The future of AI - and why I am thinking a lot about UBI

UBI = Universal Basic Income

A lot of people are discussion this now, and here is my take on how the future will be like, using my background from AI, research on algorithms etc. I am posting this here, so people can comment.

First, let me get one thing straight: Taxation on automation is not going to work.

The reason is that profit margins decreases under competition, meaning that if some company develops an AI that gives them a 30% profit advantage on other companies, over time this will be reduced by reducing prices, perhaps to something like 5%. Governments get a fraction of the profit through taxes, and this is not enough to cover the income of people who previously worked at the same company.

It is obvious: A fraction of a fraction of the same money does not add up to the same amount. This is a mathematical fact. Even if the economy grows there is a finite amount of people on this planet, and after that the only way you can grow is by making new goods and services that did not exist before. These innovations will majorly be done through AI in the long term, which will be widely distributed, and therefore increase competition.

When AI reaches the level of human intelligence on most tasks, it will be trivial to train it to become smarter. We have theories about what it means to be an optimal decision maker. We have theorem provers, domain specific languages and lot of stuff that we could ask the AI to improve. Once we have an AI that can do most physical labor, the step to go beyond the smartest humans is pretty small, since most of our brain capacity is spent dealing with the complexity of the world, not thinking. Ordinary physical labor is the grand challenge that AI needs to overcome, and beyond that the path is pretty clear: It is far easier for a machine to maximize its thinking capabilities than it is for humans.

In the toolbox we have for thinking in computer science, there is reasoning through logic and probabilities that are well developed, while concept modeling and reflective reasoning are sorely lacking. We know how to make deep belief network become smarter by looking at a lot of data, but an artificial reflective reasoner would be able to make use of small amounts data, some physical interaction with the world, and a ton of computer capacity. Beyond some limit of required data, there is little need for input for it to optimize its own intelligence.

At some point, assuming a system that improves its own thinking capabilities, it will find observations of the world less interesting because there is less to learn. One could say that the AI start to “get how it works”. What matters is access to real world data, but in the domain of physical problem solving it will be some algorithms that are well suited and all you need is applying more computer power to get stuff done.

This level of AI is sometimes called “superintelligence”, which when not engineered properly, might be an existential risk for humanity. It is kind of hard to imagine because there is no such thing today, not even close.

So, here is an example to make it easier: Imagine the intelligence of various geniuses in early 1900-1950, on the most abstract areas like logic, proof theory, computer science and theoretical physics. For example, Bertrand Russel, David Hilbert, Alan Turing, Kurt Godel, John von Neumann and Albert Einstein. Then you take the brute power of these brains, and put them on a computer chip, communicating with each other at a rate near the speed of light.

A superintelligence has no need for sleeping or eating. All it needs is energy, which it can get plenty of, all by itself, if given the opportunity. There is no need to instruct in detail how to build new devices for getting energy.

For example, assume superintelligence was invented in the year 1900. If I remember correctly, solar panels use the photoelectric effect, which was first understood by Einstein. If the computer program have the intelligence of Einstein, it could figure out how it works, and then use the intelligence of other smart people living in that same period of time to come up with something that looked like a solar cell. In the beginning, it might not be the optimal design, but over time it could make improvements. These improvements can be made faster than any human can do, because the communication and reasoning happens a lot faster.

This brings us to the control problem: How do we make sure that it does what it is supposed to do?

Some people believe you can control superintelligence by wiring it to your brain. I believe this will not work at all. It is like trying to use a floppy disc with a smart phone. I think that wiring a human brain to a virtual reality is a possibility, but not superintelligence, because humans are not built to do that. You might try to enhance the human brain step by step, by swapping some brain tissue, synthesize the pathways on chips, and plug it back in with a brain interface. This is a lot more complicated than just using a self improving program, which you give commands by having conversations and visual demonstrations of what the superintelligence works on.

Superintelligence is not going to solve every problem we have the first day. Some problems are extremely computationally expensive to solve, so it might take years before we have the capacity. I don’t believe that superintelligence will take over the world, because with a reflective understanding of itself and human values it should be able to restrict itself, if its goals are soft and not rigid constraints.

What I am worried about, is some new kind of virus or a “silly idea” done by some humans, that could wipe out humanity or take control over the Virgo supercluster of galaxies on a fast track. It could easily lock down the faith of humanity for hundreds of trillions of years. One might not think that is not such a big deal, because you will be long dead and humanity drastically changed, but think a bit about the long term future of existing in the universe: You need energy to move, and as long there is energy, there can be life. In the long term most of the observable universe that we see today could be inaccessible, so all the energy we will have could be restricted to the Virgo supercluster. Would you sacrifice all available energy in the future to have some “fun” in the 21st century? I would not.

