AI-powered browsers have been touted as the future of web browsing, but are they living up to the hype? The AI industry wants us to believe so, just like they promised with autonomous AI agents, but the reality might be a little different.
The Promise vs. The Performance
Let's dive into the world of AI browsers and uncover the truth. Recently, The Verge put these chatbot-integrated browsers to the test, and the results were not impressive. The browsers were often slow and frustrating to use, contrary to the automation benefits AI was supposed to bring.
"You have to think extra hard about crafting the right prompt," wrote Victoria Song from The Verge. Simply adding an AI assistant to a browser doesn't revolutionize the user experience.
AI Browsers in the Spotlight
OpenAI's new AI browser, Atlas, brought attention to this niche but important area of technology. With competitors like Perplexity's Comet and The Browser Company's Dia, the race is on to create the ultimate AI-powered browsing experience.
But mainstream browsers like Chrome and Edge already offer chatbot features. So, what's the big deal about full-blown AI browsers? Well, these browsers put AI front and center, encouraging users to interact with prompts before typing a URL.
The Verge tested these browsers extensively. One task they tried was organizing and summarizing emails, a feature often promoted by AI integration. However, this task proved to be a challenge, with AI browsers flagging unimportant emails and providing useless summaries. It required an incredibly detailed prompt to get any meaningful results.
"Find unanswered emails where I previously responded with interest or featured personalized requests/feedback. Then, evaluate which ones to respond to based on timeliness and keywords like 'embargo' with dates in the next two weeks. Ignore emails with multiple follow-ups to which I haven't responded."
Does that sound like a practical solution to you? While Comet and Dia managed to flag some relevant emails, others snagged spam. OpenAI's Atlas provided a technical explanation for its inability to fulfill the prompt and suggested refining it further.
Another task that proved challenging was shopping. AI browsers and models were supposed to excel at this, but they still made basic mistakes like recommending items in the wrong color. The process was equally frustrating, with OpenAI's Atlas repeatedly nagging the user to confirm items in the cart and struggling to navigate between windows.
The Problems with AI Browsers
AI browsers suffer from the same issues as AI agents: they're slow, require constant supervision, and need user approval for important decisions, defeating the purpose of having an autonomous helper.
But that's not all. AI browsers also pose security risks. They are vulnerable to prompt injection attacks, where hackers can deliver hidden messages to the AI, potentially embedded in malicious webpages. Researchers have demonstrated that Perplexity's Comet could be manipulated to grant hackers access to your bank account, and OpenAI's Atlas could be tricked into visiting your Google Drive and deleting files.
Safety should be a priority, but if AI browsers can't provide a seamless experience, their existence might be at stake. For now, they're more like a demanding child that needs constant attention rather than a personal assistant.
"My AI browser experience reinforced the idea that I spend more time doing things for AI than the other way around," Song wrote. "It's about adapting my natural behavior to accommodate AI, rather than the other way around."
So, are AI browsers the future? It seems we still have a long way to go before they can live up to their potential. What do you think? Should we continue to develop and improve AI browsers, or are there better alternatives?
