Are AI Agents the Next Frontier of AI? by Timothy Smith

Photo Source: Flickr

If 2023 represented the year of AI, especially large language models or LLMs for short, such as ChatGPT, then 2025 will become the year of agentic AI or AI agents. AI agents represent the next step in AI evolution, which pulls together different types of AI to perform complex operations autonomously. Following the revelation of LLMs and their uncanny ability to answer natural language questions, summarize massive amounts of information, and translate many languages, AI has irretrievably entered everyday people's lives at home, work, and play.

LLMs such as GPT4 and others require massive amounts of information and data at the level of the whole internet to learn the patterns of language that make them so human-like in their responses to questions and requests. The price for such training ranges from $5 million-$100 million due in part to the billions of data points, tens of thousands of computer chips, and the staggering amount of electricity needed to train these models. However, as powerful as these models appear, their learning stops at the end of training, leaving the models locked in the past and needing updating regularly. In other words, these models cannot evolve and take on new challenges.

AI agents transcend the limitations of LLMs by taking advantage of their strengths in combination with other AI tools, such as reinforcement models that get better at tasks by learning from their mistakes. Reinforcement models such as AlphGO and DeepBlue became master GO and chess players by playing the games over and over again and remembering what works and does not work to win. This reinforcement learning helped the machines defeat human grandmasters in these games. AI agents use a series of AI tools configured to solve problems rationally and make autonomous decisions with their findings. In other words, the AI agent is a software program designed to take in information, use different tools to analyze the information, do tasks based on inputs from other tools, and perform tasks autonomously, such as decision-making, recommending a course of action, and learning to adapt to changing conditions.

A commonly known AI agent found in some automobiles, such as Tesla, is autopilot. The autopilot system takes information from various sources, such as sensors that detect other objects around the car, global positioning satellite information for location, and local traffic laws to guide the automobile safely through town and country without human intervention. Such an advanced agent can incorporate other data such as traffic and weather conditions. These machines need to make rational decisions continually to avoid crashes and protect people inside and outside the vehicle.

Autonomous cars represent a very high-risk type of AI agent, with many developers and engineers to help bring this product to market. Still, AI agents will also rapidly gain ground at home and at work for millions of people through agents as diverse as nutritional and health coaches to supply chain optimization. LLMs now provide the problem-solving ability to help non-computer scientists develop helpful AI agents that serve many functions. If you asked ChatGPT to set up a meal plan and grocery list for one week for your family. It would generate something like this: (I did not include the whole week to save space)

Hi Copilot, would you please generate a meal plan for my family of four--a man, a woman, a 12-year-old girl, and a 14-year-old boy as well as a grocery list to supply the kitchen

Copilot:

Sure, I'd be happy to help with that! Here's a balanced meal plan for a week, along with a grocery list to get you started:

Meal Plan

Monday

Breakfast: Oatmeal with fresh berries and honey

Lunch: Turkey and cheese sandwiches with carrot sticks

Dinner: Grilled chicken, quinoa, and steamed broccoli

Snacks: Greek yogurt, apple slices

…

[Grocery List]

Although useful as a quick plan, an AI agent with access to your local grocery store website, a budget for food, and any food allergies and preferences could generate a much more specific menu. An AI agent could also learn from feedback from the family and even incorporate requirements such as holidays, calorie restrictions, nutritional needs such as low sodium, or the need for a meal to last two nights to reduce cooking responsibilities. If designed properly, an AI agent can provide such support. Not everyone wants help with meal planning, but the example shows how an AI agent with access to the right information could help with planning. The same could hold for vacations, home improvement, and finances. AI agents will become more available through apps like Copilot, Microsoft, or Claude from Anthropic, as well as different companies that want customer loyalty. 2025 might well be the year of the AI agent.

Dr. Smith’s career in scientific and information research spans the areas of bioinformatics, artificial intelligence, toxicology, and chemistry. He has published a number of peer-reviewed scientific papers. He has worked over the past seventeen years developing advanced analytics, machine learning, and knowledge management tools to enable research and support high-level decision making. Tim completed his Ph.D. in Toxicology at Cornell University and a Bachelor of Science in chemistry from the University of Washington.

You can buy his book on Amazon in paperback and in kindle format here.