Token Model Simulation Series
#0 Why Simulation?
#1 Fools Agreement Part 1: Introducing ‘Fools Agreement’ and Simulation Environment
#1 Fools Agreement Part 2 : Simulation Result Analysis
The photo above is a screen shot of the engineering design & simulation program Midas Gen. Architecture engineers use such complex simulation programs to verify the sustainability of buildings in various environmental conditions when designing a building.
[Fyi] Midas Gen is the product of Midas IT, famously known to be one of the best companies to work in Korea, and is acknowledged world-wide, even being used for designing globally famous buildings like the Burj Khalifa, Beijing Olympic Stadium, and Beijing International Airport.
Simulation is crucial in architecture design because once a building is built, it is difficult to change anything. Also, it goes without saying that one mistake in the design may lead to catastrophic consequences involving the lives of many.
This is why engineers use these programs to expose their design in unpredictable and complicated interactive environments to test safety before their design is materialized.
But, what about for blockchain protocol?
If architecture/engineering design is needed in building a city, token model design is required for establishing a decentralized network because a decentralized network on a blockchain operates not by the order set by a central authority but by the order autonomously formed by the market.
Therefore, teams developing a blockchain have to meticulously design a token model in order to achieve their desired objective.
Blockchain protocol design is an Internet based technology but it is more like architecture design mentioned above, rather than existing Internet services design.
Flawed design can be catastrophic
Because buildings could pose a threat to the safety of people, it cannot be built just for the sport of it. This is also the case for blockchain. Blockchain fundamentally aims to become an infrastructure on which assets of value, not simple data, are exchanged. A flaw in the design may liquidate people’s assets into thin air.
One famous incident is the DAO where tokens worth roughly 75 billion KRW were hacked. Designing a token model involves greater responsibility than designing a platform for posting pictures or sending messages.
It is hard to make fixes.
The most known advice to start-ups is ‘execute quickly and get feedback earlier.’ Unfortunately, this approach cannot be applied to decentralized network.
Because a single entity does not have control over the network, it is difficult to change the rules once they are set. Modifications cannot be easily made even when changes in external circumstances that lead to problems in the economic system occur.
Similarly, building structures cannot be changed because a problem is found after construction is complete. Extreme caution is required from the beginning of the design effort.
Too many unpredictable variables
A token model designer has to presume what results this model could bring. He assesses using tools like game theory what kind of decisions rational users would make. However, the forecasting ability of game theory diminishes mostly when the environment deviates from a simplified, theoretical model. Token model is a complex system where tens of thousands interact, which is why it is extremely difficult to find a solver.
As many variables such as natural disasters and interior load have to be considered when designing a building, there are many, complicated variables that need to be thought about when designing a token model.
Like architecture, token models need to be simulated before deployment.
Trent McConaghy wrote in ‘Towards a Practice of Token Engineering’:
Simulators came on stream in the 1970s and CAD tools in the 1980s; and no one’s looked back. These tools are crucial to modern chip design. It costs >$50M to manufacture a design on a modern process; it would be, well, stupid to not verify and optimize that design to the best possible level before committing the $50M.
Yet in the world of token design, we are building and deploying what we hope to be billion-dollar ecosystems, with barely any tools. It isn’t even 1970 yet.
As McConaghy puts it, if there is no design verification and optimization when designing a blockchain and its key engine, the token model, then we fall short to the level of the 1970s when semi-conductors were produced for millions to use depending only on the intuition and logic of people.
It is difficult, but needs to done.
Decon has always had a need for simulation as we designed token models for blockchain projects. However, it is not an easy task to simulate a network where countless free entities interact, especially the blockchain of today.
First, there is extremely little actual data to utilize in simulation. Blockchain is not yet being widely used and blockchain based services have yet to be commercialized, which is why there is an absolute deficit of data needed for simulation.
Second, the entities reacting to the blockchain mechanism are people, who are inconsistent, sometimes utterly selfish, sometimes selfless, at times reasonable, but also at times affected by unforeseen psychological factors. It is hard to model and predict human behavior.
But that does not mean simulation itself has no value. At least it is better than having nothing. Simulation cannot be perfect, but having it readily available as reference helps us make better decisions (designs).
Think about economic forecasts. Governments and central banks spend substantial amounts of money and hire economists to build models to predict growth rates and inflation rates, but most of the predictions are wrong. Still, these models are not totally useless because better decisions and agreements can be made based on these models.
Decon’s approach: Agent-based simulation with Reinforcement Learning
That is why Decon seeks to create a token model simulation by combining Agent-based simulation (ABS) and Reinforcement Learning.
ABS is commonly used in simulating social dynamics, aiming to explain complicated marco-level phenomena with interaction results of micro-level behaviors. It does not set systems, rules, nor orders top-down, and does not have a pre-determined equilibrium to reach as an objective.
An advantage of ABS is its adaptive decision making in accordance with circumstantial changes. Methods used to predict macroeconomics such as dynamic stochastic general equilibrium modeling (DSGE) selects a representative agent only show fluctuation around a given equilibrium point, and is difficult to predict abnormal circumstances like stock market crashes or economic recession.
However, agents of ABS do not just comply with predetermined behavior rules, but modify its strategy and adapt. Such feedback mechanism allows us to predict whether agents will shift their strategy to advocate the majority trend or make irrational behavior in panic.
Also, ABS uses reinforcement learning as a component. Reinforcement learning uses a machine learning method that maximizes rewards in a given environment. Using reinforcement learning, it is possible to make agents in ABS learn how to find an optimal strategy on their own with their acquired experience and data.
By utilizing reinforcement learning, there is no need to pre-define the behavior strategy of agents. This enables a more realistic simulation that requires quick adaptation and strategy development in an ever-changing environment.
In following posts, we will introduce the results of Decon’s token model simulation. At first, we inevitably had to start with a model that simplified many portions. However, through numerous tries and lessons learned, we accumulated knowledge and will continue our research to create a more realistic simulation.
If blockchain services become commercialized in the long-run, we believe simulations will be a prerequisite tool. We still have a long way to go. We appreciate your support as we take our first step towards progress.
