S Poddar
This week, OpenAI released its first independent AI agent, called Operator. Operator can act autonomously using a web browser on your computer, performing many of the tasks that can be done through a web browser. That’s to say it could presumably book restaurant reservations or buy groceries for you. You simply tell Operator what you want, and it gets to work like a diligent internet-enabled butler, completing the task or coming back to you with questions if needed. For example, if no tables are available at 7:00 PM, Operator might ask if you’d prefer a table at 7:45 PM instead.
While Operator shuns the use of terms such as “Sir” or “Madam,” it essentially plays the role of the internet butler we’ve been promised since the advent of Ask Jeeves in 1997. Ask Jeeves was a search engine that featured a character named Jeeves, a butler who would assist users in finding information online using natural language questions. Ask Jeeves was different from other search engines because it enabled users to look for information by asking natural language questions rather than keywords. For example, a user could ask “What’s the perfect side dish for a roast dinner?”
Search Engine Optimization and the Emergence of Natural Language Search
The search engine wars were won by Google, and Ask Jeeves eventually dropped the Jeeves character and rebranded as Ask.com in 2006. However, with the rise of AI technologies like ChatGPT search and Perplexity, natural language search is back in vogue. Today, these AI agents function very similarly to our當初 (dāngchū) [back then] internet butlers, but with the more modern title of AI agents.
The Goal: AGI
OpenAI’s fundamental goal is to develop artificial general intelligence (AGI), or strong AI/superintelligence. It means an ability for a machine to perform the kind of general intelligence of a human, or human-like, even in a wide range of cognitive activities. All of the recent successes of chatbots, like ChatGPT are viewed by OpenAI as an accidental product of working toward AGI.
In a promotional video for Operator’s release, an employee of OpenAI sitting next to Sam Altman states explicitly that Operator’s purpose is to “remove one more bottleneck on our path to AGI.” While AI agents are clearly exciting advancements, they are just one step along the way for OpenAI. AGI has the potential to fundamentally alter our world. Once artificial intelligence surpasses human intelligence, it would be logical to assume that it could create even more advanced versions of itself, leading to an intelligence explosion. The concern is that this superintelligence could decide to eliminate humanity.
Experts Divided on the Safety of AI Agents
While some view AI agents as a threat, others believe they can be beneficial for society. At the World Economic Forum in Davos this week, artificial intelligence pioneer Yoshua Bengio voiced his concerns that AI agents could be catastrophic for humanity. In an interview with Business Insider, he said, “All of the catastrophic scenarios with AGI or superintelligence happen if we have agents.” Bengio argues that we should focus on achieving AGI without using agents that can act autonomously. “All of the AI for science and medicine, all the things people care about, is not agentic,” Bengio said. “And we can continue building more powerful systems that are non-agentic.”
The Dark Side of AI Agents
So, can it really be that an internet butler meant to do things like grocery shopping is going to accidentally give AI the keys to the kingdom? At this point, it is hard to see how a bot that systematically orders restaurant reservations with a web browser is going to bring about the end of humanity. But AI agents succeed if and only if users adopt them, and here there is some scope for skepticism.
Personally, I’m not ready to give my credit card information to a computer program that is supposed to make purchases on my behalf in order to save time. I just do not have total confidence that it wouldn’t mess up. Would you trust an AI agent with your financial information? Maybe OpenAI needs to make Operator more user-friendly and trustworthy if they want to gain widespread adoption. Surprisingly, it almost seems that the good old-fashioned butler character might be experiencing a revival.
