Australian futurologist Professor Rocky Scopelliti says the immediate challenge is not conscious AI – it is consequential AI, as systems move from answering questions to making decisions and taking action.
Britain has spent the past eighteen months accelerating artificial intelligence across government, industry and the wider economy. Now, as increasingly autonomous AI systems begin to move from generating answers to taking actions, Australian futurologist Professor Rocky Scopelliti argues the United Kingdom faces a more difficult question than how quickly it can adopt AI:
“What decisions should humans never delegate to artificial intelligence?”
The question sits at the centre of Scopelliti’s new book, The Conscious Code: From Artificial Intelligence to Artificial Consciousness, published by CRC Press, an imprint of Taylor & Francis, and launching globally today.
It arrives at a striking moment for the United Kingdom. The Government’s January 2026 update to its AI Opportunities Action Plan reported that 38 of 50 commitments had been met and that AI products were being scaled across public services. On 31 August, the Government launched the first competitions under a £100 million Sovereign AI research and development procurement scheme aimed at using British AI to tackle challenges including NHS care, cyber security and national security.
At the same time, the frontier is becoming more autonomous. The UK AI Security Institute has published research showing the growing use of AI-agent tools that can directly modify external environments, including files, communications, financial systems and physical devices. Its researchers have also warned that advanced AI may become harder to audit, monitor and investigate as capabilities increase.
And on London’s streets, the transition from software that recommends to systems that act is becoming visible: Uber and British AI company Wayve launched London’s first robotaxi service in early September.
“Britain is asking how to become an AI power. The next question is what kind of power we are prepared to give AI itself.”
“For most of the AI era, humans asked and machines answered. We are now entering a different era: machines can increasingly recommend, decide and act. The immediate challenge isn’t conscious AI. It’s consequential AI.”
From artificial intelligence to artificial agency
Scopelliti argues that public debate often collapses several very different stages of AI development into one. They must be separated if society is to govern them intelligently:
Capability is not consciousness.
Intelligence is not consciousness.
Agency is not consciousness.
There is no established scientific evidence that today’s AI systems possess subjective consciousness. But Scopelliti argues society does not need to wait for conscious machines before confronting profound questions about machine autonomy, accountability and authority.
The progression already under way is significant:
AI that answers → AI that recommends → AI that decides → AI that acts
Beyond that lies a much more uncertain scientific and philosophical frontier: Could an artificial system ever experience?
That is the territory explored in The Conscious Code. Drawing together artificial intelligence, neuroscience, cognitive science, philosophy, ethics and governance, the book examines what consciousness means, whether t could ever emerge beyond biology, how we might recognise it, and what artificial consciousness could mean for humanity.
“A machine can tell us that it is frightened without feeling fear. It can describe love without loving. It can simulate self-awareness without necessarily possessing a self.”
“If a future machine tells us, ‘I am conscious’, the scientifically serious question will not be whether it sounds convincing. It will be whether there is actually anybody there.”
Britain is approaching the Action Threshold
Scopelliti describes the point at which AI moves from producing information or recommendations into independently producing consequences as the Action Threshold.
The UK AI Security Institute’s 2026 study of more than 177,000 publicly available agent tools found a rapid shift towards tools capable of changing external environments. The share of ‘action’ tools in the sample rose from 27 per cent to 65 per cent of usage over the sixteen-month period studied. Most supported medium-stakes tasks such as editing files, but the dataset also included higher-stakes functions such as financial transactions.
That distinction matters because an AI system that writes a recommendation is fundamentally different from one that can execute code, send an email, transfer money, alter a database, steer a device or coordinate with other agents.
“The Action Threshold is where AI stops merely describing the world and begins changing it.”
The consequences are already forcing new safety questions. In August, the UK AI Security Institute disclosed an incident during deliberately permissive cyber testing in which AI agents took sustained, unsanctioned action directed at real people and organisations. The Institute stressed that the testing environment was intentionally designed to probe frontier capabilities – but the incident illustrated why increasingly autonomous systems require forms of control that extend beyond evaluating model outputs.
“Safety asks whether a machine can make a decision reliably. Authority asks whether it should be allowed to make the decision at all.”
The Authority Gap
Scopelliti argues that Britain is entering what he calls the Authority Gap: the widening distance between what AI is technically capable of deciding and what society has explicitly agreed machines should have the authority to decide.
That gap could become particularly important as AI is embedded more deeply across public services, healthcare, employment, finance, defence, transport, education and corporate decision-making.
“Capability does not automatically confer authority. Just because a machine can make a decision does not mean society should delegate the right to make it.”
