In earlier posts, I wrote about differentiating data that has been observed from data that comes from calculations and models, particularly those that project into the areas currently impossible to observe, such as the future. The first type of data I described as bright data and the second type I described as dark data using the precedence of cosmology’s dark matter and dark energy: stuff we assert must exists because models tell us so.
Most of the data we actually have in databases is a mixture of both, the measures we want are often calculated from measures we can actually observe. For example, we may record a temperature value from a calculation from measure of current from a thermocouple. I describe data as having levels of brightness from very bright (very little modification) to very dark (data that requires extensive processing with elaborate algorithms to expose).
I proposed a form of government that would have algorithms to make all government decisions, policies, regulations, or even enforcement or judgement of individuals or groups. These decisions by algorithms could not be overruled by humans: the results of the algorithm would be law of the land. Some describe this as being some form of dystopia where humans would become serfs or even slaves to the machines. I question whether this is bad.
In comparison, our current form of government of democratic representation hands over power to select individuals make these decisions. Theoretically, these individuals are directly elected representatives who can be replaced in future elections. In practice, the actual decision makers are career bureaucrats whose jobs are outside the reach of democratic votes.
We have a ruling class of people making decisions that affect everyone. The structure of our government does permit some public participation. The public can demand data that supports the decisions, or they can challenge the data or offer data that undermines the pending decision. In this sense there is at least a beginning of having a government by algorithms and data.
Unfortunately, the current government still pretends to be a democratic republic with humans accountable for the decisions, even though the people held accountable (elected representatives) are not actually the ones making the decisions. A consequence of this fiction is that all decisions are ultimately made by humans. In particular, humans have the final say of choosing policy based on their personal interpretation of the data that that person chooses to accept.
Given the scale of governance that we must manage, it is impractical to impose on all these bureaucratic positions short terms that must be periodically renewed by democratic elections. The general population would be overwhelmed by having to make so many choices and they would end up checking off recommended lists handed out as colored flyers at the polling booths. There is no effective democratic input into the acceptance of qualifications of the people who will make the decisions that will most directly impact their lives.
Assuming that it is preferred to make policies based on acceptable data using acceptable algorithms, an alternative form of government (requiring a completely different constitution) would have computers make all policy decisions based on currently accepted data and algorithms. In such a government, no human would have any say in the final policy choices, neither a final decision nor a veto. No human would be accountable for the policy decision based on all of the acceptable data and algorithms available at the time. This would be rule by machines instead of humans, but we should be able to agree that the best policies would be the ones that are best ones measured by the best data and the best algorithms.
Having a government where people democratically elected representatives to make decisions on their behalf may have worked well earlier where the data was hard to retrieve, retrieve, process, and interpreted. In such times, we needed specialists to work with this data, often by hand using paper or over-generalized statistics. We now have technologies making the entire data process easier and transparent to large populations.
Government by algorithms and data (with no human decision maker) may soon be possible. Not only do we have the ability to collect extensive data covering a multitude of relevant variables, but we have the ability to process that data to select optimal policies among a list of possible alternatives.
The key to turning this into a real government is popular acceptance. One path to such acceptance is to have a democratic role in the process. Instead of having the population choose among various personalities to represent them in policy making, the population could be participating directly in the collection of data, the cleansing or identifying good data from bad data, and the verification and validation of the algorithms including the choice of competing algorithms.
Such a government by machine policy-making could be a more advanced form of democracy that is better optimized to serve the public as a whole. Instead of relying on individual’s competence and accountability to make the right choices, we would instead rely on the competence of the data and the algorithms that the population as whole accepts.
The democratic role in this government is in crowd sourcing the entire life cycle of data and algorithms to make the policies.
To be effective, such as system needs to evolve with new data and new algorithms. One consequence is that such a government would not have laws in the current sense where laws remain in force perpetually until a separate law revokes the former one. A government by data could be limited to making rules that remain in force only for a maximum period of time, certainly less than a decade.
Retaining a preexisting policy requires a new processing of the latest data with the latest algorithms that happen to choose that as the best policy among alternatives, where one alternative is to let the old policy expire without any replacement.
In order to have the best unbiased data to make new decisions, the currently in force policies should be minimized in order to allow the most freedom for the population to express themselves through actions and choices as they respond to their own individual challenges and opportunities.
Given such a model of limited number of policies in force and the policies having limited life-span, we would need to change our justice system. In particular, it makes no sense to impose prison terms that will extend past the time when the offended policy would expire. In this regard it may be sufficient in many cases to simply record in the data itself one’s guilt and this data will influence future opportunities for the individual. In other cases, we have to choose alternatives to imprisonment.
A current example are people serving prison terms of possession of drugs that were illegal at the time of their trial but are legal (or not as seriously punished) today. The ongoing imprisonment does not make sense today even though it was acceptable in the past.
I think this is a fundamental consequence of the evolution of society, it is inevitable that future sensibilities will be very different than current ones. Basing decisions on data instead of moral absolutes negates the justification for long prison terms as a form of punishment.
Prisons become obsolete when laws cease to pretend to be absolute and indefinite. Despite a person’s past transgressions, we need that person’s unencumbered participation in society in order to collect better data to make better policies in the future.
This form of government takes humans out of the policy making roles. In the example of USA, there would be no purpose to a capitol city such as Washington DC. All of the expected accountability of representatives and bureaucrats that inhabit this city would be replaced by machines that probably run elsewhere in the clusters based on optimal geography for computers and communications rather than human congregations.
Meanwhile, the population as a whole has broader opportunities to participate in the process of policy making and selection. Individually they may focus on their particular interest in the process. For example, some may be particularly interested in the collection and validation of specific forms of observations. Others will be interested in the various stages of data modeling and interpretive algorithms. There will be some form of consensus (undoubtedly with a vocal minority of objectors) for what data to feed the algorithms, and what algorithms would ultimately select the preferred policy, if any.
