Blockchains will save us from A.I.

Since the emergence of money, the price mechanism has acted as a coordinating signalling mechanism between humans by creating incentives to reallocate resources from areas of surplus to areas of scarcity. Like diligent ants, humans construct supply chains, engage in marketing, dig up earth, set up factories and even perform tasks they consider very unpleasant all in service of prices. On a society wide level, the name we give to this emergent intelligence is economy.

The economy is such a complex and vast super intelligence that individual human intelligence is completely dwarfed by it. I’ve spent a great deal of my adult life trying to fathom the genius that is the economy and one observation I’ll share is that the best economists are those who approach understanding it with utmost humility. Explaining the economy with models and equations is like trying to explain a human with all its emotional complexities using only an IQ test score. Furthermore, fitting your ideology to the economy is like projecting your psychological analysis onto someone you barely know simply because some of their behaviour reminds you of your own preexisting biases. The economy has grown so complex that no individual human can even construct the simplest of products entirely from scratch. We act not as individuals but as members of hive whose purpose we aren’t even sure of. Can we really answer questions like “Is the economy good for humanity/the planet/other animals?” We can speculate and give short run snapshot answers but we don’t really know where the economy is leading us. We know a few things, though. We know that the more one participates in the economy by responding to price signals, the higher the probability that they’ll be rewarded materially which is why political pressure to “create jobs” is always at the forefront of any election cycle.

Enter Artificial Intelligence

With the advent of self improving machine learning, the latent existential dread of AI is finally moving from paranoid science fiction to public consciousness. In particular, there is an immediate fear that AI will lead to mass unemployment as even white collar jobs are threatened. What most commentators miss is that the methods by which A.I. learns is an imitation of the methods that the economy uses to teach individual humans. This insight will help us understand how blockchains can be used to boost the intelligence of the human economy. To understand why, we first need to compare AI learning to economy-driven human learning.

Machine Learning

To simplify machine learning, the current approach to training machine intelligence is to simulate ecological selection by rewarding choices that are correct and culling choices that are incorrect. The workflow to train an AI is:

  1. Species: A software agent is created that can internally calibrate itself without human direction.
  2. Population: millions of randomly calibrated instance of this software are spawned.
  3. Environment: researchers provide the population of instances with vast datasets and provide meta information about the data. For instance, an A.I. trained to recognize red cars is given a vast library of car images, each with a colour label attached.
  4. Selection: the instances are asked to answer a question (pick a red car) and the least successful instances are culled.
  5. Mating and reproduction: the surviving instances are often used as templates to produce instances that can answer more nuanced questions (e.g. maroon bus) and are provided more data. At this step the researcher doesn’t know why an instance is accurate but can only observe that it is accurate. This feature is what distinguishes AI from regular software.

Economy Learning

In order for the complex economy intelligence to emerge, individual humans have to be trained using a similar evolutionary algorithm. In the case of machine learning, human researchers carefully curate and select artificial behaviour. In the case of human economy, scarcity performs this selecting role:

  1. Species: humans can perform physical tasks in response to internal experience. Given time, they can calibrate their plans and skills to achieve different outcomes.
  2. Population: since humans can coordinate in small numbers without outside assistance, economy as an intelligence does not tend to exist in very small groups. On the contrary, economic intelligence appears to scale quadratically with human population (Metcalfe’s Law).
  3. Environment: Faced with resource scarcity and choice surplus, humans establish ratios of scarcity (prices) so that they have a dynamic dataset on which to form decisions.
  4. Selection: The goal of any human in an economy is to either accumulate the greatest abundance of resources in the least unpleasant manner (whatever that means) or to achieve the greatest amount of satisfaction (whatever that means) using as few resources as possible. These 2 approaches are logically equivalent. The constant mediating force is scarcity. Humans achieve this by conceiving plans and then applying those plans. Plans that use more resources than they bring in are culled by scarcity (loss) and plans that bring in more resources than they use are selected by scarcity (profit). This applies to all economic behaviour, by the way, not just entrepreneurship.
  5. Mating and Reproduction: Good plans are emulated (market entry), improved upon (competitive innovation) and sometimes spawn superior offspring in one way or another (disruption).

Now we have identified what guides improvements in machine learning (selective researchers) and what guides improvements in economic intelligence (scarcity). It would appear that every emergent, intelligence requires important forces or actors who guide their development. Let’s abstract this concept and refer to these actors as Guides.

At the time of writing, the progress of AI is limited by the number of guides that exist. There are only so many human researchers involved in the guidance process. Imagine how much quicker AI would develop if it could just spawn more guides on demand. There would probably be a runaway machine learning effect. But this is currently impossible since human guides can’t be spawned on demand.

