Does AI regulation help or hurt innovation? Lessons from Lord Tim Clement-Jones
7 min. read&w=3840&q=75)
Lord Tim Clement-Jones, member of the House of Lords.
Lord Tim Clement-Jones has worked on how to govern AI since 2016, long before it became a dinner-table topic. He co-founded the UK Parliament’s All-Party Parliamentary Group on AI, chaired the House of Lords Select Committee on AI, and wrote the book Living with the Algorithm.
Today he’s a member of the House of Lords focused on AI and digital policy, and a self-described conditional optimist. AI can be the best thing that ever happened to us, as long as it stays our servant and never becomes our master.
We got into his AI kill switch proposal, what South Korea gets right about AI governance, how standards can keep pace with a moving target, and what AI means for jobs and young people.
Key takeaways#
Does AI regulation slow down innovation?#
Regulation speeds up AI adoption, because it builds the public trust that wide-scale use depends on. Tim’s most unpopular opinion among policymakers is exactly this. The business people he meets want a level playing field, clear liability if something goes wrong, and something they can insure, and insurers are asking for regulation too. Skip trust and you get backlash: Gen Z is growing wary of AI, and mistrust of data centers is rising.
For risk and legal: before your next AI rollout, write down who is liable if it goes wrong. If nobody can answer, that’s your first guardrail.
How can AI regulation keep up with the speed of AI?#
Through flexible standards on top of a thin layer of law. Acts of Parliament take years, and secondary regulation is faster but still slow. Standards can be updated by agreement between standards bodies. Tim’s model is legislation that mandates standards for risk assessment, testing, training, monitoring, and transparency. ISO 42001, the international standard for managing AI systems inside an organization, already sits inside EU AI Act conformity assessment and is gaining ground in UK corporate governance.
For CTOs: check how far your organization is from ISO 42001, and treat the gaps as your AI governance backlog.
What is the difference between human in the loop and human in command?#
Human in command means a person decides why and how AI is used and can show how they checked its output. A human in the loop easily becomes a tick box: reviewed, big tick, done. For high-risk use, Tim wants a trail you can audit: how the AI was prompted, what came out, and how a person verified it. Lower-risk tools, like a recommender system, can stay light. Agentic AI, meaning systems that plan and take actions on their own across tools, is pushing more work into the high-risk zone.
For product teams: on agentic pilots, log the prompt, output, and reviewer for every high-risk action, and spot-check ten of them each month.
How should companies decide where to use AI?#
Start with purpose: how does this deployment benefit the humans involved? Tim asks whether each use of AI supports human flourishing and improves life chances. A business that can answer that knows why it’s adopting AI, earns the trust of its workforce, and avoids a new wave of resistance against the technology.
For budget holders: add one line to every AI business case: who benefits, and how would they notice?
Which country is getting AI governance most right?#
South Korea, in Tim’s view. It introduced a basic AI act similar to the EU AI Act and paired it with a digital bill of rights. One side covers AI development, safety, and ethics. The other gives people redress when AI affects them. Tim argues the two need to go together.
For the board: check whether your AI policy covers both sides: rules for your teams, and a clear way for customers or employees to challenge an AI decision.
What is an AI kill switch?#
An AI kill switch is the power to stop an autonomous AI system, for example by cutting its access to data centers. Tim proposed one as an amendment to the UK’s Cyber Security and Resilience Bill. He calls it a headline grabber on purpose: his core ask is for the AI Security Institute to inspect models and enforce standards, with the kill switch as the backstop. Its weak spot is borders. A capable agent can move to a data center in another country, so it only works with international agreement.
For security teams: name the person who can switch off each AI agent you run, and test that they can do it within the hour.
Will AI take entry-level jobs?#
It is already thinning out entry-level professional roles, and new jobs will arrive with a lag. Tim uses two photos of Fifth Avenue: horses and buggies in 1903, almost only cars in 1913. The farriers took years to become mechanics. Blue-collar work is exposed too, now that an amateur with a phone camera and AI can be talked through a repair step by step.
For L&D leaders: list the entry-level tasks AI now does across your teams, and redesign how juniors learn those skills.
What is cognitive offloading, and how do you prevent it?#
Cognitive offloading is handing your thinking to AI until you can no longer judge whether its output is right. It is the near-term risk Tim focuses on most. He wants AI used in schools and universities, alongside exercises done without it, so critical thinking stays sharp. Use it or lose it applies to the brain as much as the body.
For team leads: keep one no-AI ritual in your team, like writing the first draft of a strategy before any model sees it.
What skills should young people build in the age of AI?#
Judgment, creativity, critical thinking, and above all drive. Tim points to Daniel Susskind’s book What Do I Tell My Children? and its advice to follow your passions: a person with a mission pursues a goal more relentlessly than any machine. AI skills become a given, because AI will be a coworker in almost every job.
For HR leaders: in your next graduate hiring round, ask candidates what they would pursue even if nobody paid them.
What does proportionate regulation look like in practice?#
Sometimes it means removing rules. The best thing Tim has delivered is a private member’s bill that deregulated live music in pubs and restaurants. Venues for up to 500 people can now host live music without a license, where they once needed a magistrate’s approval, and live music in pubs and clubs has flourished since.
For leadership teams: find one approval step in your organization that slows down small AI experiments, and raise its threshold.
Listen & watch#
DELIVERED is a production of Infinum and Your Majesty.