Swarm AI may sound a bit dystopian, but just like humans, AI agents often achieve the best results when they work together to solve problems.
Human teams build on the skills and experience of individuals to produce results that would otherwise be impossible. The same goes for the Swarm AI system. When multiple AI agents and people work together, these human-AI ecosystems can achieve outcomes that no single agent could ever accomplish.
Professor Hussein Abbas’ research explores the potential of human-AI ecosystems. His research considers how these ecosystems can be shaped, how they can be protected, and how human ethics can foster positive human-AI collaboration.
Working with multiple AI systems as part of a complex human-AI network can feel tedious, he says, but the results are worth the effort.
“Trying to solve complex problems presents many opportunities,” he says.
“In mathematics, there is a well-known concept called a ‘superadditive set.’ When certain conditions are met, a collection is always more than the sum of its parts. By working together, we can be more than, but not less than, the sum. That’s the swarm’s opportunity. As more humans and AI agents interact, that ecosystem can do bigger things with more powerful intelligence.”
Think of a flock as a team of sheepdogs working together to herd (of sheep, robots, data). Human “shepherds” work with sheepdogs who oversee flocks of sheep. A network of sheepdog agents, under the supervision of a shepherd, can achieve more sophisticated solutions than individual agents or humans. Each sheepdog may take on a different role to solve a larger problem, for example catching a lost sheep or directing the movement of a herd, or they may take on a similar role to achieve results faster.
One of the swarm AI applications Professor Abbas is working on is an unmanned aircraft management system for swarms of aircraft working together to clear birds from airport runways.
“What’s great about this problem is that it brings together a practical application that requires consideration of both human and animal ethics. Safety is paramount here, but it’s also a problem where the uncertainty of bird behavior brings surprises,” he says.
Education to build a trustworthy human-AI ecosystem
To take full advantage of swarm AI, humans must first learn how to manage and operate multiple agents simultaneously. Currently, AI training primarily focuses on the use of a single agent, but the ethical, technical, and safety challenges grow exponentially when interacting with multiple agents.
“AI is not just a computer science subject,” says Professor Abbas.
“AI education for humans in this ecosystem needs to span ethics, safety, digital engineering, and other areas. We need to create meaningful AI curricula that don’t require students to be experts in every field.”
Professor Abbas says this is not just a matter of curriculum design and educating people.
Education of AI systems, or machine education rather than machine learning, is also important. In this model, humans give swarm AI systems a structured curriculum based on the same principles used to train humans to work in these ecosystems. The system is then tested for reliability, safety, security, robustness, and ethical integrity with human teammates.
“We design curriculum for education and then build trust through rigorous testing, evaluation, verification, and validation. We’re asking, ‘How can we subject these systems to conditions to know if and when they will fail or become vulnerable?’
A common language and communication approach is essential for a positive human-AI ecosystem.
“My starting point is to understand how we can communicate with AI and how AI agents communicate with each other,” says Professor Abbas.
“You can build more trustworthy systems and interactions if you have confidence in common understanding from the beginning. By ensuring human-AI communication and creating a safe interaction space, you can reduce risk and mistakes.”
Professor Abbas points out that swarm systems and their super-additive benefits can create super-additive risks if they are only protected at the individual agent level.
“That’s why we don’t just protect a single AI system; we make sure the entire interaction space and ecosystem is also safe, ethically aligned, responsible, and secure.”
AI swarms are everywhere
Professor Abbas says that while AI is often discussed in terms of a single platform or agent, in reality it is rarely talked about as a single system or entity.
“What appears to be a single AI is working in parallel and interacting with many other AI systems and humans. In banking, no single agent does everything. In an ecosystem with humans, there are many software agents transacting on behalf of real humans.”
As the impact of AI expands across sectors, so does the need for a trusted, collaborative, and ethical human-AI ecosystem that combines strengths to deliver impactful, human-centered outcomes.
“We need to rethink the design of social systems. Many AI agents are now actively participating. We need to not only build smarter AIs, but also interact with these ecosystems, develop their skills, and test their trustworthiness.”