Our current thinking about AI is dominated from the perspective that we learn through movies and media. We are terrified of economic instability resulting from loss of jobs. I do not think “figuring out what to do with our lives” is the big problem here. It is more like “how to prevent society from total collapse” in the transition phase.

It is not like waving with thumb rules as “technology creates new jobs” is going to work. There is some research that seem to indicate an upper boundary of how much you can exploit very smart agents in a competitive environment. This means that in order to keep up, you would have to improve your own decision making to get an income in a pure free market. Humans are not very good at being rational, and very far from ultra-rational. However, this is an area that AI will excel at, so it can keep improving its own rationality beyond the ability of humans. By rationality I mean epistemic and instrumental rationality, which are frameworks for how you gain optimal skills for learning the environment and achieving goals.

Superintelligence won’t need money, because it can be its own supply chain. It could take raw materials and produce what it needs for various purposes in the other end. There is no way it would find the current economy an efficient way of getting stuff done for the long term.

This brings us to the topic about what we should do in the transition period: What will happen to the economy?

As I pointed out before, there is no solution that makes the current economic system with all its political baggage sustainable at near complete automation. There are several reasons to believe this:

  1. Humans are heavily indoctrinated to accept the current economic system at face value, without questioning
  2. Politics have gotten more and more decoupled from real issues for most people, and does very little service except for those with a lot of power and influence
  3. Corporations are mainly concerned about short term profit and maximizing stock values
  4. The level of fundamentalist thinking with strong authoritative figures, drawing their power from economic downturns, brings instability to countries where the well being of majority of people is ignored
  5. People without an income are excluded from society, making it near impossible to raise a voice for important decisions
  6. Mass media uses fear mongering tactics to make profits, and this leaves a lot of people without the ability to get constructive feedback on their opinions, without involving themselves in research that is beyond their current understanding

I have no hope to bring the general population up to date with the situation. People are smart, but they do not have free hands to make the optimal choices under pressure.

However, there is one idea that people at least get their head around: A basic income for everyone to provide for basic needs.

So, I want to point out some things that are less known about basic income:

  1. Basic income IS NOT a simplistic model where you tax rich people and give it to the poor
  2. Basic income IS a regulatory mechanism that can both incentivize and deincentivize people to work

For example, if you want people to work, then they need education. To educate people, you give them what they need to study and design an environment to support learning. You can not do this and let the same people starve at the same time! Giving people what they need to get an education is a way to incentivize people. You don’t have to tell people to educate themselves, they do so on their own, when given the opportunity.

What people do depends on how you distribute resources. In my country, Norway, while being far from an utopia, there is a social security net helps people find a new job, get treated for physical or mental disorders, get an education, and start a company. I have been in all these situations, receiving help from the government, but it has never made me less ambitious or less wanting to do good for society. On the other hand, I think I am getting more and more eager to help as I have gotten older. Somehow, I feel what I do today is not enough, since I have so little resources and little time, but I assume some of these problems will get fixed as AI advances. I am thirsty for achieving more great things!

The idea that rich people will reduce their fortunes under basic income is also wrong. Most of the fortune that rich people have comes from people being better off in general. For example, results of academic research that rich people would have to pay more to get done otherwise. Besides, feedback effects in the economy might also lead to super rich getting a LOT more super wealthy. It depends on how the system works.

Many people have the simple, but wrong idea, that if you have some money, then giving it away, you will loose all of it. If you give away some money, and other people make better use of it, then you could save it back later, ending better off than by not giving it away. Just look at banks, and see how they lend out money to everyone that can afford it. They do it because keeping the money themselves without anyone using it gives less profit. Governments print money all the time, essentially because people can make better use of it than the government. They even buy properties from the government with the money that the government gave them in the first place! Poor people often translates money into meals and clothes, so they don’t see how money works on the grand scale. It is an information system to forward goods and services, so everyone in society can benefit. No one believes that rich people are those who keep systems running. It is the general population that is the pillar of society, and rich people are just those who profit more than anyone else from it.

So, when extremely rich people see new concepts as basic income, they would be very stupid to think it is charity. If I was an evil dictator, I would implement a basic income for everyone in order to make my self richer! People get money on their hands but this accumulates back to piles that I can use to control those who got a taste of wealth! Mwuhahaha!

Most people are not like this: I am getting a paycheck every month without working, so I am going to play video games all day long.

