How to shape the development of AI for public services (January 2024)
A response to the question “How can AI be implemented to improve public services?”
Introduction
The key to unlocking AI for public services lies in public sector certainty (‘strategy’) and the interplay between our cultural evolution, this GPTs construction and research and a thriving, competitive, innovative, private sector. AI as a GPT is unique - we can imitate and automate the human - in its profound capacity to impact our culture and way of life.
The way in which the public sector engages with AI technology can and should influence the cultural soul and direction of AI development. Whilst doing so, such engagement provides the basis for market success for a competitive and innovative AI landscape that is adapted to a particular nation's demands. Responding to this broad and deep opportunity requires a similarly unique, broad, and culturally grounded system-based strategy unique to each nation.
Governments should have strategies to nurture private development of their national AI markets and utilisation of AI services in the public sector. This requires the following five elements:
A published set of high level goals that can shape the focus of public, private and academic investment.
A parallel infrastructure plan that ensures the exponential demands of AI technologies are not unnecessarily constrained.
A national AI-body responsible for AI capability and culture that identifies challenges and gaps in market development that is endowed with the resources to resolve them.
A short and medium term purchasing strategy that gives power to public service providers to experiment with AI technologies safely in a manner aligned with the national strategy.
A mandate for regulators to both support the national AI strategy and innovation diffusion.
This essay outlines the need for these elements for enabling the public sector to maximise the opportunity presented to itself by AI technologies.
Note: In this essay I focus on approaches at national level public sector systems.
Why a systems approach is needed in AI to help guide public sector choices
No technology exists in a vacuum, and we humans are fundamentally influenced by the machines we employ. As AI models attain greater levels of complexity and capability their selection becomes critical. This is of paramount importance to nations’ capabilities and identity, it is not as simple as picking the most ‘effective’ technologies in isolation. Technologies emerge as components in a feedback system, they are a product of, and likewise shaper of the culture and environment that created them. Choices on AI will have an impact long after the initial technology has gone.
The procurement and financing of any technology play a critical role in its development, a notion that is especially pronounced in the case of AI. For example with LLMs (Large Language Models) we actively define - or more accurately, train - their entire behaviour and character. The LLM is influenced by both architectural and technological choices and also the data that we chose to instruct and align our models with.
Imagine your government is choosing between two LLM chatbots that staff will use to query their internal policies, documents, along with other highly ranked materials. What if one model was trained on literature from across the world, and one just from your nation’s x.com feed?. Both test well, interpret the wealth of internal information, and do a good job. Over the long term these LLMs would provide a different experience to its users. Those users in turn would build different expectations on how LLMs should behave and provide different service to their customers (the public) - a simple example of the circular influence on culture, from a procurement choice.
High level, long term goals to set all sectors developing in the same direction
The first thing that the public sector must do is clearly identify their biggest challenges that may be solved with AI. These can form the code of a set of goals for AI in the public sector from which long term investment decisions can then be made by all actors in the system.
There are countless examples across millennia of public services, procurement, investment and techno-leadership that delivered great leaps in a society’s capability to develop and leverage technology. From the Apollo moon landing program (computing, materials, medical imaging, culture); To the UK Longitude awards of the 18th Century; We can even look back at the techno-public infrastructure of the Roman Empire of aqueducts, roads and public utilities that allowed them to define an era (as is China’s stated goal with AI). Are there public goods of the AI age? Are GPU clusters the new roads that the public sector is failing to build, and how can we find out, quickly?
A system based approach to improving public services through the use of AI must focus on the cultural system that produces it, and that it feeds back into. This requires a clear set of goals for what that system is working towards, and guidance on what is expected from its contributors (a clear set of rewards); What AI is being applied to achieve, not how AI could be used to achieve a particular objective.
This is a combination of goals for the deployment of AI, guidelines that shape potential private sector suppliers' choices, and building a safe environment for experimentation with the public sector in the provision of public services. This must be culturally aligned to a nation's identity and objectives, well communicated, shared and enforced.
Such high level long term goals for a nations’ AI in public services should include:
A digital healthcare system that uses automatic insights to manage disease from population to individual.
A thriving private market, and widespread adoption for AI tools that are ‘good’.
A trusted data sharing system for training models on the nations’ data.
A citizen engagement platform for policy formation and delivery.
The use of AI to bring the quality of education received by all students to outstanding.
The existence of this AI plan will directly shape the behaviour of those developing new technologies and align the market development of technology more closely with those who would adopt it.
Investing in the right infrastructure
With high level objectives, longer term insights and an exponential technology provide a unique challenge to public services providers. A nation’s infrastructure requirements for AI are not well understood, save for some fundamental requirements for the running of the technology: GPUs, energy; And building the technology: researchers and data. These, and others should be prioritised at a national level as infrastructure that is available in-country to those developing AI technologies that align with the nation’s strategy.
Nations should urgently consider their own capacity for AI development infrastructure, including:
Building compute capacity that can be used by a country's researchers.
Energy generation to underpin demand of new super compute clusters.
Encourage research and development and knowledge transfer between countries.
Opening up public sector data to accredited national research institutions and organisations.
Policy makers and public service providers are under pressure to make the most of this exponential technology. There is hype, the technology is here, and people are excited about what will happen next. There is corresponding pressure to do big things and to jump forward to where our imagination takes us as opposed to what is technically practicable. This is an exponential trap. As technologies that grow exponentially are frustratingly slower at their emergence and then dramatically more disruptive or impactful later on.
