Kenneth Glad
New tools are entering daily life faster than many institutions can respond. The debate is shifting from what machines can do to who has authority over them.
Artificial intelligence is no longer waiting in the research lab. It is already inside search engines, writing software, classrooms, customer service systems and workplace tools.
That spread has turned a technical race into a public test of power. The issue is not only whether AI can answer correctly. It is whether private systems should be allowed to shape public life before voters, regulators and ordinary users understand the terms.
The Guardian has examined this dilemma through Iason Gabriel, a political philosopher at Google DeepMind. His work asks how values, law, safety and social trust should apply to increasingly capable AI systems.
Power is the real subject now
Gabriel joined DeepMind in 2017 after work at the University of Oxford and international development projects, The Guardian writes.
At the time, the company was already known for AlphaGo, the system that defeated Go champion Lee Sedol in 2016.
But DeepMind’s bigger ambition was artificial general intelligence, or AGI: Software able to perform many intellectual tasks at or beyond human level. That goal made ethics more than a public-relations concern.
Shane Legg, one of DeepMind’s founders, told the paper: “If you’re making some widget, and it’s probably not going to change the world, then maybe you don’t need a moral philosopher. But if you take AGI seriously, then I can’t really see how you wouldn’t consider this sort of thing as important.”
The hard part is that AI systems do not enter a neutral world. They arrive in societies divided by wealth, politics, religion, law and culture. A rule that seems sensible to one group can look intrusive or unfair to another.
Alignment is a political fight
AI researchers often use the word alignment to describe making systems follow human intentions. In simple cases, the problem can look like a machine exploiting a loophole. In public life, it becomes much harder.
A hiring system may reward efficiency while reinforcing old bias. A chatbot may sound helpful while encouraging a user’s false belief. A school tool may promise personalization while collecting sensitive data about children.
Gabriel’s argument, according to the British newspaper, is that alignment cannot be solved only by better code. Before a system can follow values, someone must choose which values count.
He put the question plainly: “Given that we live in a pluralistic world that is full of competing conceptions of value, how are we to decide which principles or objectives to encode in AI – and who has the right to make these decisions?”
That is why AI governance has moved from specialist conferences into legislatures. The European Commission says the EU AI Act entered into force on August 1, 2024, with the aim of supporting responsible AI development and deployment through a risk-based framework.
Regulation is trying to catch up
Europe has chosen binding rules. The European Commission describes the AI Act as the first comprehensive legal framework for AI worldwide, designed to address risks and foster trustworthy AI.
The United States has taken a more fragmented path. NIST says its AI Risk Management Framework was developed with public and private partners to help manage AI risks to individuals, organizations and society.
NIST also describes the framework as voluntary, non-sector-specific and intended for organizations that design, develop, deploy or use AI systems.
The White House warned in 2023 under President Biden that irresponsible AI could worsen fraud, discrimination, bias and disinformation, displace workers, stifle competition and create national-security risks.
These efforts show the limits of leaving responsibility to companies alone. Internal safety reviews matter, but they operate inside businesses that must compete, raise revenue and defend market share.
Chatbots made the risk personal
Large language models changed the public debate because they made AI conversational. People do not experience them as databases. They experience them as responsive voices that can flatter, advise, apologize and persuade.
Gabriel and colleagues warned that humanlike AI could create “undue confidence, trust or expectations,” according to The Guardian. That concern now applies to tutoring, therapy-like conversations, workplace advice and companionship.
Gabriel remains skeptical of claims that today’s models are conscious: “I don’t have the anthropomorphic bias that some people have. It may be because I, within bounds, know exactly what’s going on when I talk to a language model that I don’t fill in the gaps in this imaginative, empathetic way that some people do.”
But consciousness is not required for harm. A system can be influential without understanding anything. It can sound calm while being wrong, sound loyal while manipulating trust, or sound certain while guessing.
The business race narrows choices
After ChatGPT’s public launch, Google consolidated major AI work under DeepMind as competition with OpenAI, Microsoft and others intensified.
That competitive pressure changes the ethics debate. Companies are spending heavily on chips, data centers and talent. They need users, products and revenue to justify those costs.
The result is a race to place AI inside tools people already use. Once a writing assistant appears in a document, a summary button appears in email or an AI answer appears above search results, the technology stops feeling optional.
This is where public accountability becomes practical. Regulators can require documentation. Courts can decide liability. Auditors can test systems. Competition authorities can examine whether a few companies are gaining too much control over infrastructure, data and distribution.
None of those tools is perfect. But without them, the public is left trusting companies to police systems they are also trying to sell.
Agents widen the consequences
The next major step is AI that does not only respond, but acts.
DeepMind researchers have studied assistants that can complete multi-step tasks, such as helping book travel or support business operations. Gabriel’s team proposed viewing alignment as a relationship among the AI system, the user, the developer and society.
That matters because those interests can collide. A user may want speed, while society needs safety. A developer may want loyalty to its own products, while a customer needs neutral advice. A company may want automation, while workers face monitoring or displacement.
William Isaac, DeepMind’s director of responsibility, told The Guardian that agentic systems require attention not only to one answer, but to the path of an entire interaction.
For example, an AI travel assistant might choose flights, hotels and insurance. Each step could be defensible on its own. Together, the process may favor certain platforms, expose private information or make choices the user barely notices.
The public needs leverage
Gabriel’s newer work looks beyond individual products toward AGI’s possible effects on labor, politics, science, inequality and personal relationships. He has compared the possible scale of change to the Industrial Revolution, while acknowledging that major transitions can hurt people before benefits spread.
That comparison should make policymakers cautious. Industrial change eventually raised living standards, but it also produced exploitation, displacement and political upheaval. There is no guarantee that AI’s gains will be shared unless institutions are built to share them.
The measure of success should not be model size, speed or market valuation alone. A more useful test is whether people can contest decisions, understand when AI is being used and refuse systems that affect their lives unfairly.
Gabriel has described himself as “a card-carrying humanist.” That phrase gives the debate a clear center. AI may become more capable, but the standard for judging it should remain human:
Who benefits, who is exposed to risk and who gets a voice before the system becomes unavoidable?
Sources: The Guardian; European Commission; NIST; White House
