This episode of “Impact of AI: Explored” is a fast‑paced, honest conversation between us (Gerjon and James) and AI governance specialist Louise Humpington (https://www.linkedin.com/in/louisehumpington/) about accountability, harm, and the very human work needed to govern AI well. We explore why “strategic positioning” beats simple “progress” narratives, what a “mycelium network of governance” looks like, and why caution from women and Gen Z is not a bug but an intelligent signal.
1. Introduction
In this episode, we sat down with strategist, AI ethicist, and executive coach Louise Humpington to talk about AI governance in the real world: who is responsible when things go wrong, and how we design systems that do more good than harm. We dig into the recent joint letter to US Congress from OpenAI, Anthropic, and Microsoft, and why Louise sees it less as pure progress and more as smart strategic positioning by big tech.
From there, the conversation ranges across law, international development, organisational design, and human rights all threaded back to a core question: how do we govern AI in a way that is agile, accountable, and genuinely human‑centred?
2. Meet the Guest
Louise Humpington is a Strategic Advisor and Executive Coach focused on AI governance, ethics, and human rights, with a background that spans philosophy, financial services litigation, international development, and zero‑waste retail. She holds legal qualifications (PGDL, LPC) and has worked on regulatory services, ESG, and corporate social responsibility before moving into AI ethics and governance consultancy.
Today, Louise helps organisations navigate emerging regulation such as the EU AI Act, design multi‑disciplinary governance structures, and build what she calls a “mycelium network of governance” across teams and sectors. She also writes the Substack newsletter “From Intentions to Impact,” where she explores the ethics and governance of AI, human rights, and the gap between good intentions and real‑world outcomes.
3. Setting the Stage
AI is moving faster than traditional governance and regulatory models were ever designed to handle, which raises a simple but uncomfortable question: when harm happens, who is actually accountable? As we discuss with Louise, laws, frameworks, and company structures tend to assume neat, linear chains of causation, but modern AI systems are multinodal, evolving, and entangled across vendors, data sources, and use cases.
We also explore why broad public resistance to AI from concerns about bias and surveillance to job displacement and cognitive decline is not just noise to be “managed” but critical feedback that should shape how we build and deploy these systems. Listeners (and readers) can expect a grounded conversation that moves beyond buzzwords into concrete questions: what harms can this tool cause, who will bear them, and what structures do we need in place before deployment rather than after the fact?
4. Episode Highlights
Highlight 1: “Strategic positioning,” not just progress
We open by asking Louise about the joint letter to US Congress from OpenAI, Anthropic, and Microsoft widely hailed as a sign of progress and responsibility. Louise agrees that the increased collaboration and willingness to engage on governance is positive, but she also calls it “strategic positioning” on several levels:
- Big tech understands that legal and regulatory frameworks may eventually place liability at their door for harms caused by downstream bad actors using their tools.
- Participating early in shaping guardrails allows them to influence how far that accountability actually reaches and how it’s distributed.
One standout idea: these companies are not just “being nice”; they are anticipating how deep the impact of their systems might run and positioning themselves accordingly in the governance conversation.
Highlight 2: The “mycelium network of governance”
Later in the episode, Louise introduces one of the most memorable metaphors of the conversation: governance as a “mycelium network.” Instead of governance sitting with a small, centralised expert group, she argues it needs to be distributed across technologists, regulators, human rights practitioners, end users, and front‑line staff all connected by a communication layer that translates across different languages and perspectives.
As Louise puts it, we need people who can bridge law, technology, human rights, and business to “decipher” what each side is really saying and build a shared language for risk, harm, and accountability. That’s a radical shift away from siloed governance structures and a big theme in how she thinks about AI oversight.
5. Deep Dive: Grounded Governance in a Multinodal AI World
One of the core threads in this conversation is how poorly traditional governance models map to the reality of AI systems. Historically, legal and regulatory thinking assumes a fairly linear chain: system → failure → harm → accountable party. With AI, we’re dealing with complex, multi‑layered, constantly updating systems where models, data pipelines, agents, infrastructure, and user inputs interact in unpredictable ways.
