AiN # 26: Multi-Agent Systems or Agentic AI

Welcome to the Augmented Intelligence Newsletter (AiN) by Dr. Chan Naseeb.

Hey, in this issue, I explain and talk about What are the Multi-agentic Systems or What is Agentic AI? What differentiates LLMs? What is RAG, and Agentic RAG and how these technologies such as AI, RPA, LLMs, RAG, Agentic RAG, and Multi-Agent Systems are going to impact the future of the businesses?

Multi-agent systems (MAS) and the Agentic approach have been getting much attention recently, thanks to AI[1] and the momentum accelerated by Generative AI [2]. For example, you might have seen the term Agentic AI being used very often. While these are thrilling developments, the Multi-Agent approach to solving problems has been around for quite some time and has some very sound theoretical underpinnings. For your reference, have a look at this paper to see how large scale multi agent system was proposed to use for personnel distribution [3] back in 2002, all that was without LLMs [4].

Times have changed, and LLMs have introduced at least an order of magnitude improvements in the quality we can expect from MAS. Today there are many LLMs and Foundation Models such as GPT4, Llama (Meta), BERT, Bard, Gemini, Falcon, LaMDA, Granite (IBM) [5] have been developed and many more application specific such as LLM for timeseries [6] are being released on regular basis. All this was made possible thanks to the very fast usage of ChatGPT [7]. However, as many of these are being developed based on the internet data, or other public data which has several issues, there is a dire need for building AI systems and agents which are trustworthy [8].

Agentic AI Systems?are architected to resolve complex problems with minimal human charge. These systems are composed of multiple conversable agents that talk with each other and can be orchestrated centrally or self-organized decentralizedly. Thanks to the pace dictated by generative AI and LLMs, the usage of multi-agent systems has increased in the enterprise to automate complex processes or solve complicated tasks.

A multi-agent or "self-organized" system is a computerized system with multiple interacting intelligent agents. These systems can solve challenging or impossible problems that would be impossible for an individual agent or a monolithic system to unravel. Different techniques, such as methodic, functional, procedural approaches, algorithmic search, or reinforcement learning [9], can be leveraged to infuse intelligence into these systems.?

The rise of agentic process automation is on the verge. Two powerful technologies are converging to shape the prospective automation landscape: robotic process automation (RPA) and agentic artificial intelligence (AI). RPA has revolutionized repetitive, rules-based tasks, streamlining workflows and increasing industry efficiency. Agentic AI is gifted with the capacity to perceive, reason, and learn autonomously to handle complex decision-making and problem-solving tasks. From automating mundane tasks to navigating ambiguity and uncertainty, the synergy between RPA and agentic AI promises to revolutionize businesses' operations.?

Artificial Intelligence?(AI) continues to change many aspects of modern life, from chatbots handling customer requests to digital AI assistants answering everyday questions. Retrieval Augmented Generation?(RAG)?technology refers to AI models that utilize external sources to generate original content. Rather than using static training datasets, RAG uses new, up-to-date data obtained from indexed documents or web pages. In other words, it is the process of optimizing the output of a large language model so it references an authentic knowledge base outside of its training data before generating a response.

Large Language Models (LLMs) are trained on extensive volumes of data and use billions of parameters to generate original output for tasks like answering questions, translating languages, and completing sentences. RAG [10] extends LLMs' existing robust capabilities to specific domains or an organization's internal knowledge base without retraining the model. It is a cost-effective approach to improving LLM output so it remains relevant, accurate, and useful in various contexts. This methodology enables RAG-powered bots to remain appropriate in quickly changing business dynamics while providing richer results than previous techniques.

Agentic RAG is?an evolution of classic RAG, combining AI agents to enhance the RAG approach. This technique employs autonomous agents to analyze initial findings and strategically select adequate tools for data retrieval. One of the most promising developments in?AI?is the combination of two unique concepts:?Retrieval Augmented Generation (RAG)?and autonomous AI agents. Together, they form Agentic RAG [11], an innovative development that will transform how businesses engage their customers, researchers explore new areas, and individuals communicate daily.

I regularly write and talk about AI, business, technology, digital transformation, and emerging trends.

1. LinkedIn
2. Medium
3. Youtube

Subscribe to this newsletter or click 'Follow' to read my future articles. You'll be able to read the previous issues here. Also, let me know in the comments if you want me to write about a specific topic that interests you.

Enjoy the newsletter! Please help us improve it by sharing it with your network.

Have a nice day! See you soon. - Chan

References.

1. Understanding AI, ML, DL and Data Science
2. Power of Gen AI: daily life
3. Personnel distribution through Multi-agent system
4. Tokens in LLMs
5. IBM Foundation Model Granite
6. LLMs for Timeseries
7. ChatGPT
8. Trustworthy AI
9. Reinforcement Learning
10. RAG and Customer Care
11. Agentic RAG