Sovereign AI - in plain English

Unless you live under a rock you will be familiar with “consumer AI”, be it ChatGPT, Claude, Gemini, Midjourney or even the older and somewhat stupider Siri and Alexa.? For most people this is the face of AI and it’s what comes to mind whenever the topic of AI is mentioned.? Well that and Skynet…

There is another layer of AI tools that are in common use but aren’t so well known because they are aimed at businesses rather than the general public.? These are called Enterprise AI, and examples include Microsoft Copilot, Salesforce Einstein, and IBM Watson.?

There’re some important differences between consumer and enterprise AI, the key one being privacy of data.? Information that you provide a consumer AI will probably be used to train it and could be made public.? There are many examples of where this has gone wrong, such as Samsung engineers asking ChatGPT to review highly confidential design information, only to have that information made available publicly to all ChatGPT users.? Whoops.? My advice, don’t ask ChatGPT about that strange rash you’ve discovered after a big weekend in Vegas.

Other differences are the data that enterprise AI is trained on, which is likely to be semi-confidential and somewhat verified, so in theory enterprise AI should give more accurate answers in relation to a company’s business decisions.? Such as “what is the risk of giving this person a loan?” – the answer to that needs accurate financial records about the customer, not just the opinion of the Reddit community…

Consumer AI and Enterprise AI are commonplace and fairly well understood concepts, and it’s pretty obvious which one should be used for what purpose. There is however a third type of AI that’s less understood and a bit controversial about what it should and shouldn’t be used for.? Sovereign AI.

1984?

Big brother is what probably springs to mind when I say Sovereign AI – the all-seeing state, not the crappy 90’s reality-TV show.? But that’s not a fair assessment - if anything Facebook presents far more of a big-brother risk than a well-used Sovereign AI.

I’m now going to invent some acronyms to save me trying to spell sovereign over and over (yes this article is written by an actual human being, one that can never remember the rhyme about i before e except after g…).? So let’s call them cAI, eAI, and sAI.? Hopefully I don’t need to explain those.

What is sAI you ask?? Well there’re lots of definitions out there, not all of them line up as it’s still a new concept, so I’ll give you my view, but be sure to check that if someone else is talking to you about Sovereign AI they mean the same thing as sAI.

Where eAI is only concerned with data and decisions that impact a single business (such as a bank, retailer, or law firm), sAI takes a system-wide view of things.? For example, sAI would look at a person in totality, not just as a recipient of a social benefit, or an applicant for a driver’s licence, but as a complete person.? This doesn’t just apply to people, it would apply any entity, such as land.? Rather than just viewing a block of land as a farm, sAI would understand it is also part of a geographical feature, a flood plain, a flight path, a source of natural resources, and a culturally important area.?

Taking a System-wide View

A system-wide view also allows sAI to understand the interconnectedness of things, to take a 10,000 foot view.? A bank’s eAI might be able to draw some conclusions and inferences on the state of the economy based on analysing banking transactions.? But it would be making assumptions based on a limited view of the system.? A reduction in transactions at restaurants would indicate people are going out less, but why that is happening isn’t something that can said for certain based on just one bank’s view of the world.?

A more pressing and socially valuable example could be understanding the mental health of a nation.? Using an eAI our health system might be able to detect an increase in suicide rates, but understanding why is not something that can be inferred from health data alone.? It would take a system-wide view of health, financial, employment, weather, social media, world politics, climate change, and family stability (to name just a few) to get a real understanding of why suicide rates increased, and more importantly what could be done about it.

This is the fundamental difference between eAI and sAI – the ability to understand the wider picture, and to provide recommendations that utilise all of the levers available, within the resources available. Addressing mental health challenges would take a coordinated effort across numerous government agencies, not-for-profits, and commercial businesses.? Even if all of these organisations possessed an eAI and invested in the cause, it would be unlikely to succeed.? Pulling a lever in an organisation will have an impact on other organisations, and often just results in moving the problem between organisations.

Only through a coordinated approach can we understand and address complex issues that require system-wide responses.? This is where sAI shines.

It's not just about Government

Often sAI is mistaken for a government only system, something that the deep state uses to monitor citizens and make nefarious decisions about who gets what.? This is not true at all (well maybe it is in China but who actually knows). If I was referring to a Government AI I would have called it gAI. sAI should be viewed as something that belongs to all the people of a nation, the information it holds and the benefits it provides are for and about the nation. Government is the appropriate place to create and manage our Sovereign AI, but government shouldn’t be viewed as the only user of it.

New Zealand suffers from some of the lowest productivity rates in the developed world, and the use of technology has been highlighted as a key way to address this challenge. If we were to establish a New Zealand sAI why would we lock it away for only government to use??

Sovereign AI could be the key to lifting our productivity, growing our economy and making us more competitive globally. Mind you, they probably said the same thing about Skynet, so let’s not rush into this without proper care.

?Tread Carefully

[The next bit about risks was originally written by Claude.ai, but don’t worry, I’ve fixed it to remove that familiar taste of chewing cardboard that comes with GenAI text – I mean it would be rather hypocritical to write an article about AI and not make use of it myself.]

While the potential benefits are significant, implementing sAI incorrectly or without the right guardrails has public disaster written all over it. To do this properly New Zealand must navigate several key issues:

Privacy and Data Protection: With the recent implementation of the Privacy Act 2020, any Sovereign AI system must adhere to strict data protection standards. How do we balance the need for data integration with individual privacy rights?

Transparency and Accountability: New Zealand's commitment to open government requires transparency around AI decision-making. Ensuring these complex systems are understandable and accountable to the public is crucial.

Fairness and Non-Discrimination: Given New Zealand's diverse population, it's essential that Sovereign AI systems are designed to be fair and unbiased, particularly considering the needs of Māori, Pacific peoples, and other minority groups.

Treaty of Waitangi Considerations: Any implementation of Sovereign AI must align with the principles of the Treaty of Waitangi. This includes ensuring Māori partnership in development and governance, and respecting Māori data sovereignty.

Building Domestic Capability: To truly harness the potential of Sovereign AI, New Zealand needs to invest in building its domestic AI capabilities, from education and research to infrastructure development. In my opinion, which I'm not known for keeping to myself, it is vital that New Zealand's sAI is built in New Zealand, stays in New Zealand, and is run by New Zealanders.

Its our Opportunity

As a small but innovative nation, New Zealand has the opportunity to lead by example in the ethical development and use of sAI. By leveraging our reputation for honesty and innovation, we can improve the wellbeing of our people, lift the economy of our nation, and protect the future of our environment. ?

New Zealand has the opportunity to be a showcase of AI development that respects indigenous rights and multicultural values.

Sovereign AI represents a transformative opportunity for New Zealand to enhance its governance, address key national challenges, and play a leading role in shaping the global future of AI. By approaching this technology with our kiwi innovation, ethical consideration, and respect for cultural values, we can harness its potential while staying true to who we are as a nation.

Comments?

What are your thoughts on the role of Sovereign AI in New Zealand's future? How can we ensure its development aligns with our national values and aspirations? Share your perspectives in the comments below. ?