So whilst we can see great areas for leveraging what we think AI technologies are in the public sector, neither the technology nor the public sector is really ready to implement them - whatever they may be. With a published national plan, key infrastructure can be identified and be invested in that supports the development of AI technologies in the private, public and research sectors. This lowers the barrier for innovation, and potentially increases the competitive markets that eventually the public sector service providers will be drawing on for their AI solutions.
A body responsible fit for the technical challenges of AI
With new technologies come new opportunities for the improvement of existing services and the provision of new ones. Both the requirements for investment in key infrastructure, and a national set of objectives for AI, require a body to continually update them as the system evolves. Governments need an expert body that is responsible for understanding the strategic, technical, and public sector landscape and has the resources to affect them for the benefit of the nation and its cultural priorities.
This AI Body should have responsibility to identify and define key technical, market and innovation challenges in the application of AI and report to a decision making body that has the resources to address them. This would include feeding back into the national plan and infrastructure projects already outlined, and would comprise a team with deep technical expertise able to directly change how a nation's technical capability can be applied to improve citizen life and provision of public services.
Outcomes of this AI Body’s work should be work such as:
Sponsoring the creation of local open source LLMs that are aligned (trained) that adhere to best practice and have data sets that represent a nation's character.
Building large research clusters (GPU supercomputers) that are open to research targeted at solving the objectives of the public service provider plan.
A change in the nation’s infrastructure plan to increase the long term (green) power generation for a country by x% to underwrite future compute demand of AI models.
Coordinating and educating regulators nationally and internationally in particular industries on the emergent AI technologies in their respective industries.
Establishing trusted data handling protocols for sharing healthcare data with AI companies.
Prizes aimed at non-market solutions; A $5m prize to reduce the complexity of legislation by 50% without material to the system, or the overall meaning and intent of the law (Prove your working!).
This could be created by extending the remit of existing institutions, for example the UK could grant the Alan Turing Institute greater resources and scope, turning it into something like a ‘BBC for AI’.
Short and medium term guidelines for procurement and experimentation
A short and medium term purchasing and experimentation strategy should be used to give power to public service providers to experiment with AI technologies safely in a manner aligned with the national strategy (procurement rules of the road).
There is strong opportunity for both less dramatic implementation of mainstream AI productivity improvements already demonstrated in the private sector, and experimental ones. Examples abound, such as using LLMs for internal communications; knowledge sharing; training of staff; translation and process automation. There will also be opportunities for experimentation and creative applications which have not been thought of yet.
To unlock these improvements the public sector needs to provide clear purchasing guidelines that align a nations’ risk appetite with its plan and to encourage the ecosystem of experimentation. This needs to be done quickly as LLMs are already used in the wild by employees, whether their employers want them to or not!
This should include a code of practice laid down by the AI-Body outlined earlier (scale dependent) that aligns purchasing decisions, private sector decisions with a nation's AI strategy. For example:
The data sources that a model is trained on must be:
Public (as in the sources published, not the data)
20% from-the-country-of-use
Appropriately match the population demographics of the intended use case (tough!)
Handled with clear, public, ‘editorial’ guidelines
Define the specific training requirement a user must have in order to assume decision making power from the AIs outputs.
All AI content to be flagged and visible to all users, who being appropriately trained can (hopefully) act in an informed manner to the AIs output.
A mandated ‘code of practice’ for example for AI suppliers to the Public Sector would not only set the tone, behaviour and provisions to the Public Sector, but protect public sector procurement decision makers. Those who are responsible for smaller scale or experimental deployments of AI technologies can use it to short cut decision making processes on the ethics or risks of adoption.
Such a strategy must include a requirement for public service providers to actively seek out current private sector best practice - which has significantly evolved in the past few years - and can be informed by the expertise of the AI-Body. Clear rules for experimentation and procurement plans (without large projects) should build competence, capability and also, being within the context of a national objectives of the application of AI, move public service provision towards those longer term goals.
Regulators enabled to accelerate sustained innovation in AI
In order to have a proper public sector procurement system for AI you need a competitive market. If the same old large suppliers are repackaging last year's AI technology from a repurposed failed project from elsewhere then you’re probably not operating a competitive marketplace. This is the job of national regulators, and it is a wider issue than AI in particular as it concerns innovation as a whole.
Competitive markets are required to truly support the development of AI technologies that are aligned with a national plan. I suggest that this be addressed by creating a mandate for all regulators to have ‘encouraging and promoting innovation’ added to their remit and a top level responsibility for all economic and technical regulators created to encourage innovation and innovation diffusion in their industries. These responsibilities must include a presumption that innovation is favoured in any trade off innovation against current statutory duties and an ombudsman-like process for dealing with barriers to innovation.
To demonstrate their understanding of innovation within their industries it should then be the job of the regulator in each sector to report on where these bottlenecks in the innovation process, and in this case specifically AI technologies, and suggest measures of redress. This work can go hand in hand with the AI-Body outlined earlier - in a similar way to how specific industry regulators will work with a national competition authority.
Specifically for AI Technologies regulators could be tasked with providing specific options on how to change their industries to improve the innovation process. This should include options for new regulatory interventions or structural changes that would require changes to policy. This would, in the case of AI, provide feedback both into the body tasked with addressing bottlenecks and put the regulator in a position of being ideal advisers to public service providers on technological adoption and new technologies.
Summary
Through a systematic response, that is unique to the character of a nation, that outlines strategic objectives, procurement rules, new responsibilities and invests to address long term bottlenecks, nations can unlock the power of AI in public services. This requires both patience and bold action. Patience, to avoid investing early in large scale projects and bold action to declare clear objectives, invest in existing and new infrastructure and create new bodies.