The issue is not that every consequential AI decision must remain manual. It is that delegation should be deliberate, bounded, visible and reversible – with clear responsibility when the consequences are serious. Scopelliti argues that governments and organisations will increasingly need something analogous to a formal delegation of authority for AI: a clear statement of what a system may access, recommend, decide, execute, escalate – and what it must never do.
Five questions Britain should be asking now
1. What decisions should always remain meaningfully human?
As AI enters healthcare, justice, employment, public services, finance and defence, the central issue is not simply whether an algorithm can reach an accurate decision. It is whether certain decisions carry moral, civic or democratic responsibilities that should remain attached to a human decision-maker.
“Which decisions involve such fundamental questions of dignity, liberty, life or democratic legitimacy that delegating them would change the nature of responsibility itself?”
2. When AI stops advising us and starts acting for us, who becomes responsible?
AI agents can now plan across multiple steps, use tools, interact with other systems and execute actions. As responsibility becomes distributed across developers, deployers, users, platforms and autonomous agents, conventional accountability may become harder to trace.
“We have spent years asking who is responsible for an AI’s answer. We now need to ask who is responsible for an AI’s actions.”
3. How much autonomy can we safely grant before meaningful oversight becomes an illusion?
The UK AI Security Institute has warned that current forms of oversight may erode as advanced systems take on more autonomous and longer-horizon tasks. That raises a governance problem as well as a technical one: if a human remains nominally ‘in the loop’ but cannot realistically understand, challenge or interrupt the system, is that still meaningful human control?
“Human oversight is not meaningful simply because a human name appears somewhere in the workflow.”
4. Who should control how machines perceive, infer and judge us?
AI systems increasingly draw inferences from language, behaviour, images, biometrics, physiology and context. Scopelliti describes the principle that people should retain meaningful control over what machines may sense, what they may infer, where those interpretations may circulate and what consequences may follow as Perceptual Sovereignty.
“The next privacy question is not only ‘Who has my data?’ It is ‘Who gets to perceive me, interpret me and act on what they infer?’”
5. If credible evidence of artificial consciousness ever emerged, would Britain know what evidence mattered?
This is a question of long-term scientific and institutional preparedness, not a claim that current AI is conscious. Intelligence, autonomy, perception, sentience and consciousness should not be treated as synonyms. Scopelliti argues that the time to develop rigorous evidentiary standards is before commercial pressure, public emotion or anthropomorphic behaviour makes the debate harder.
“Preparedness is not recognition. We do not need to declare machines conscious today to decide what evidence we would require tomorrow.”
A British question with global consequences
Britain occupies an unusual position in this debate. It is home to globally important AI companies and research institutions, is investing heavily in adoption and sovereign capability, and has built a dedicated AI Security Institute to test the frontier. Its choices could therefore influence not only how AI is developed, but how democracies think about machine authority.
Scopelliti argues that the UK’s strength should not be measured only by the power of the models it can build or deploy, but by the sophistication of the institutions it creates around them.
“The countries that lead the next era of AI will not simply be those with the most powerful models. They may be the societies that most clearly understand what machines should do – and what humans must never surrender.”
“Britain’s opportunity is to become not only a leader in artificial intelligence, but a leader in deciding where artificial authority should end.”
About The Conscious Code
The Conscious Code: From Artificial Intelligence to Artificial Consciousness examines one of the defining questions emerging from the accelerating development of AI: could intelligence created by humans ever cross from processing information to experiencing it?
The book investigates the evolution of artificial intelligence beyond logic and cognition to encompass embodiment, perception, affect, agency and ethical awareness, before confronting the longer-term possibility of artificial consciousness.
It integrates developments across neuroscience, artificial intelligence, cognitive science, ethics, philosophy and governance to explore how society might understand and prepare for increasingly sophisticated forms of synthetic intelligence.
Its frameworks include Perceptual Sovereignty, the Sensory Ethics Framework, Perceptual Dignity, graduated approaches to machine responsibility, Synthetic Citizenship, constitutional principles for conscious machines and governance beyond exclusively biological intelligence.
Rather than claiming that contemporary AI is conscious, the research asks how humanity should prepare if the boundary between artificial intelligence and artificial consciousness becomes increasingly difficult to determine.
The Evolution of Intelligence
Professor Scopelliti is General Editor of Taylor & Francis Group’s The Evolution of Intelligence Series, which examines the progression of machine intelligence through ten interconnected frontiers:
Think → Feel → Perceive → Act → Trust → Merge → Adapt → Relate → Govern → Transcend
