AI-assisted Governance – Pillar 4

If we survive relatively unscathed until about 2040 – the time of the formation of the Human Federation (HF) – our civilisation will be ready to gradually deliver unimaginable wealth to everyone on the planet. But to deliver the world of abundance we will need a very efficient World Government – the executive body of the HF. But even before then, a reformed democratic system should propose solutions for the governments to be more efficient and effective. That is the objective of the fourth pillar of Consensual Presidential Democracy (CPD).

How can we do that when nearly all governments world-wide are today run by politicians, who are not top experts in efficient delivery of services such as health service, education or economic development. Yes, they have the support of a civil service and thousands of advisers and consultants but in the end they themselves have to make the final decision. The problem is that quite often such a decision requires really deep understanding of the subject matter.

The consequence of that is that many of the projects initiated by ministers run over time and budget and some, especially the most expensive ones, which will have an impact for decades are unnecessary. One of the best recent examples is the HS2 railway project in Britain, which is to be completed in 20 years, cost over £100bn and which only now, a few years after its initiation. In the same week, in February 2020, when the UK government gave the final ‘go-ahead’ for the project, a Chinese company offered to complete it in just 5 years, for less money and with trains running faster. That is what choosing the wrong type of project can lead to.

Examples like the one above, prompt some academics to suggest a silver bullet solution – a technocratic government run by experts. Its technocratic ministers would respond to the parliament as needed or as the law may require. The apparent logic behind a technocratic government is that it should be the parliament, which tells the government what to do, and it is the government, which knows how to do it. That would also increase the separation of powers. Such governments have been set up in many countries mostly in the ‘hour of need’ but only as a temporary solution, rather than a ‘normal’ feature of delivering services to the nation. The British civil service could have been considered a kind of a technocratic government had not all its departments been headed by Secretaries of State and Ministers (altogether 118). An exception is perhaps Singapore with its longest, and probably most effective, technocratic government, which achieved an incredible growth of prosperity for the nation over a few decades. Today’s China, which is modelled to some extent on the Singaporean economic system has, for all purposes, a technocratic governement as well.

So, why are such governments still a rarity? The main problem of technocratic governments is their accountability. That’s why they are usually disliked by both the public and politicians even though they are more likely to deliver value for money for the society than a government led only by politicians. Unless the whole political system is a blend of democratic and authoritarian rules, as is the case in Singapore, such government are not here to stay. Therefore, in the pursuit of effective and efficient government we need to look for other options. What I propose here may significantly impact, if implemented, political decision-makers at any level of governance, i.e. ministers, governors, mayors, councillors etc. The solution that I consider involves the support of politicians and decision makers at all levels of governance by AI assistants. This will happen anyway on a grand scale in just the next few years, when almost every profession such as medicine or engineering will be supported by such AI assistants.

If you think it sounds incredible, then just look at the offerings of one company – Generis. It has already a number of industry-specific AI Assistants. For example, CARA (Case Analysis Research Assistant) can work in most ‘soft’ areas such as law, pharmaceutical, medicine or government. It is competing with ROSS, which is primarily an AI legal assistant, which has already delivered incredible results, especially in the Anglo-Saxon world, where law is case-based. There are other such AI Assistants in the legal area where they deal with thousands of documents per case, so are engaged in similar tasks as in most government departments, such as:

  • Due diligence – Litigators perform due diligence with the help of AI tools to uncover background information
  • Prediction technology – An AI software generates results that forecast litigation outcome
  • Legal analytics – Lawyers can use data points from past case law, win/loss rates and a judge’s history to be used for trends and patterns.
  • Document automation – Law firms use software templates to create filled out documents based on data input
  • Intellectual property – AI tools guide lawyers in analysing large IP portfolios and drawing insights from the content
  • Electronic billing – Lawyers’ billable hours are computed automatically (E.A.Rayo ‘ AI in law and legal practice, 2019).

