The Evolution of AI: From Generative Models to Intelligent Agents
Introduction to AI Tool Use
Generative AI models have been making waves in the tech world, but the next big leap involves AI that can operate other software on a computer or via the internet. This development is crucial for making AI more useful and intelligent.
Claude’s Role in AI Development
Anthropic has been collaborating with various companies to integrate Claude-based AI helpers into their workflows. For instance, an online tutoring company has utilized Claude to assist their workers. One notable application is an AI agent designed to manage online shopping returns by locating receipts in Gmail, filling out return forms, and scheduling package pickups.
Challenges and Cautious Deployment
Google and other companies are cautiously rolling out these AI agents. The main challenge is ensuring that AI agents correctly understand and execute tasks. Large Language Models (LLMs) can sometimes misinterpret instructions, leading to errors in task completion.
Controlled Environments for AI Agents
Limiting AI agents to specific tasks within a company’s workflow can make the technology more reliable. Similar to how physical robots are used in controlled environments to minimize errors, AI agents can be more effective when their roles are clearly defined.
Potential for Profit
Even these early applications of AI agents can be highly profitable. Some large companies have already been automating common office tasks using AI for over a year. Adept, a company in this space, is secretive about its clients and the tasks its agents perform, but their strategy is clear.
“Our agents are already in the 90s [percent] for reliability for our enterprise customers,” Luan says. “The way we did that was to limit the scope of deployment a bit. All the new research we do is to improve reliability for new use cases that we don’t yet do well on.”
Training AI for Better Understanding
Adept aims to train its AI agents to better understand the goals and steps required to complete tasks. This approach could make the technology versatile enough to assist in various workplaces. Luan emphasizes the importance of understanding the reward of the task at hand, not just mimicking human behavior.
The Bigger Picture: Advanced Machine Intelligence
The core capabilities needed to make AI agents more useful are also essential for advancing machine intelligence. Currently, the ability to plan and achieve specific goals is a hallmark of natural intelligence that LLMs lack.
The Evolutionary Perspective
While it may take a long time for machines to achieve human-like intelligence, the concept of tool use is significant. In the natural world, early hominids used crude stone tools for tasks like cutting animal hides. As tool use became more sophisticated, so did human intelligence. Perhaps now, it’s time for one of humanity’s most advanced tools—AI—to develop its own tool-using capabilities.
5 Comments
And yet, people still miss the good ol’ phone calls!
Life used to be simpler when talking to a person was the norm.
Emma Leigh: But have they really improved our lives?
So, are we really just talking to machines now?
Considering the rise of AI, are we just embracing more digital isolation?