Plator Ltd

The first Global Dialogue on AI Governance will be a success if there are tangible commitments by world governments to unite behind this UN dialogue and actually drive change - rather than just signing up to another good-intentioned set of principles, guidance, and governance policies. From the Bletchley Park Summit in 2023 through to the open letters signed by many eminent scientists and people around the world, these have all been (in effect) posturing - not actually enforcing specific adherence in a common global agreement, and not calling out those who aren't working together collaboratively across the world. The success of the UN Dialogue will be formal commitment to make a change.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight
• Interoperability of governance approaches

I think protection and promotion of human rights should be cross-cutting, not a theme in its own right - it underpins all other themes. As I noted in my verbal intervention at the April webinar, my main focus is on promoting AI accreditation across various fields as a key lever industry can quickly adopt, moving us toward a highly responsible society, a bit like highly responsible organisations such as NASA where "safety first" is defined by principle. Everybody in the supply chain (end users, trainers, developers, suppliers, deployers) should be accredited to a certain level in their own area of specialism, such as law, coding, or general-practitioner AI, as to their level of competence, quality and credibility. Licensing has been dismissed in AI, but other industries - pharmaceuticals, and the full transport system - have exactly these sorts of models. Take car transport: we have driving licences for different types of vehicle, garages to ensure car safety, the Highway Code, and even fuel and carbon emission standards - all applying to road safety. And yet none of these apply to AI safety and ethics.

First, the significant increase in data collection and what is no longer covered by GDPR. When somebody asks a chatbot about their health, their weight, their food, their heart, the thoughts they're having, this is all deeply personal information, and it's going into models. Intermediate providers could be bad actors collecting significant amounts of data on individuals for malicious intent. Beyond that, the behaviours, thoughts, feelings and actions of individuals around the world are being collected by large model providers. From a human rights point of view, where does it leave people - exposed, when effectively digital avatar twins of themselves are being created? Far too many terms and conditions are out there for anything these days; you give your life away and then find your data has been sold off without you realising you gave consent. Second, legal decisions should never be handed over to an AI process. The one outstanding thing that is still uniquely human is law - it is made by humans, agreed by humans, and changed by humans. If we cross the boundary to allow AI models and AI companies to drive legal decisions, we have crossed a boundary too far. Third, as noted, protection and promotion of human rights should itself be treated as a cross-cutting principle, not a single thematic area.

In terms of governance gaps and related developments, there is enormous confusion, especially among everyday citizens and SMBs (and even larger enterprises). There is information overload. We are seeing frameworks, government frameworks, and governance frameworks proliferating around the world. We don't need more frameworks. We need common standards. We need common agreements. We need specific accredited kite marks of quality that people recognise, so they know that work has been done to a certain standard, and that training, deployment and use are happening in the right way. Otherwise we just open ourselves up to misinformation, manipulation, and significant breaches of law around the world, simply because people don't understand how to use AI: when they can't use an image, when they can't use a generative output, let alone the security risks that end up appearing in the crime statistics.

The AI Dialogue's most important role is to move the world from posturing to formal commitment. From the Bletchley Park Summit in 2023 through to the open letters signed by many eminent scientists, we have seen plenty of good intentions but very little enforced adherence — and very little willingness to call out those who aren't working together collaboratively. The Dialogue can change that in three ways. 1. First, by acting as the convening body that brings governments to tangible commitments rather than another set of principles — including a mechanism to publicly identify those who sign up and don't follow through. 2. Second, by becoming the connective tissue between the inner circle — the UN, governments, think tanks and frontier labs — and the rest of the world. Right now, material is published at the top but very little reaches universities, SMBs, local communities, or even trained practitioners working in AI and ethics. The Dialogue should be deliberately structured to cascade information downwards and feed responses back upwards. 3. Third, by championing common standards, common agreements and kite marks rather than yet more frameworks. Cooperation only becomes real when there is something specific that countries, organisations and individuals can be measured against — recognised across borders, in the same way driving licences, pharmaceutical approvals and aircraft safety standards are mutually recognised today. If the Dialogue does these three things — commitment, cascade, and common standards — it becomes the place where international cooperation on AI governance actually happens, rather than another forum where it is talked about.