Most people are like this: I want to learn something new, start a family and helping other people.

The reason some people isolate themselves from society is often that they don’t get the opportunity to make advancements in other areas. For example, the educational system fails them, the economic system fails them, and they fall outside the system. They become invisible, non-existing for most people.

Some people think that poor people are lazy and rich people are clever, despite the very obvious fact that given a large population, some people get unlucky and some get lucky, simply by chance. There is much less “blame” to put on people that are in a worse situation than yourself, at least compared to what you get impression from media.

Alan Watts called this “the freedom to make mistakes”.

With current pace of technology, we can look forward to abundance and prosperity. Will we say no to this future, just because we feel it is uncomfortable to think outside the box? Of course not!

The economic system is basically a distributed algorithm with people’s actions as input, and distribution of wealth as output. There is no special law that is for economic systems, that does not apply to algorithms in general. It can be studied and improved just like any algorithm, and the tools we have keep getting better and better.

Today, we know how to secure crypto-currencies through proof-of-work. We know how to “create money” out of algorithms, in the most literal sense. We know how to make these systems secure. We know how to use fingerprints and facial recognition to identify people. We know how to produce certificates that proves that people are who they say they are. We know how to build every piece of the system that is required to regulate the economy and incentivize people, even with strong influence from rapid advance in AI technology.

What we are lacking is the architecture, the overall concepts that makes it all work. This is a difficult problem, because the world is extremely complex. However, instead of thinking “politicians should do something to fix the economy”, which is not going to help at all, since they do not understand how to fix it, the problem is “we should find a way to get more accurate understanding of how new concepts for economics behave in complex environments”.

If we could get more accurate beliefs about these things, then I have no doubt that we could make the necessary technological steps to implement it and make it work in the real world. I don’t think that you can fix the old system through gradually changes and see what happens, because we don’t have enough time. It is much quicker to start from scratch and design it from the ground up, then copy ideas from it and adapt existing systems. For example, stock markets could relatively easily be modified to support a new technology, once it is has proven itself on smaller scale. Heck, you could design as an investment opportunity, which is something crypto-currency is already doing.

Some people have doubts about convincing others to switch a new system. Let’s say Elon Musk or someone else makes a new kind of money, called “Capitalism 2.0”. Will people switch to it?

I don’t think this is a problem. Just look at how fast Facebook made it in recent years. If people get the choice between “living on the street” vs “being socially included, supported and secured basic needs” then it does not take that much to decide which option is the better one.

Then there are people that say that “cleaning toilets is no fun, so people will rather stay home than do it against marginal benefits”. They picture that you tax the people who works, so they feel the “burden” of pulling the weight of others.

You know what? This is just stupidity. I have no problem seeing people doing the “dirty work” if they are well payed. Some algorithms give long term rewards to people who work , such that the margin they get from working pays off over time, while the employer only need to pay a tiny bit of it. For example, a progressive negative fortune tax! Another idea: Let us say that people work a week, and then go on vacation for 3 weeks, while other people take their turn. A system could give the people who work free tickets, for things to see and experience around the globe, with free transport and traveling in groups with friends. Such things are possible because technology makes it dirt cheap!

Cars, trains, planes and boats can be automated, and they can be produced by automated factories, using raw material from automated mines. The software to do all this can be open source and constantly improved. Resources can be licensed to non-profit organizations which whole purpose is to make efficient use of materials. So, if you are very into the use of a specific resource, you could serve as an expert for one of those organizations against well payed rewards. Advanced in nanotechnology will lead to cheaper materials, we have an excess of CO2 in the atmosphere that could be put to better use. It is, after all, an important ingredient for the materials that nature uses in biology.

Fear is a strong emotion that can make people become hysterical and easy to control through indoctrination. Children today are much more anxious about the future than they were 50 years ago. I am scared of what the future might bring, but I am not going to let this control my decisions! Getting a better understanding and working on the problems is always going to be the best option. I hope you also choose to think stuff through before making up your mind.

It is not a solution to the economic problems by building a bunker under the ground with oxygen supply. We don’t go back to the old ways to produce food, just because of some economic uncertainty. What people are making decisions about today are many order of magnitudes of wealth beyond “do it yourself”. We make decisions to save minutes, or to reduce costs with a few percentages. The constraints we have are near local optima for society, and we don’t need to go to the bottom of the graph in order to find a higher local optima. All we need is a substantial thinking outside-the-box for a decade or two, and I am sure we will find out how to deal with the transition from the current economic system to something that works with high level of automation.

Good luck, whoever you are! It is going to be an exciting time to live in.