Information to Action The Evolving Path of Generative AI

What differentiates AI from Gen AI is its ability to generate content be it text, video, audio, image, or a combination of all. So far Gen AI has improved upon its ability of generation in a very mature manner in integration with multiple techniques embedded with Gen AI such as RAG (Retrieval Augmented Generation), GRAG (Graph Retrieval Augmented Generation), TAG (Table Augmented Generation), In Context Learning, Transfer Learning, and others.

With this evolution, the confidence of users in the confusion matrix has deepened and the LLMs (Large Language Models) are seen as close friend / co-pilot, assistant and more, where precision, faithfulness, answer relevancy, and recall are harmonized to give the correct response for a given context. However, there is still a lot of work that needs to be done to make it perfect, but one can say it is on the path of maturity very visibly and the day is not far when it will be difficult to make a difference between the ground truth and with the generated content.

This is what the claim most of the research in space of Gen AI is alluding to. But, again, this is only one dimension of Gen AI capability, that we are talking about, the “Information Retrieval.” What about when you have the information with you and you have to act upon it based on the information given, like taking certain decisions, performing tasks in a value chain, aggregating/deriving information to a specific context, slicing/dicing the information to different segment/users and many more. So, as a next step, a lot of work already has started and will keep happening in the coming time to add one more dimension to Gen AI capability from thinker to doer. When we think of any organizational construct, there a two major pivots, “Doers” and “Thinker,” and it is said to improve efficiency the ratio of two must be in the right way so a business can optimally reap the benefits. While the thinker is helping you to retrieve information and the doer will help you to implement the information at the right place in the right context.

?Let us see this from the point of view of the Taker / Shaper (Taker – Adopting the LLMs as its, Shaper – Tweaking the LLMs to an in-context window) with this, If I can draw an analogy specifically in an IT Organization, the thinker will be your domain SMEs, how will train the LLMs to a specific task / sub-task, create the prompt templates, validate the response, etc. The doers will be more focused on executing the task based on asking in prompt, and it is prompt which will guide which information to seek and what to act based on information retrieved in a very subtle manner.

?This shift is making LLMs add one more wing to their construct as Agents, who can listen to the LLMs response as an attentive listener and call the agents in sequence or in parallel to perform the task. This construct will add the flavor of traditional AI / ML along with Gen AI to microservice architecture so that the actions from agents are aligned to the ask in the prompt. However, a few architectures are already in place like LangChain and others, but when it comes to attributes of making them perfect agents are still in question.

Multiple research works are in progress to improve upon reasoning, sensing, coordinating, acting, learning, and adapting attributes of agents in a modular way so they can work cohesively in the multi-agent system while leveraging a variety of data sources and knowledge representations to interact effectively with both digital and physical environments.

So, there is still a lot of work needed to be done to make Gen AI from Information Retrieval (Thinker) to Doer (Intelligent Agents) who can sense and respond appropriately without any confusion in retrieved information which itself is in question in many scenarios where context is ambiguous as some time human mind is.?

So, what do you think, when Gen AI + Agentic Architecture will mirror the fantastic work that is been done in autopilot mode or the time horizon will keep evolving with the intersection of creativity and neurons?