The final policy decision itself would be the output of that process. The data we permit into the algorithms and rules we permit will determine the policies that we must follow. No human will be responsible for that final choice. The only way to overturn such a policy would be to provide new data or new algorithms that would produce a different result. But that would require democratic acceptance of the data and the algorithm.
Part of the stakes involved is that once we accept data favorable for one desired policy we must accept the same data for choosing another policy that may be less desired. The incentive is on making sure the data is the data we most trust to make our policies.
This leads back to my initial observation about data having different levels of quality that I call brightness. Actual observations are the bright form of data, where the most recently observed data is the brightest data. No matter how recent the observation the observation will be of a historic event.
Policies that matter most to governance are intended to influence future outcomes. We may erase or rewrite our past through our choice of selecting acceptable data, but the goal of governance is to pick the policies that will confront future challenges or exploit future opportunities.
Such a goal cannot rely on bright data of historical observations alone. In particular, a goal of policy making is to influence people’s behaviors to be better prepared for something that has not happened yet but is very likely to happen in the future. To be practical, the policy making must accept as data the results of models, predictions based on some type of projection from past events. The prediction itself is the data that will determine the policy, not the past.
In recent years, there is a popular recommendation for people to make preparations for disasters. Often this is in the form of making sure the home is well stocked with materials to weather though a disruptive event that may mean weeks without access to supplies including water or power. An individual following that advice is allocating his resources for something that is very unlike his historical experience: he is preparing for an event that he feels has a reasonable chance of occurring in the relative near future. That preparation involves a compromise to allocate resources for a predicated future at the expense of more preferred used of those resources for immediate enjoyment. The compromise comes in the form of expending money and of taking up valuable space ones home for storage, or even modification of the home for preserving that store.
The government policy making likewise will need to prepare entire populations for projected future events whether they be catastrophic or beneficial. These policies take resources away from current enjoyment and reallocates them for preparation with some confidence that the investment will pay off in the future.
I very much doubt the current human-driven decision making governments of representatives or bureaucrats are capable of enacting policies that will mobilize the population for some future not-yet occurred event. Like my description above about bright data, we don’t yet have solid observation of the coming event. We have to make a policy decision based on modeled data.
Certainly, the government can make enforceable policies based on human assertions of their private analysis of data and models, but that analysis is unavailable to the general population. To be effective, these policies must be acceptable by the population so that they actually follow the policies instead of ignoring them or trying to undermine them. In a democratic representative form of government, decisions made by humans inevitably turns the human decisions into an election issue for the next election with a sizable population interested in removing that person because they don’t like the policy. They don’t like the policy because they are not convinced of the competence and fairness of the human making that decision. Consequently, the policy maker will undermine the optimal policy in order to make it less risky for his re-election chances, or for the continuation of control by his preferred party.
We face such a need for a quick policy decision with a viral epidemic. Based on observed data alone, we see that the disease is serious for many infected and fatal for a dangerously high proportion. The current observed data shows that the spread is not easily contained through quarantines, but so far the data shows that quarantine actions are keeping the spread at a manageable rate.
The current government is making policy decisions based on human written policies, many of them written long in the past without any input of the current circumstances. These policies are filtered through individual’s political aspirations whether it be their own election or the change in balance of power between their preferred party and the opposition. It is hard to fault the population from concluding that the ultimate policies are politically driven instead of data driven.
If instead we had a government by data where we trusted an algorithm to process the trusted data to select the proper policy, we would still be hampered if we only allowed currently observed data of daily new cases and daily new deaths distributed geographically.
Unless we permit the algorithms to project the trends into the future, thus inventing modeled data for future values, the algorithms would conclude on policies that may not be distinguishable from the current politically biased decision making. Government by data must permit dark data, data generated by models, as equals to bright data of trusted observations.
It is possible to have a democratically acceptable government by data where the population accepted policies determined by algorithms and data without any human accountability or intervention. The democracy comes in the form of open access to collecting, selecting, and checking both the data and the algorithm for policy making. The output of that algorithm would be the policy that a sufficient super-majority would follow to make the policy effective. Further assuming that we accept algorithmic projections from current trends, we would have policies that would commit the population to various sacrifices to prepare for a future possibility that is very likely but not guaranteed.
Applying this form of government with the data we are seeing now, the best data and best algorithms would likely conclude that there will soon be a time when the disease is completely out of control. Everyone will be at risk of infection no matter what they will do. Many will get symptoms, some severe, some fatal. In such a time, we will need to accept that symptomatic people will need to show up for their jobs or for new jobs assigned to them. We will need to accept that uninfected people will need to interact directly with infectious people with at best partially effective preventive measures.
In addition, we will need to augment key infrastructure with people not currently doing those jobs. To do that, we need to start training those people for their new roles at least as auxiliary to be surprised by experienced people. Most obviously, we need to start preparing a force to help medical professions particularly to administer to the people with severe conditions. Similarly, we need new work forces to assist in distribution of resources and maintenance of equipment for those who are unable to do so themselves. Also, we can predict an attrition across the board for all supply chains where people will be unable to work temporarily or permanently. We need to find ways to fill in those vacancies with similarly newly trained people.
Perhaps an appropriate immediate response would be the solicitation of volunteers is not outright conscription to get people to begin training for roles that will be needed from them in the near future. Given the current conditions of a very small number of actual cases, this is too much of a dramatic action for a democratic republic form of government to make. They will be accused credibly of over-reaction and perhaps with self-serving motives. In contrast, if we had a government by data and urgency, the democratically trusted algorithms using democratically acceptable data may impose such an immediate course of action. And perhaps enough people will cooperate so that we will be ready when the prediction comes true.