Charles Darwin has famously said, “It is the long history of humankind (and animal kind, too) that those who learned to collaborate and improvise most effectively have prevailed.” Teamwork and collaboration has been always a value that organisations, societies and people have held in high esteem.

Research has found that the benefit of teamwork and collaboration is also reflected by Artificial Intelligence. As efforts are on in the tech world to make AI more engaging and useful to humans, researchers are getting together teams of stand-alone Large Language Models (LLMs) into teams, called Multi-Agent Systems (MAS). So if you ask a MAS to write a detailed report on the stock report and ask it to suggest ways to improve it or play it, MAS would assign each AI agent in the team separate tasks, coordinate and build on the work of the agents, discuss and deliberate to find a solution that one of them would have not found alone. Sounds very familiar? We humans have been doing this for generations in every aspect of life. The MAS will be able to do all this without any human intervention or direction.

How do the LLMs communicate?  Because LLMs use written text for both their inputs and outputs, they can communicate among themselves independently. These teams work by feeding each agents solution to the others, to better the output. Many commercial uses of MAS are already being explored – medical consultation, legal opinion, military and other strategic decisions. The world that is emerging is exciting and already many tech giants are in the game. Microsoft released AutoGen, an open-source framework for building teams with LLM agents. Working on AutoGen an entrepreneur in Australia made a team of an image generator and a language model. This team fine tunes the image output to the original human prompt through collaboration.

Like with independent LLMs, the problem of bias, prejudice, false results or ‘hallucinations’ as they are called, could also be a part of the MAS teams output. In fact, one of the dangers being discussed is that the safeguards against harmful outputs built in independent LLMs could be circumvented by the team. The MAS teams could be used to attack and discredit a successful independent LLM. All this also sounds familiar, does it not?

Presently, GPT- 40, recently unveiled by OpenAI, can respond to you in a near human-like way. Based on your inputs, for example of food, suggest the best restaurants to go to. Soon enough, these AI agents working in teams would be able to suggest a comprehensive food trip covering places, how to reach there, make your bookings based on your budget and preferences!

Technology is working towards making the human experience effortless, burden less and easy. That is the premise behind this breakneck march of AI in our lives. However, it is also important to remember the evolutionary observation of Charles Darwin. ‘Those who learned to collaborate and improvise most effectively have prevailed’ It casts a different hue on the technological innovations happening in our present times.

Published in Lokmat Times on 30-6-2024