Latest Updates

Enabled all pages to be accessed with https (previously only http was possible) through SSL encryption with registered certificate

This is somewhat of a teaser-video, serving as first introduction to Tree of Knowledge platform. It explains the biggest problems facing current statistics, says that Tree of Knowledge can solve these problems and redirects to some other videos for the detailled explanation of how exactly these problems are solved

In order to better explain Fundamental Theory building and how this platform works the Home-, About- and News- pages were redesigned:

This new video shows how Fundamental Theory Building allows the social sciences to move away from making individual, disjoint ad-hoc investigations of specific phenomena and towards collaboratively working on a powerful, shared fundamental theory.

This new video gives a general introduction to Tree of Knowledge and explains all the basic concepts and ideas that make this technology work

This video compares the fundamental theories that are created in Tree of Knowledge to conventional statistical models and explains the 10 biggest advantages of fundamental theories

In this video the biggest obstacle to studying complex systems is explained: not being able to isolate phenomena and the resulting noise. The video also shows how Tree of Knowledge can help deal with this obstacle and thereby open up many areas of research that have sofar been too complex to study

This video shows what a game changer Fundamental Theory Building is: where conventional statistics leads to very unreliable findings (see replication crisis), fundamental theories are highly reliable, rigorously proven and universal

This video is the predecessor of the "How it works" video published on 18th April 2025. It's not quite as good as misses a good Intro.
Creating this video involved learning to use the softwares: Inkscape (for vectorgraphics) and Sozi (vectorgraphic animation).

This video looks at what is special about exact/hard sciences and shows how the Fundamental Theory Building approach is able to produce exact Social Sciences.
Creating this video involved learning to use the softwares: Blender, Elevenlabs and VideoPad Editor.

The platform had been developed in the years 2018 to 2021 in Python 2.7 - the version I was most familiar with at the time. This locked me into a set of python libraries and library versions that all depended on Python 2.7.

Starting in 2023 AWS however stopped supporting Python 2.7 and effectively shut down the platform.
I had to migrate everything to Python 3, which included finding a new set of mutually-compatible libraries and rewriting large amounts of code.

The software Tree of Knowledge was used in two different research projects:

• A project involving modelling people's satisfaction in doing tasks of varying complexity, resulting in this paper
• This other project focussed on traffic simulation

The main take away from these projects is that the further work is required to make this software easily usable in any scenario

In order to explain the novel Fundamental Theory Building approach and it's overwhelming advantages, six different videos were created, each highlighting a different aspect of this approach. These are:

For doing Bayesian Inference, Tree of Knowledge has to run simulations over and over again with different parameter values. For large simulations and simulations with many unknown parameters, this can take quite some time.
To speed up this process, I implemented an architecture where the Tree of Knowledge backend puts simulation tasks on a celery queue and there is a pool of workers (servers) that take tasks from the queue and run the simulations - see diagram below.
If the simulation is very big, then automatically further workers are started and join in.
The code for these workers can be found in this GitHub repository.

There already had been great work done in modelling the probability of a woman to give birth to a child in any given year. For example, this research used a dataset called the "National Longitudinal Survey of Youth" to model this "birth probability".

In order to demonstrate Tree of Knowledge's ability to get more accurate and reliable insights by training on multiple datasets, a very basic test was done - the same birth probability was modelled, but was trained on two datasets instead of the one:

• National Longitudinal Survey of Youth
• Fertility rate by age of mother

You can see that this is very different to a meta analysis, as the two datasets are very different: the National Longitudinal Survey of Youth contains data on individual people; the fertility rate data contains data on age-range-groups.
The test was a success, you can see some of the results below.

Behavior rules calculate the next timestep's value for a variable, using any of the current actor's own attributes or attributes of an agent related to the current actor.
When creating behavior rules you had the option to use one or more of the following options:

• "If Condition" - condition that has to be satisfied for the rule to fire
• "Parameters" - unknown rule parameters that will be inferred using Bayesian Inference
• "Uncertain/Probabilistic rule" - probability for the rule to trigger; also inferred with Bayesian Inference

New option "Aggregation"
This allows variables such as a country's "Average fertility rate" to be calculated as aggregated value from many people/population groups.

New option "Random Number"
This allows for random values anywhere in the rule - see example below. The number is drawn from a uniform distribution between 0 and 1. This allows variables such as a country's "Average fertility rate" to be calculated as aggregated value from many people/population groups.

