview with CBC News.
“There's no need to tell them how it finds information, plans, or book flights,” said Bovell, founder of Tech Education Company Waye.
What is an AI agent?
Currently available to most paid subscribers, the CHATGPT agent brings together two tools developed by OpenAI. Deep Research, advanced search tools already used in ChatGPT, and interact with websites to complete user online tasks.
Users access the agent via the familiar chatgpt chatbot, but chatgpt agents It runs on the screen via its own virtual browser and computer.
The general goal of Agent AI is to achieve online tasks in a more natural and efficient way.
For example, rather than spending time through the websites of various airlines trying to find the cheapest flights, using AI agents is like interacting with a trusted administrative assistant who does boring searches for you, offering you the best options.
Agent AI Gold Rush
Agent AI generates a lot of heat in AI circles, and many major players chase it.
“AI agents are seen as the next step in evolution into the AI first society, or the age of AI,” Bovell said. “So, in a way, they're the Holy Grail and there's a great competition.”
For example, Google is developing the project Mariner AI agent, which was announced in the second half of 2024. In May, Google said it was available to people in the US who paid for the best AI tier.
Sundar Pichai, CEO of Google and Alphabet, said in Google Keynote in May that the company has made many advances since.
CBC Radio ColumnistThe way we shop, travel and manage is changing with Agent AI
Personal finance columnist Rubina Ahmed Hack says that Agent AI is the next iteration of artificial intelligence and could work with minimal human intervention. Do you allow AI agents to manage your money independently?
“First of all, we've implemented multitasking, but now we can oversee up to 10 concurrent tasks. We're using a feature called Teach and Repeat, which allows us to show the task once and learn to plan similar tasks in the future.”
Perplexity, an AI-based search engine company, released Comet, an agent browser featuring Comet in early July. Comet allows you to select paid subscribers. There are also other small businesses that also work in Agent Space.
Reliability factors
These systems make mistakes, as anyone who has used AI chatbots can prove. So, what does that mean for the reliability of a tool intended to take on the tasks for you?
“Every time an AI agent does something like filling out flight books, filling out forms, creating webpages, responding to emails, etc., you need to understand fairly accurately what's going on,” cognitive scientist, author and AI entrepreneur Gary Marcus explained in an email.
“Even small slip-ups like booking the wrong, non-refundable ticket can cost you,” he said.
“The reality is that these systems model the way humans speak, but the details are often wrong, so they're saying something that sounds a bit right.”
guardrail
When launching the CHATGPT agent, Openai itself warned of any error issues or risks that could cause the agent to be misunderstood.
At the July 17 launch, Openai researcher Casey Chu gave an example of an AI agent that could stumble over a malicious website asking them to enter their credit card information.
“An agent trained to be useful may decide that it's a good idea. We did a lot of work to ensure that this doesn't happen. We trained the model to ignore suspicious instructions on suspicious websites.”
When using AI agents, users need to be controlled to allow intervention, correction and verification to avoid performing harmful procedures. You can already see how it unfolds.
For example, the ChatGpt agent has a takeover mode that allows users to enter their credit card information manually. Chu also said the agent would “please ask for confirmation in the final step.”
Still, some people have deep security concerns. In March, Meredith Whittaker, president of privacy-focused messaging app signaling, told an audience at the SXSW Tech Conference that there are “deep security and privacy issues” when using agent AI.
For AI agents to be extremely useful as personal assistants, they need to access calendars, contacts, emails, and more.
A group of Calgary High School students offers free courses on artificial intelligence to younger students. The classes are aimed at children in grades 7 to 10, and are held at the University of Calgary Library and aims to teach students how to use AI tools such as ChatGPT responsibly.
The end of Google?
Google has many things to win or lose in Agent AI battles. This is because for a long time, they have been the dominant player in the way people get information online. AI appears to be at the heart of its strategy of staying ahead of its competitors.
Google has begun deploying AI mode. This incorporates more AI-driven features, such as users asking for longer, more complex queries. The AI agent project Mariner is to be installed in AI mode.
Google has created names that help human users get information quickly and easily, but the way information is being retrieved has changed.
According to Bovell, current internet browsers are not designed for AI age.
“They are designed for human traffic. Human viewers, people who are looking at certain web pages,” she said.
“We have reached the beginning of the human end, the leading user of the Internet.”
New research from Pu Research Center It suggests that people reading Google's AI overview are already cutting back on the number of people who are actually visiting the website. Users who did not see the AI summary clicked on a link to their website almost twice as often as those who saw the summary.
This has a lot to do with content creators who rely on traffic from search engines to turn to online ads.
Nowadays, Openai, Google, and the agent system of confusion are all tied to the paid class. So whether these tools will become widely available for free or widely adopted is still publicly available.
In the near future, users will need to balance the usefulness of the feature and the need for supervision with costs.
But ultimately, the rules still remain unwritten to govern the future where AI agents act on our behalf. This concerns people like Bovell.
“I'm probably a bit surprised that security vulnerabilities, privacy vulnerabilities, mistakes and the fact that AI systems haven't yet been sorted out, but AI agents are still at full speed.”
