Google CEO Sundar Pichai wrote in January 2020 that artificial intelligence needed to be regulated because it was “too important not to.” His argument did not reject oversight; it called for rules that addressed harms without unnecessarily limiting social and economic benefits.
The intervention reflected a turning point. AI was moving from research laboratories into health, advertising, transport, public services and security, making a purely voluntary approach increasingly difficult to defend.
Agreement on Regulation Did Not End the Debate
Many policymakers, companies and civil-society groups agreed that AI needed governance. The difficult questions concerned scope, enforcement and who bore responsibility when a system caused harm.
Weak rules could allow discrimination, manipulation or unsafe automation. Overly vague or burdensome rules could favour the largest firms, which had the resources to navigate them, while making experimentation harder for smaller organisations.
Good regulation needed to be precise enough to protect people and flexible enough to adapt with technology.
The Internet Offered a Cautionary Example
Pichai noted that the internet connected people and information at extraordinary scale but also enabled misinformation. The comparison showed why technical capability did not guarantee a beneficial outcome.
AI could improve diagnosis, accessibility and energy use. It could also amplify falsehoods, automate unfair decisions or concentrate power in a small number of platforms.
Governance had to consider how a model was deployed, who controlled it and which groups faced the consequences of error.
Risk-Based Rules Offered a Useful Direction
High-impact applications deserved stronger requirements than low-risk tools. A spelling assistant and a system influencing medical treatment or employment should not face identical obligations.
Risk-based regulation could require testing, documentation, human review and independent audit in sensitive settings. Lower-risk innovation could operate under lighter rules, provided it did not create hidden harm.
This approach focused attention on consequences rather than treating “AI” as one homogeneous category.
Google’s Principles Needed External Accountability
Pichai pointed to Google’s AI principles: social benefit, fairness, safety, accountability, privacy, scientific excellence and limits on harmful or abusive use. The company had adopted them after employee backlash over Project Maven, a Pentagon drone-analysis contract.
Corporate principles could guide internal decisions, but they were not enforceable public law. Their credibility depended on transparent application, independent challenge and willingness to decline profitable work.
Regulation could establish a baseline that did not vary with the values of each individual company.
Europe and the US Favoured Different Emphases
At the time, European institutions were developing comparatively comprehensive AI proposals, while US officials warned against rules they considered innovation-killing. The UK sought a distinctive position emphasising ethical AI and national research strengths.
These debates were often presented as a choice between safety and competitiveness. In practice, trustworthy systems could support adoption and create long-term advantage.
The real trade-off was between carefully designed obligations and vague slogans that left both innovators and the public uncertain.
Technical Literacy Was Essential for Policymakers
Rules needed to address data quality, model drift, explainability and cybersecurity without demanding impossible guarantees. Policymakers did not need to write code, but they needed independent technical expertise.
Companies also needed to explain systems clearly rather than treating complexity as a reason to avoid scrutiny. A shared factual foundation made proportional regulation more achievable.
Public agencies had to build capacity so they were not dependent entirely on the firms they oversaw.
People Needed Rights and Remedies
Regulation mattered most when something went wrong. Individuals needed to know when an automated system affected them, understand the basis of a consequential decision and challenge it through an accessible process.
Data protection, anti-discrimination law and consumer protection already offered some tools. AI-specific rules could strengthen them where scale and opacity created new gaps.
An oversight framework without practical remedy would do little to build trust.
Innovation Benefited From Clarity
Developers often wanted clear rules before investing in a product. Defined standards for testing, documentation and data use could reduce uncertainty and prevent a race to the bottom.
Regulatory sandboxes and open standards could help organisations test novel approaches under supervision. They needed guardrails, however, so that “experimentation” did not become an excuse to expose people to avoidable risk.
Responsible innovation meant learning before harm became irreversible.
The Call Was a Starting Point, Not a Solution
Pichai’s statement recognised that voluntary company commitments were insufficient for technology with broad societal impact. It did not settle what sensible regulation looked like, particularly given Google’s own influence and record.
The lasting lesson was that AI policy required balance, but balance did not mean inaction. Effective rules could protect dignity, safety and democratic accountability while allowing useful systems to develop.
The test was whether regulation gave people real protection and gave builders clear responsibility—not whether it sounded permissive or strict in the abstract.
