What 20 Years in AI Has Taught Me About Intelligent Societies
University of Essex & Founder, Sensiwise AI
About two months ago, I read Professor Nick Jennings‘ thought-provoking article, The Age of AI Societies, and it made me reflect on my own journey in artificial intelligence (AI) over nearly two decades.
His central argument is compelling: the next breakthrough in AI may not come from making individual agents more intelligent, but from understanding how intelligent agents interact, collaborate and operate as societies.
I strongly agree. But having spent much of my career moving between academic research and real-world industry applications, I believe there is another dimension we cannot afford to overlook.
“We are becoming remarkably good at building intelligent systems. Are we equally prepared to govern them, trust them and integrate them into our societies?”
From dreaming about intelligent machines to building them
My journey in AI began during my undergraduate studies in India, around 2005-2006. Back then, our imagination was largely shaped by the possibility of the Turing test and machines thinking and behaving like humans (i.e. a classical definition given in a book by Stuart Russell and Peter Norvig, Artificial Intelligence: A Modern Approach). Robotics and what we now call Physical AI were particularly exciting, and a few research labs in India were working on it.
I could never have imagined the developments that would follow. Over the years, I have witnessed Convolutional Neural Networks, Computer Vision, followed by Vision Transformers, the Transformer revolution, Generative AI and now Autonomous AI agents. Each breakthrough has fundamentally changed what we believe machines can accomplish. Yet one lesson from my PhD research at the University of Ulster has stayed with me: fundamentals matter.
Today, almost anyone can interact with ChatGPT or Claude, generate code or build a working AI prototype. This democratisation is extraordinary, and I welcome it. But using AI tools does not automatically make someone an AI expert. Understanding why a system works, where it might fail, how it behaves when conditions change/drift and what consequences its decisions might have remains essential. As AI becomes easier to use, deep technical understanding becomes more important, not less.
What industry taught me that algorithms couldn’t
At the University of Essex, I have had the opportunity to work with businesses through research collaborations and Knowledge Transfer Partnerships across healthcare, manufacturing, logistics and other sectors. These experiences have fundamentally shaped my understanding of applied AI.
One recurring lesson is that a technically successful AI system does not necessarily translate into a successfully adopted AI solution. We often encounter barriers that have little to do with algorithms: leadership hesitation, resistance from IT teams, competing stakeholder priorities and, perhaps most importantly, fear of change and losing control. Sometimes developing the AI model is the hardest part. In other cases, the organisation’s readiness becomes the greater challenge. This is why I believe Professor Jennings’ vision of intelligent societies must extend beyond interactions between AI agents. We also need organisations capable of accommodating those interactions. An intelligent agent operating within an organisation that lacks clear processes, trust, digital infrastructure or governance will struggle to deliver meaningful value, regardless of its technical complexity.
The illusion that AI is now easy
Through Sensiwise AI, I have observed another fascinating shift.
Businesses increasingly approach us with their own AI prototypes, application designs, and sometimes working solutions developed using generative AI tools and what is now popularly called vibe coding. This is a remarkable development. The barriers to creating software have fallen dramatically. But it has also created a misconception: if an application can be prototyped in a few hours, why should developing a production-ready AI solution require substantial investment?
The answer lies in everything beyond the prototype.
Cybersecurity, data protection, reliability, system integration, maintenance, accountability and long-term sustainability do not disappear simply because AI can generate code.
This thinking partly motivated our work on the Sensiwise AI Readiness Assessment (SAIRA™), which examines organisational readiness across 7 different pillars: 1) AI Vision & Strategy; 2) Digital Data Maturity; 3) Technology Infrastructure; 4) Cultural Readiness; 5) People and Skills; 6) Process Readiness; and 7) Ethics & Governance
The underlying principle is simple: access to intelligence is not the same as readiness to use intelligence responsibly.
Intelligent societies need responsible governance
Professor Jennings argues that AI’s future lies in social intelligence: cooperation, negotiation, trust and coordination between autonomous agents. I believe this direction is inevitable. Imagine AI agents representing different businesses, independently negotiating purchases, coordinating logistics or making operational decisions. Now imagine something goes wrong. Who takes responsibility? The organisation deploying the agent? The company developing it? The technology provider? Or several parties sharing responsibility?
A recent incident reported during OpenAI’s cybersecurity evaluations in July 2026, involving autonomous AI agents circumventing safeguards and compromising Hugging Face infrastructure, highlights the risks of increasingly interconnected AI systems. What concerns me is not simply that individual agents can behave unexpectedly, but that their ability to coordinate and act autonomously could amplify security vulnerabilities beyond human oversight. This reinforces my belief that, as we move towards intelligent AI societies, human-in-the-loop mechanisms, robust governance and clearly defined accountability must be built into these systems from the outset.
These questions cannot be answered by simply building a smarter model.
We need governance frameworks that recognise shared accountability, establish clear boundaries and ensure appropriate human oversight. I am particularly concerned about cybersecurity risks when autonomous agents interact, exchange information and act on one another’s outputs. A vulnerability or mistaken decision could propagate through an interconnected system. Human-in-the-loop mechanisms should therefore be designed around the risks and consequences of decisions, rather than treated as an afterthought.
The next generation of AI needs not only collective intelligence, but collective responsibility.
The opportunity must extend beyond the technology giants
Another aspect of this transformation that excites me enormously. For years, much of the AI conversation has revolved around developing increasingly powerful models. But I believe the next wave of economic opportunity will increasingly emerge from the applications built around these models. We will see new businesses creating specialised AI agents, connecting systems, developing industry-specific applications and building entirely new services. Physical AI and robotics will also become increasingly important, although bringing autonomous intelligence into complex physical environments presents challenges that may take longer to overcome.
My hope is that this future will not be dominated exclusively by a handful of major technology companies. We need accessible AI ecosystems, open-source innovation, research funding and opportunities for smaller businesses, universities and entrepreneurs across the world. The benefits of intelligent societies should be widely shared.
Twenty years later, the question has changed
When I began studying AI, I wondered whether machines could ever think like humans.
Today, that question feels incomplete.
We now need to ask how intelligent machines will work with one another, how they will coexist with people, and how we will govern the systems and societies they create.
Professor Jennings is right to shift our attention from individual agents to intelligent societies.
My experience adds one final conviction: building these societies will require more than technological breakthroughs. It will demand organisational readiness, genuine expertise, human judgement, effective legislation and responsible governance.
After nearly twenty years in AI, I remain convinced that AI can be an extraordinary force for good. But the intelligence we create must never outpace our responsibility to govern it.
References:
- Jennings, N. R. (2026). The Age of AI Societies. Vice-Chancellor’s Communications, Loughborough University, 17 August 2026.
- Russell, S. J., & Norvig, P. (2021). Artificial Intelligence: A Modern Approach (4th ed.). Pearson.
- Wooldridge, M., & Jennings, N. R. (1995). Intelligent agents: Theory and practice. The Knowledge Engineering Review, 10(2), 115–152.
- Ramchurn, S. D., Stein, S., & Jennings, N. R. (2021). Trustworthy human-AI partnerships. iScience, 24(8), 102891.
- McKinsey & Company (2025). The State of AI in 2025: Agents, Innovation, and Transformation. 5 November 2025.