If you consider the continuous self-learning of such AI assistants like IBM’s Watson, or more popular, but also potentially game-changing solutions, like Amazon’s Alexa or Google’s Assistant, then within a few years, work in many companies of these industries will change beyond our imagination. The easiest way to imagine such an assistant at work is to visualize that you have a humanoid robot driven by Amazon’s Alexa-type application. Today, such an application can communicate in perfect, easy to understand accent, in about 60 languages but only one way. We can only understand what the app is saying but it has serious difficulties to continue a natural contextual dialogue. Therefore, quite often its response is just ‘I don’t know that one’. Only the very best, most expensive robots, linked to superfast computers, such as Sophia by Hanson Robotics, can have a longer meaningful dialogue. However, according to the company, it needs another 2-3 years before its Sophia will be fully conversant on most subjects. According to Ray Kurzweil, the most renowned futurist, we will need to wait till 2029 (he is precise about the date), when AI will achieve human level intelligence (in terms of processing power not intelligence as such).

At that time, almost every decision made by a political decision-maker or any consultant will be executed as the AI assistant had suggested. Until then, these robots will be capable of advising on a narrow subject matter using its database of knowledge. Such databases are already being produced as plug-ins (see CARA and ROSS above), purchased as a service and then maturing through self-learning in a concrete environment, e.g. at the Ministry of Health. Therefore, realistically, we can expect a widespread use of such assistants by about 2025, although probably with very limited cognitive capacity yet. However, even at about 2030, such AI assistants will not have a multi-disciplinary knowledge and intelligence (human level cognition) in every discipline, otherwise they would have become Superintelligence. Therefore, a ‘Master’ Assistant serving for example the Minister of Health, will be a generalist and will still need to be supported by several of his ‘colleagues’, each in a different discipline. On the other hand, from the point of view of the user, the whole process of knowledge acquisition, interpretation, compilation and presentation of final answers by such AI assistant will be largely seamless. The quality of its response and decisions will largely depend on the quality of data to which it has access and its overall skill level it has learned on a given site.

The benefits gained by the government of a country implementing such an AI-assisted governance will be immediate and significant. First of all, most decisions will be made many times faster, with full justification and with various options costed. They will also be correlated with other decisions made in a similar way by AI assistants helping across all government departments. There will be fewer missed deadlines and unwanted projects. The savings will be truly vast if implemented at all levels of government.

Finally, there will be very few purely ‘political’ decisions to win the votes in the coming elections since the planning horizon for most of such projects will cover a decade or more. Additionally, should there be a legal requirement (say in 10 years’ time) that each decision made by a minister must be justified by an AI assistant – an entirely apolitical entity, populism will be most likely rooted out. That should not be a surprise at all. If you agree that in about 30 years’ time Superintelligence will become our benevolent dictator, then what would be practiced in the intermediate period is just a preparation for what will happen on an unprecedented scale in every step of our life anyway.

Such implementation would allow politicians to have a personal, direct control on even the largest initiatives and projects, executing them with incredible effectiveness and efficiency. The added benefit will be a continuous parliamentary scrutiny, should such a politician be an MP. To make the best use of these assistants, say from 2025, they will probably be best used as additional advisers to humans. However, they should be physically present in a humanoid form in their ‘place of work’ for three reasons:

  • If it is in a physical, humanoid form, hardly distinguishable from humans, it will also move around almost like most of us, explore and learn about its environment, listen to conversation and analyse the problems ‘first hand’
  • It will have the ability to practice its learned skills and improve on them in a real physical environment
  • Finally, it will also learn our values, emotions, how we make errors and simply what is good and bad. It willlearn our preferences, rather than simply, goals. That can only be experienced in a real physical environment by a real (not augmented) physical humanoid robot.

Gradually, through self-learning and additional augmented reality capability, such AI assistants will become better and better in making decisions than most human advisers. It is at this stage, that some legislation may be needed to minimize the risks for humans from such advanced robots. The first law might be to recognize a concrete AI Assistant, as having some rights – e.g. only certain people will be able to make highest level decisions, and if needed, switch off the assistant. Secondly, laws may be introduced, requiring a politician to execute a decision made by such an AI Assistant because that might be in the best interest of the nation or a given community. The only exception might be in case such an assistant’s decision is challenged by a panel of human specialists. In any case, expect some interesting laws to be introduced quite soon regulating the sphere of initial coexistence of humans and AI assistants.