There is no shortage of relevant material to build on: the Bletchley Park Summit and subsequent AI safety summits, the open letters signed by many eminent scientists and people around the world, the various national and regional AI strategies, GDPR as a data-rights precedent, and emerging accreditation thinking from professional bodies. Beyond AI itself, there are very mature models worth connecting to — pharmaceutical regulation, the full road transport system (driving licences for different vehicle types, garages ensuring car safety, the Highway Code, fuel and emission standards), aircraft safety, and "highly responsible organisation" cultures such as NASA where safety first is defined by principle. The added value the AI Dialogue can bring is not another framework on top of these. It is convergence. Specifically, the Dialogue can: * Pull these strands together into common standards and agreements rather than parallel ones. * Translate safety-critical models from other industries (transport, pharma, aerospace) into AI-specific accreditation pathways — for example, a Bronze / Silver / Gold scheme across AI law, AI marketing, AI sales, advertising standards on AI, with self-registration at the lower tier and peer review at the higher tier. * Establish kite marks end users actually recognise — so that training, deployment and use can be identified as meeting a known standard. * Act as the bridge between the inner circle (UN, governments, think tanks, frontier labs) and the universities, AI interest groups like Thames Valley AI, SMBs, professionals and local communities who today are largely outside the conversation. The added value, in short, is turning a crowded landscape of good intentions into something coherent, recognisable, and enforceable.

The AI Dialogue needs to bring in more professionals from the wider SMB business community and a broader range of stakeholders, including AI communities such as Thames Valley AI (University of Reading, Henley Business School), the Centre for AI Safety, BlueDot Impact, and similar groups. These are key AI communities making important links between wider society and the central "inner circle" of AI players. There is a massive gap between what I would call the inner circle — the United Nations, governments, think tanks, and the frontier labs — and the rest of the world: normal people and normal businesses. High-quality, but often dense, inner-circle material is being published, but not much of it is communicated or shared in a way that cascades down effectively. Even intermediate, higher-level tiers can struggle to translate it for the next layer, including universities, let alone trained AI safety and ethics practitioners like myself working in the field. I do my best by speaking to local communities, but SMBs are still largely devoid of what is happening in AI. The Dialogue should therefore be structured to push content actively down through these tiers, not simply publish it at the top.

Several groups are underrepresented as discussed above. SMBs are totally devoid of what's happening in AI. Local communities have very little access to current governance thinking. And even trained practitioners working in AI and ethics — outside the inner circle of the UN, governments, think tanks and frontier labs — are largely on the outside looking in. Inclusion needs an intentional cascade: regional and local forums, ambassador-style intermediaries who can carry the UN-level messages downwards and feed responses back upwards, and direct engagement with universities and SMB networks — rather than relying on these voices to find their own way to the table. Funding and easily accessible grants to SMBs, specialist interest groups like Thames Valley AI and low / zero cost access to "higher order" training, or professionals, networking events (international summits command £000s of entrance fees) - so its not about who can pay the most, its about who can contribute the most

Webinars are one format, but they aren't enough on their own. The most effective approach would be a tiered system of national, regional and local forums with a deliberate cascade of information — where accredited individuals operate at each tier like ambassadors for the UN, able to communicate the messages, translate them for their context, and so forth. This turns the Dialogue from a single event into a continuous, multi-level conversation rather than a one-way broadcast from the top.

The strongest concrete approach, in my view, is accreditation modelled on what already works in other safety-critical domains - car safety, aircraft safety, pharmaceuticals. A workable model is a Bronze / Silver / Gold professional accreditation pathway across specific domains - AI law, AI marketing, AI sales, advertising standards on AI, and so on. Licensing would be self-registered at the lower level and peer-reviewed at the higher level, and would be stripped from anyone engaging in deep fakes or misinformation. It's all about responsible, ethical narratives. The model can be self-adopted, peer-qualified at higher tiers, doesn't have to be heavily regulated, and would be fairly easily implemented. It also gives end users, developers, deployers and suppliers something they can actually recognise - a kite mark - rather than yet another framework.