However, the human economy isn’t guided by researchers but by scarcity as noted above. And this is why blockchains have given us a head-start in the race against automation. Blockchains give us the power for the first time in human history to create scarcity on demand!

Scarcity as a Service (SaaS)

As alluded to in a previous article explaining how Bitcoin has already created a self improving intelligence, scarcity as code means that incentives are now programmable by engineers. We no longer are bound to passively respond to the true scarcity of nature but instead can spin up scarcity out of code, giving birth to intentional economies designed to serve specific human needs.

Hypermediation

One of the common misconceptions about smart contracts is that they’re self executing code that live on the blockchain. It’s better to think of the blockchain as an inert machine that can perform calculations and smart contracts as the triggers that activate the blockchain into action. Once the smart contract has run its course (this has to happen in the span of a block), the blockchain returns to being inert.

The Oracle Dilemma

Since smart contracts cannot self activate and since they are unable to query the broader internet, they are reliant on humans to activate them and to supply them with external information, if necessary. On the surface, this appears to break down the trustlessness nature of the ecosystem but as we shall see this turns out to be an immense strength of design. Traditional centralized services such as Uber and Airbnb pretend to dis-intermediate the middle man by essentially centralizing the middle man. In an ecosystem of programmable incentives, we will be able to coordinate a networked intelligent of middle men, a phenomenon Max Borders has coined hypermediation. To illustrate the magic of hypermediation, I’ll sketch out an example.

The Hero Gotham needs

Suppose we want to create an decentralized application called BatmanDapp. The purpose of the dApp is to act as a specialized subreddit for all things batman. One of the rules of membership is that all posts must be batman related. Suppose one of the users starts posting spam. Other users can report that user to have them banned. Of course there exists the risk that false positives are banned. Perhaps members are reporting someone because she expresses opinions on batman that other members disagree with. In a traditional site, there are arch-moderators who can intervene and adjudicate. These are usually employees or owners of the app. But we want BatmanDapp to be 100% decentralized and self organizing and yet we don’t want to allow for the possibility of abuse or hijacking by bad actors. How could this be achieved? Consider the following strategy:

  1. Require all members to place a small cryptocurrency deposit to gain admittance to BatmanDapp. When they leave, they can claim back their deposit.
  2. When a member suspects another member of acting in contravention of the rules, the plaintiff can lodge an appeal of expulsion against the defendant by placing a crypto deposit in a smart contract equal in value to the membership deposit.
  3. The defendant has a week in which to respond to the appeal. If they do not respond in the given time frame, the smart contracts that constitute the dApp will treat the defendant as expelled. Half of their membership deposit is forfeit.
  4. If the defendant wishes to challenge the expulsion in the given time, they can activate their defense by selecting a judge. A judge can be any person whom they trust.
  5. Once the defendant selects a judge, the plaintiff must choose a judge to act in the interest of prosecution.
  6. The 2 judges then coordinate to select a third judge whom they both trust. The 3 judges can then request both parties to submit evidence. After deliberating together, they have to deliver a unanimous verdict.
  7. Upon the verdict, the loser’s deposit is split between the judges (loser pays). If the defendant loses, they are expelled. If the plaintiff loses, their deposit is forfeit.

There are 2 insights to take away from this layer of hypermediated judiciary. The first is that these judges are free floating agents who accumulate a reputation for fairness. They need not be qualified or belong to some special and protected group as would be the case with employed site moderators or the judiciary of modern nation states. They need not have any interest in Batmandapp at all. They simply need to be willing to rule according to the rules of the dapp. The second follows from the first. The judges are guided by incentives programmed into the judiciary smart contracts. They scurry like ants in a colony, doling out blockchain justice wherever they are needed. We’ve replaced central moderators not with a decentralized judiciary but with a layer of judicial hive intelligence, mediated by the scarcity of cryptocurrency tokens.

Build your own Human Hive

Traditional companies funded by ICOs are not particularly interesting or novel other than in the manner in which they circumvent restrictive financial regulations on investment. The magic of blockchains and smart contracts come into their when ecosystems of smart contracts are webbed together to create a pattern of scarcity that motivates humans into action without central direction as in the case of bitcoin mining or in the fictional example of Batmandapp.

When mobilized in this way, humans are no longer individually competing with AI. Instead we begin to act as a unified super intelligence. The human economy is becoming self aware and it seems that AI is a very long way off from threatening to displace so complex and formidable an entity. If anything, the human economy will consciously mate with AI to produce something even more complex. Perhaps this offspring is what we call the Technological Singularity.