AI agents have emerged as a significant force in the world of artificial intelligence, yet their exact definition remains elusive. People are divided on what constitutes an AI agent, and even among tech giants, there is no consensus. In this article, we will explore the various aspects and potential of AI agents, delving into their functions, challenges, and the future they hold.
Defining AI Agents: A Complex Endeavor
At its core, an AI agent can be described as AI-fueled software that takes on a series of tasks that were previously handled by human service agents. It goes beyond simply answering questions and can cross multiple systems to achieve complex goals. For instance, Perplexity's AI agent assists with holiday shopping, while Google's Project Mariner helps with flight and hotel bookings, household item shopping, and more. However, the lack of clarity in defining these agents leads to confusion about their exact capabilities and applications.Even within different companies, the perception of AI agents varies. Asana views them as extra employees taking care of assigned tasks, while Sierra sees them as customer experience tools. This diversity in definitions shows the complexity of this emerging technology.The Role of AI Technologies in AI Agents
AI agents utilize a variety of technologies to function effectively. Natural language processing, machine learning, and computer vision are some of the key techniques incorporated. These systems enable the agents to perceive their environment, reason about it, make decisions, and take actions autonomously. For example, in dynamic domains, these technologies allow the agents to operate seamlessly and handle various tasks without constant human intervention.Challenges in Creating AI Agents
One of the major challenges in developing AI agents is crossing different systems. Many legacy systems lack basic API access, making it difficult for software to access and integrate multiple systems while solving problems. This complexity can limit the agents' capabilities and performance. Additionally, as AI becomes more capable, there are assumptions about its growth, but not all areas are guaranteed to improve at the same rate. Rodney Brooks pointed out that AI has to deal with tougher problems than other technologies, and achieving rapid growth like chips under Moore's law is not a certainty.The Future of AI Agents
Over time, as AI continues to evolve, AI agents are expected to be able to do much more on behalf of humans. There are already dynamics at play that will drive this evolution, such as GPU price/performance improvements, model efficiency, and quality. However, it will likely take multiple models rather than a single large language model to make agents work effectively. Fred Havemeyer believes that the most effective agents will be multiple collections of different models with a routing layer to direct requests. The goal is to create truly autonomous agents that can take abstract goals and reason out the individual steps independently.While we have seen promising steps in the development of AI agents, we are still in a period of transition. There are advancements and breakthroughs needed for these agents to operate as envisioned. It is important to understand that we are not there yet, but the potential of AI agents is vast and holds great promise for the future.
