Good AI
10 August 2026
As we launch AskGGi, Jaco Marais explores the promise, risks and contradictions of Ai, arguing that public-purpose leaders must combine technological innovation with critical judgement, strong governance and actual human intelligence.
Points raised in this article:
- AI brings opportunity, but serious governance risks remain
- Environmental impacts demand far greater scrutiny and transparency
- Regulation shows harmful uses of AI can be constrained
- AskGGi aims to combine AI with human judgement.
I am trying my very best to write this blog celebrating the launch of AskGGi, our very own AI-empowered knowledge agent, while not also apologising for the various misgivings I have about this new technology.
As a GenX millennial, I have been privileged to be both an early adopter of new technologies as well as a lifelong sceptic. I remember moaning as a young teenager, 'Just because we can doesn’t mean we should,’ whenever the latest computer, mobile phone or internet capability got launched onto an unsuspecting public, always promising a new age of democracy.
And now we have AI! Can you remember the first time you tried it out for yourself? Now think of when that was and then think of how much has happened and how it has developed in such a short period of time.
All in the name of progress! Or have we (collectively) taken one giant leap for technology while inadvertently taking an equally enormous step back for mankind?
Bad AI
Many of you will remember that a cornerstone of GGi’s purpose is to help create a better, fairer world. But the rise of AI, with its implied authority and scale, inevitably raises serious questions about fairness. Let’s not forget that AI was invented by disconnected hyper-capitalist tech giants who might just have the teensy-weeniest bit of a God complex.
A quick prompt for Claude produced a list of principal AI risks most relevant to senior leaders in UK public-purpose organisations — drawn from the National Audit Office’s (NAO's) May 2026 guide for audit and risk committees, the GDS AI Playbook for Government, and current ICO guidance.
- Data quality and governance. The NAO calls this a ‘foundational risk’: government data is often incomplete, inconsistent or siloed, and poor-quality inputs undermine any AI system built on top of them.
- Privacy and data protection. Handling citizen data through AI (especially agentic AI) raises UK GDPR compliance questions; the ICO is publishing a statutory Code of Practice on AI and automated decision-making covering exactly this.
- Bias and unfair automated decisions. ICO reviews of AI-assisted decision-making (e.g., recruitment, policing) have repeatedly found insufficient transparency, weak human oversight and poor bias monitoring — a particular concern where decisions affect citizens' rights or entitlements.
- Openness, transparency and public trust. Public bodies are held to a higher accountability standard than private firms; opaque AI decisions in benefits, policing or service allocation risk eroding public confidence.
- Cyber security. The AI Playbook flags new attack surfaces specific to AI — adversarial inputs, data poisoning, and misuse of generative models — on top of conventional cyber risk.
- Over-reliance and loss of institutional knowledge. The NAO warns that automation bias and growing dependence on AI tools can erode staff judgement and organisational memory over time.
- Skills gaps and vendor dependency. A broader cohort of staff needs basic AI literacy, while specialist skills shortages leave organisations more reliant on external suppliers — raising vendor lock-in risk.
- Accuracy and hallucination in generative AI. Outputs that sound authoritative but are wrong pose particular risk in public-facing or high-stakes contexts.
- Governance and accountability gaps. Who owns risk when AI informs a decision is often unclear, and regulation (the ICO's ADM Code, sector-specific rules) is still catching up with deployment.
- Cost and value uncertainty. The NAO has publicly challenged government assumptions behind projected AI savings (e.g. the £45bn productivity figure), warning that costs and benefits are often not rigorously tested before rollout.
Water
Claude failed to include any mention of the risk that I am personally most conflicted about when promoting the use of AI to do anything: the environment.
Data centres rely heavily on evaporative cooling, and a single hyperscale facility can use as much water in a day as a town of 10,000 people; globally, AI's water demand is projected to exceed all UK household consumption by 2027. That's a particular concern here at home: the UK already faces a projected daily water deficit of nearly 5 billion litres by 2050, yet 84% of proposed UK data centres are planned for already water-stressed regions, including the South East.
Compounding the problem, data centres aren't currently required to disclose how much water they use at all – a gap regulators and industry bodies are now pushing to close through mandatory, location-based reporting.
And the EU AI Act's environmental provisions are essentially energy-only. Water consumption, mineral extraction, and e-waste aren't covered at all — reinforcing the point that water-specific disclosure is a genuine regulatory gap, not just a UK oversight.
Environmental activist Erin Brockovich has launched a crowdsourced national mapping campaign that has logged over 8,000 community reports of AI data centres, pressuring the industry for greater transparency on water use, energy demand and local impact.
Hope
Amid the headlines about AI risk, it's worth pausing on how much human effort is already going into getting this right — and it isn't confined to Britain. At home, the National Audit Office is arming committees to challenge inflated savings claims before billions are committed; the ICO is building the UK's AI rulebook from real audits of police forces and recruiters; and a British cybersecurity code has become a global standard now backed by 18 countries.
Abroad, the pattern repeats: Brussels has stood up an entire AI Office ahead of new enforcement powers; a small US government team has quietly evaluated over 40 frontier models in partnership with the very companies building them; and Five Eyes cyber agencies co-authored the world's first joint guidance on securing autonomous AI agents. Even amid rising rivalry, China has mandated safeguards protecting vulnerable users of AI companion apps, and Washington and Beijing restarted a stalled AI safety dialogue this year.
Meanwhile, when Stanford scientists used AI to design a working virus for the first time, they flagged its dual-use risk publicly — feeding into a single shared risk assessment co-authored by over a hundred scientists from 30 countries.
And since February 2025, the EU has made the most dystopian uses of AI — mass biometric surveillance, social scoring, and AI built to manipulate children and vulnerable people — simply illegal across a market of 450 million. This is a clear boundary between the sovereignty of humans and AI.
AskGGi
So AI might be the only way forward or a massive step backwards, but for now this is the direction of information sharing between humans.
GGi needs to remain that independent, critical friend driving improvements to AI and AI use, using actual intelligence to hold artificial intelligence to account.
We invite you to integrate AskGGi into your daily workflow — not just as a tool, but with the same spirit, vigour, humanity, and emotionally connected intelligence that defines Erin Brockovich.
The body of knowledge numbers over a million words and has been added to by various contributors since 2009. Let’s continue to co-create this resource to empower leaders taking initiative to create better, fairer systems that demonstrate lasting value for society.
The test version of Ask GGi is available to try now. We’d appreciate your feedback.