Generative AI chatbots such as ChatGPT, Microsoft’s Bing and Google’s Bard are barely even a thing yet, but the internet is already buzzing about their supposed replacements.
AI agents including Auto-GPT and BabyAGI are essentially bots that use bots for you, and they’re being hailed as a set-and-forget solution to have AI accomplish any task.
Auto-GPT allows the creation of AI agents, which can autonomously use OpenAI’s GPT-4 to achieve a set goal.Credit: AP
While chatbots can save you time by doing quick web searches and comparisons, summarising the results in plain language, AI agents take that idea one level further. They have access to the internet and your company’s files, can write and execute their own code, and solve high-level tasks by breaking them down into sub-jobs, creating additional agents and delivering results automatically once done.
Unsurprisingly, given AI and bots have taken over from crypto and blockchain on the cutting edge of get-rich-quick hype, these agents instantly became the talk of programming circles and hustle bros alike. But what can you actually do with them? The answer is, so far, practically nothing.
Despite an abundance of Twitter threads showing how an AI agent could be used to create entire websites and services automatically, or conduct market research on competitors, or surface sales prospects, or instantaneously write a script for a current affairs podcast, these applications range from experimental to hypothetical, and none appear to produce an output anywhere near usable.
To be clear, this is very new technology (Auto-GPT is hardly a month old!) and will grow quickly in capability. AI agents are now by far the most popular repositories on code-hosting service GitHub, and thousands of developers are working to figure out how best to make these things useful.
Even Meta is getting in on the buzz, with Mark Zuckerberg telling investors on Thursday that the company saw “an opportunity to introduce AI agents to billions of people in ways that will be useful and meaningful,” through Facebook, Instagram, WhatsApp and Messenger. Though he didn’t mention any specific applications, Zuckerberg gestured vaguely in the direction of ad creation, customer support, “and over time, video and multi-modal experiences as well”.
But the fact is AI bots themselves are in a nascent stage, they are prone to hallucinating, lying, making wild assumptions and getting confused if they try to remember too much information. Having a bot with memory manage dozens of others to accomplish bigger tasks might magnify the hype and get investor tongues wagging, but it’s also going to magnify the instability.