Louise’s take is that trying to govern AI with purely linear accountability chains is like trying to navigate a 3D maze with a flat map: you will inevitably miss critical interactions and emergent harms. Instead, she suggests we need governance architectures that mirror the multinodal nature of AI itself a toolbox of laws, policies, technical controls, processes, and cultural practices that work together rather than a single “silver bullet” regulation.
This also implies a more continuous approach to oversight. Most briefs given to technologists run up to deployment and maybe to a first audit, but not much beyond. Louise argues that for AI systems, we need a baked‑in feedback loop from real‑world use back into design and updates, so that harms, near‑misses, and unexpected use cases actually reshape the system over time. That’s not just a compliance issue; it’s also a commercial one, because products that adapt to real risks are more resilient and more trusted.
6. Real-Life Stories & Examples
The episode is full of concrete stories that make these abstract governance questions feel very real. A few that stood out to us:
- The “rogue” agents and the missing basics
We talk about the now‑familiar media stories of AI agents “going rogue,” including agents deleting entire databases in seconds. Louise’s point: often the root cause isn’t the AI itself but basic data governance failures like putting production and backup databases in the same cloud environment, or skipping decades of IT best practices just because there’s “AI” in the stack. - Aviation as a model for psychological safety
Louise draws a compelling parallel with aviation, which has become the safest form of travel mile‑for‑mile partly because of an active no‑blame culture around incident reporting. When something nearly goes catastrophically wrong, the industry’s instinct is not to find a scapegoat but to understand what failed and how to prevent it in future a mindset she believes AI‑using organisations need to adopt. - Patagonia vs “Patagonia”: trust built and then undermined
To illustrate how trust is earned and lost, Louise references outdoor brand Patagonia, which historically built strong trust by being brutally transparent about the environmental impact of clothing manufacturing and positioning itself as part of the solution. She contrasts that with the company’s more recent decision to sue a drag artist and environmental activist performing under the name “Patagonia,” which she sees as a missed opportunity for collaboration that undermines their previously strong trust narrative. - Zoom and the power of plain language
On a more positive note, Louise shares how a direct conversation with Zoom’s trust and safety team dramatically changed her comfort level with their AI features once she understood, in clear language, the protections already in place. The lesson for us in tech: transparency and legible explanations are not “nice to have”; they materially shift adoption and trust, especially for sceptical or previously harmed users.
7. Key Takeaways
Here are the core ideas we walked away with from our conversation with Louise:
- Governance is strategic positioning
Big tech’s public engagement with AI regulation is both welcome and self‑interested: they are trying to shape where future liability lands and how broad it becomes. - Linear accountability breaks in a multinodal world
AI systems don’t behave like simple, static products, so our governance structures cannot rely solely on linear “harm → responsible party” chains. - We already have many of the tools we’re just not using them well
From IT governance and backup practices to leadership development and safety cultures, much of what we need isn’t brand new; it’s better coordination and application. - We need a “mycelium network” of governance
Governance must be distributed across technologists, regulators, human rights experts, end users, and cross‑disciplinary translators who can bridge different languages and priorities. - Psychological safety is a governance technology
Without cultures where people can safely raise concerns, near‑misses turn into crises; aviation shows what’s possible when you get this right. - Women’s and Gen Z’s hesitancy is rational
Women and under‑represented groups are over‑represented among those harmed by AI (biased hiring, facial recognition, healthcare, credit scoring), and Gen Z has lived through crisis after crisis driven by systems not designed for them. Their caution is data, not a problem to bulldoze. - Trust and transparency are commercial advantages
Clear, plain‑language explanations of risks, protections, and data use combined with genuine engagement with the most sceptical users are powerful drivers of sustainable adoption. - Ask different questions before you ship
Before deploying any AI system, ask “Who is the end user, and what harms could this tool cause them?” and design governance, guardrails, and feedback loops from that starting point.
8. Closing Thoughts
As technologists, we tend to focus on what AI can do; this conversation with Louise reminded us to spend at least as much time on what it can break and who pays the price when it does. Governance, in her framing, isn’t an after‑the‑fact compliance chore but a design discipline that spans human rights, law, psychology, and classic IT good practice.
For us, the big challenge to our own mindset is to treat caution and mistrust especially from those historically harmed by systems as valuable input, not resistance. If you’re building or deploying AI, we’d love to hear: how are you involving the people most at risk from your systems, and what does your own “mycelium network of governance” look like today?
