The year 2024 is all about AI agents. But what exactly are AI agents? To understand this, we first need to take a look at developments in the field of generative AI. One of the most significant trends is the shift from monolithic models to complex, composite AI systems.
From Monolithic to Modular Systems
Traditional AI models are based on the data they were trained on. This means they only have knowledge of the world that is available to them and are often used for specific tasks. Additionally, these models are difficult to adapt. Even if it is possible to improve or fine-tune a model, it requires a significant amount of data and resources.
For example: Let’s say I want to know how many vacation days I still have left. If I ask this question to a generative model, the answer will likely be incorrect, as the model doesn’t have personal information about me. This limitation can only be overcome by embedding the model into a larger system that can access the necessary databases.
Composite AI Systems: The Future
This is where composite AI systems come into play. These systems consist of multiple components that work together to solve complex tasks. In our vacation example, this means that the AI model wouldn't just give a generic answer, but also access a database where my personal vacation data is stored. The model would then create a query, retrieve the data from the database, and generate the correct response: "You have 10 vacation days left."
This principle makes it much easier to adapt such systems compared to adapting individual models because it is faster and more flexible. A well-known example of such a system is "Retrieval-Augmented Generation" (RAG), where models are combined with a search function to retrieve specific information from external sources.
The Evolution: AI Agents
This is where AI agents come into play. A significant breakthrough in AI is the ability of large language models (LLMs) to analyze and plan for complex problems. Essentially, an AI agent is such a model that is not only capable of processing data but also of acting strategically, using external tools, and adapting to various challenges.
An agent can not only access databases but also utilize external tools to perform calculations, search the web, or run programs. So, in our vacation scenario, an AI agent could predict the weather, calculate the amount of sunscreen needed based on sun exposure, and even recommend the number of sunscreen bottles to bring.
What Makes AI Agents Special?
Three key abilities differentiate AI agents from traditional AI systems:
- Reasoning and Planning: Agents can think through complex problems, break them down into smaller steps, and solve them systematically. Instead of providing a simple, possibly incorrect answer, they follow a path of careful planning and testing.
- Action Capability: AI agents can use external tools to complete tasks, whether it’s querying a database, performing a web search, or using a calculator for mathematical calculations.
- Memory: Agents can remember past interactions and conversations, allowing them to work contextually and give personalized responses.
One of the best-known configurations for AI agents is the so-called ReACT model, which combines reasoning and action capabilities. Agents working under this principle don’t just think; they also actively act, call external tools, and correct themselves if mistakes occur.
An Example: My Vacation in Thailand
Let’s imagine a concrete scenario: I’m planning a vacation in Thailand and want to know how many bottles of sunscreen I should bring. An AI agent could break down this complex task into several steps: First, it accesses my vacation data to determine the number of remaining vacation days. Then it checks the weather forecast for Thailand to calculate the hours of sunlight. Next, it researches the recommended amount of sunscreen per hour and finally performs a calculation to determine how many bottles I need.
Conclusion: The Future of AI
2024 will be the year of AI agents, as they represent a new level of autonomy and problem-solving. While simple, programmatic systems will still make sense for narrowly defined problems, AI agents offer a solution for more complex tasks through their flexibility and adaptability.
The Public Cloud Group is here to accompany you on this exciting journey. We help you find the right solution for your specific challenges, whether through the use of composite AI systems or the integration of agent-based intelligence. Our experts support you in developing and implementing these innovative technologies to make your business processes more efficient, flexible, and future-proof.
Don’t hesitate to contact us – we are your partners on the path to the world of AI agents.