This webservice/online platform had been hosted on servers from the Cloud Provider "Heroku". The server was getting too slow for running intensive simulations and Heroku however only offers limited options for scaling. It was therefore necessary to migrate the webservice to AWS.

In order to monitor and manage this online platform (especially the Knowledge Base), specific Admin pages were created which are only accessible for users with admin rights.

Result Inspector
These pages allow you to inspect the details of the Bayesian Inference.
Notably, they allow you to see

• How well each simulation performed - how accurately the simulations mimicked the real world/ how well it reproduced the observed data.
• The parameter values that led to specific simulations working well/not so well
• Posterior distributions depending on the threshold ε
  According to Approximate Bayesian Computation (ABC) logic, you can try out different values for the threshold ε and see how the posterior distribution is affected by this threshold.

Simulation Inspector
Simulation rules sometimes have unexpected consequences. It is therefore very important to be able to inspect every/any simulation in detail and see where it potentially went wrong...
From the Result Inspector pages (see last paragraph) you can select a simulation to be shown in more detail. The system will then re-run this simulation while exactly logging everything that occurs, and display the folllowing simulation details in the Simulation Inspector pages:

• Comprehensive table of all values at all timesteps
• Graph to show the evolution of an attribute. It can also show this attribute evolution for many simulations simultaneously - allowing for comparisons and trend analysis
• Executed rules
  For every timestep, it shows exactly which rules were executed in which order - allowing for detailled debugging

The agent's behavior rules is what makes everything run. Tree of Knowledge's core mission is to find the correct behavior rules for humans and other social agents.

A Rule is a calculation of the next timestep's value for an attribute. To allow you to create versatile and expressive behavior rules, you have the following options:

• "If Condition" - condition that has to be satisfied for the rule to fire
• "Parameters" - unknown rule parameters that will be inferred using Bayesian Inference
• "Uncertain/Probabilistic rule" - probability for the rule to trigger; also inferred with Bayesian Inference

The below images will give you more details on the new rule creation interface.

With the Simulation Editor, you set up a simulation. This includes specifying:

• the initial configuration of a simulation
• the simulation duration
• the timestep size

Generally, you would want to run a simulation for one of the following reasons:

1. "Inference from a dataset"
   This is done by specifying a simulation that mimicks the real-world environment in which a dataset was captured. The Inference then happens through fitting this simulation to the real-world data.
   If you uploaded a dataset, then this simulation environment will be set up for you - using the meta data you specified during the uploading


2. "Prediction of a completely new setting"
   You can specify any social system. If you press run, it will be simulated and you can see what mechanisms drive it and how it unfolds.

The Figure descriptions below will give you a better idea of the Simulation Editor's functionality.

Uploading data to Tree of Knowledge involves 6 steps of clarifying meta-data about the dataset. These are considerably more steps than with any other statistics software. On the other hand, once you've gone through the uploading process, the knowledge base knows everything there is to know about the data and handles all data related tasks for you. This includes:

• automatically transforming integrating your data with the existing data in the Knowledge Base
• autmomatically preparing a simulation of the system observed by the data
• automatically managing the inference of this simulation

A tutorial on this upload process can be found here

The Knowledge Base has a specific structure ("knowledge representation") that allows for any data from any dataset to be integrated and strored in this structure.

At the core of this structure are the following elements:

• Object & Object Type
  An object (e.g. Adam Smith) is an instanciation of an object type (here: human). The object type human inherits all properties from its parent object types, which are [object -> living thing -> animal -> mamal] as well as [concept -> legal entity].
  The parent-child relationships of object types are captured by the Object Hierachy Tree - see image below.


• Attribute
  Objects of a certain type (e.g. human) have attributes (e.g. height, age, weight,...) which specify details about this object.


• Datapoint
  A datapoint usually corresponds to a single cell in a data table. Each datapoint describes the value of an object's attribute at a specific point in time.
  Any dataset (crosssectional, timeseries, panel data) can be split up into these datapoints

This data structure/knowledge representation is not only used for integrating all the data, it is also used for the simulation agents - allowing things like simulation initialisation, parameter inference and rule evaluation to be completely automated.

This involved: hosting a django webservice on Heroku; making the home-, about- , news-, contact-pages; setting up user accounts with salted-hash password encryption; registering and setting up the domain treeofknowledge.ai