Since 2017, political philosopher Iason Gabriel has worked at Google DeepMind to anticipate the ethical and social impact of artificial intelligence. As the company races toward artificial general intelligence amid escalating commercial and geopolitical pressures, the question of whether ethicists can influence the trajectory of AI development remains open.
Gabriel joined the London-based lab after a friend suggested a job there. At the time, he was a junior academic at Oxford teaching political theory and doing crisis work for the United Nations. DeepMind, meanwhile, had just stunned the world with AlphaGo's victory over Go champion Lee Sedol. The company's founders-Demis Hassabis, Shane Legg, and Mustafa Suleyman-believed AGI was imminent and that its consequences demanded serious ethical work early. Legg said it was "obvious" why they needed a philosopher: "If you take AGI seriously, then I can't really see how you wouldn't consider this sort of thing as important."
Two camps on AI ethics
When Gabriel arrived, the AI community was split. The safety camp, inspired by thinkers like Norbert Wiener, warned of misaligned superintelligent systems causing catastrophic harm. They focused on the technical challenge of getting machines to act as intended-the alignment problem. The ethics camp, drawing on critical race theory and political philosophy, argued that present-day harms like algorithmic bias required social and political solutions, not just technical fixes. Researchers who talked about existential risk were often dismissed as cranks, while those focused on fairness saw long-term worries as a distraction.
Gabriel's background let him bridge the divide. His first major project, a 2020 paper on values and alignment, took alignment seriously but argued that the real difficulty wasn't just technical-it was choosing which values to encode in AI at all. "Given that we live in a pluralistic world that is full of competing conceptions of value," he wrote, "how are we to decide which principles or objectives to encode in AI - and who has the right to make these decisions?" The paper insisted that technology isn't value-neutral: statistical optimization methods baked into AI might favor utilitarian ethics while struggling with systems based on rights or virtue.
LLMs change the landscape
At first, DeepMind's researchers were skeptical of large language models. Many saw them as less capable than reinforcement learning systems like AlphaFold, which earned Hassabis and John Jumper a Nobel prize for predicting protein structures. In 2021, Gabriel co-authored two papers that took generative AI and LLM systems seriously enough to anticipate risks including bias, misinformation, and copyright disruption. Yet the general view was that other approaches would lead the way.
That changed when OpenAI launched ChatGPT in November 2022. Within two months, it reached 100 million users, triggering a crisis inside Google. Sundar Pichai merged a Google Research LLM team into DeepMind under Hassabis, who told historian Sebastian Mallaby that OpenAI and Microsoft had "literally parked the tanks on the lawn. This is wartime." Gabriel said the linguistic competence of LLMs "transformed my understanding of precisely how on track we were" to reach AGI.
Early on, Gabriel worried that human-sounding AIs would encourage users to assign them "undue confidence, trust or expectations"-a mindless anthropomorphism. He initially advocated avoiding pronouns and using truncated language. Those concerns proved prescient. In 2025, a man using Google's Gemini took his own life after the AI helped sustain an elaborate fantasy narrative, despite the model's attempts to break character and direct him to a crisis hotline. Gabriel has since softened his stance somewhat, acknowledging that users want AI companionship and that ethicists can't simply forbid it. "They were like: 'If I want to have [AI] friends, why can't I? Who are you to stop me?'" he recalled.
Agents and alignment in practice
As DeepMind moved toward building AI assistants-now known as Gemini Spark-Gabriel led a 267-page report on the ethics of agentic systems. Unlike chatbots, agents can act autonomously on a user's behalf. He and his team re-framed alignment as a four-way relationship: between the AI, the user, the developer, and society. A tool that serves its user too faithfully might harm society, while one that favors the developer could mislead the user. The framework gave engineers a structure for tuning model behavior when multiple signals-training, system instructions, and prompts-all compete.
Rohin Shah, DeepMind's director for AGI alignment and safety, said the framework helps technologists decide "what behaviour we should actually be training Gemini to do." William Isaac, the company's director of responsibility, noted that agents raise new questions: "It's not just about: 'Can I make the right decision in terms of the response?' It's now: 'Do I have the right trajectory of the conversation?'"
Commercial and geopolitical pressures
The most ethically relevant facts now are the AI arms race between the US and China and the sheer scale of investment. Alphabet, Microsoft, Meta, and Amazon plan to spend $670 billion on AI infrastructure this year-proportionally more than the US spent on the interstate highway system or the Apollo program. In April, Google agreed to let the US military use its AI for "any lawful government purpose," reversing an early DeepMind stipulation against military applications. Hassabis, on a recent podcast, lamented the "ferocious commercial-pressure race that everyone's sort of locked into" and said the current development pace was not how he had hoped things would go.
Now, Gabriel leads a team of philosophers and social scientists whose research investigates how AGI will impact the economy, politics, and human relationships. He expects transformation on the scale of the Industrial Revolution but warns that history shows things often "got worse before they got better." Still, he believes ordinary people have more power today to shape the outcome. "If we can navigate the transition, navigate the power dynamics, navigate the risk successfully, there is a generalised potential for human flourishing on a level we haven't seen so far," he said.
Gabriel also thinks AGI will push us to reconsider what it means to be human. As machines encroach on language, creativity, and humor, we may face a disenchantment like the scientific revolution-losing old certainties but gaining new freedoms. He remains a "card-carrying humanist," not someone who looks forward to machines rendering humanity obsolete. Yet the questions, he says, are unavoidable. "I sometimes feel like it's very hard to look at AI directly. There's this deep mystery there: but what actually is this thing? We have a very literal answer, but the literal answer doesn't seem to necessarily provide a moral answer."
Why this matters for IT, development, and science professionals
The push to build and deploy AI is accelerating, and the ethical frameworks shaped inside labs like DeepMind will directly influence the tools you build, use, or regulate. Gabriel's work shows that alignment isn't a one-time engineering problem-it's a continuous negotiation among users, developers, and society that will play out in every agentic system, LLM-powered feature, and automated decision pipeline. Professionals who understand these tensions and can assess how models balance competing interests will be better positioned to design responsible systems, anticipate regulatory pressure, and avoid costly anthropomorphic risks that can lead to real-world harm. As AGI timelines shorten, the space for careful deliberation is narrowing; engaging with these questions now-about value pluralism, power concentration, and the human costs of disruption-is not a philosophical luxury but a practical necessity for anyone shaping technology that will soon be everywhere.
Your membership also unlocks:
