The following was developed from a recent seminar which I delivered.
This morning I want to speak about artificial intelligence … but not as a technical topic. Nor primarily as a policy issue. But as a question of who acts, who decides, and who is answerable when systems act in our name.
For the last few years, we have asked a particular set of questions about artificial intelligence: is it accurate? is it biased? Can we trust its outputs?
These remain important. But I don’t think that they are now the most important questions. Because something has shifted.
We are moving from a world where Ai advises to a world where Ai acts. And when something acts in your name, the moral universe changes.
This is not speculative.
- Google is building systems it describes as personal Ai agents, able to operate across daily life (Google AI, 2026a).
- At Google I/O, it launched tools capable of planning and executing multistep tasks, not simply answering queries (Google AI, 2025; DataCamp, 2026).
- Microsoft is redesigning Copilot so it can coordinate workflows, send communications, and act across applications (Microsoft, 2026a; Microsoft, 2026b).
- Apple has moved to Ai systems that operate across apps, able to carry out tasks rather than simply respond (Apple, 2024; Heater and Silberling, 2025).
What binds all of this together is simple: The goal is no longer just intelligence. The goal is agency.
From assistance to action
Up until now, most of our interaction with Ai has been conversational. You ask. It answers. You decide. You act.
But the systems now being designed alter that sequence: You ask. The system plans. The system executes. The system acts.
And that raises a fundamentally different question: If a system acts, whose action is it?
Why this matters for care
In many sectors, that question is complex. In social care, it is foundational. Because care has always been about judgement exercised in the name of another person. Every act of care involves interpretation, discretion, responsiveness and responsibility. And above all answerability. If something goes wrong in care, we know how to ask the question: Who made this decision? Why was it made? Can it be challenged?
The danger in the current moment is not that Ai will become capable. It already is. The danger is that responsibility becomes diffused. Spread across systems, workflows, algorithms, platforms until no single person can say: “This was my decision.”
When that happens, we have not improved decision-making all we have done is to have weakened accountability.
There is another dimension to this. Human decisions are limited in scale. System-based decisions are not. An individual error affects one person. A system-level error may affect thousands or millions, before it is even noticed.
And this is precisely the trajectory of current development. Systems acting across multiple services. Automated decisions chaining into further actions. Outputs becoming inputs to other processes The issue is no longer just whether a system is right or wrong. It is whether we can trace, interrupt, and account for what it does.
At the same time, we are witnessing an extraordinary concentration of power. A small number of global technology companies are investing vast sums, indeed hundreds of billions, to build the infrastructure that will underpin these systems. This is not only a race for innovation. It is a race to define the platforms, the rules and the environments in which AI operates.
So, when we ask who acts, we must also ask: Whose system is acting? On whose terms? Under whose control?
Let me put this as clearly as I can. We are not just delegating tasks to Ai. We are delegating prioritisation, classification, escalation and decision-making pathways. In other words, we are delegating authority. And delegation without clear boundaries is not delegation. It is abdication.
This is where I think care has something critically important to say. Because care has never treated authority casually. In care authority is defined, authority is limited, authority is supervised, authority is always open to challenge. And crucially authority is never separated from the person who holds it.
So, what must guide us now? Not technical possibility. Not even economic efficiency. But a set of very clear principles:
- There must always be a named human accountable
- There must always be a way to challenge a system decision
- There must always be a possibility to reverse or override
- There must always be clarity about who authorised the system to act.
If those are absent, then whatever we have built, it is not a support to human decision-making. It is a substitution for it.
So, the question before us is not, will Ai become more powerful? It will. The question is not will Ai change social care? It already is. The real question is this: As systems begin to act, how do we ensure that responsibility remains human?
Technology will continue to evolve. Systems will become more capable. Actions will become more automated. But there is one thing that must not change.
When a decision affects a life, when an action shapes a person’s experience, there must always be someone who can say: “I am responsible for this.”
If we can preserve that then Ai can serve us well. If we lose it, then we will have gained efficiency but lost something far more fundamental. Not capability. But accountability.
And without that, there is no care.
Donald Macaskill
References:
Google AI (2026a) Gemini Spark. Available at: https://gemini.google/overview/agent/spark
Google AI (2025) Gemini Agent overview. Available at: https://gemini.google/to/overview/agent
DataCamp (2026) ‘Gemini Spark: Google’s always on AI agent explained’, DataCamp Blog. Available at: https://www.datacamp.com/blog/gemini-spark
Microsoft (2026a) ‘Copilot Cowork: A new way of getting work done’, Microsoft 365 Blog, 9 March. Available at: https://www.microsoft.com/en-us/microsoft-365/blog/2026/03/09/copilot-cowork-a-new-way-of-getting-work-done
Microsoft (2026b) ‘Introducing a new design for Microsoft 365 Copilot’, Microsoft 365 Blog, 28 May. Available at: https://www.microsoft.com/en-us/microsoft-365/blog/2026/05/28/introducing-a-new-design-for-microsoft-365-copilot
Apple (2024) ‘Introducing Apple Intelligence for iPhone, iPad, and Mac’, Apple Newsroom, 10 June. Available at: https://www.apple.com/newsroom/2024/06/introducing-apple-intelligence-for-iphone-ipad-and-mac
Heater, B. and Silberling, A. (2025) ‘Apple Intelligence: Everything you need to know about Apple’s AI model and services’, TechCrunch, 9 September. Available at: https://techcrunch.com/2025/09/09/apple-intelligence-everything-you-need-to-know-about-apples-ai-model-and-services
