Virtual evolutionary organizations are organizations that reside in virtual space which leverage evolutionary computation. Evolutionary computation can take the form of human based genetic algorithms, interactive genetic algorithms, or any combination of evolutionary processes which result in a continuous state of improvement of components of it’s architecture. The result is a digitized form of smart architecture which evolves to suit the preferences of it’s participants.
An early example of human genetic algorithm would be the Free Knowledge Exchange Project. This project utilized the human being as a component in the system. A couple of quotes below provide a description of the Free Knowledge Exchange project:
The Free Knowledge Exchange (FKE) project intro-
duces the concept of evolutionary knowledge manage-
ment based on concepts of GA. It used a human-based
genetic algorithm (HBGA) for the task of collabora-
tive solving of problems expressed in natural language
(Kosoruko , 2000a).
The selection of problems is performed according to
their importance, based on expressed interest of par-
ticipants in each particular problem. This measure of
fittness based on the summed interest of all participants is used to include a problem into the generated web pages shown to people. This process happens in inter-faces of HBGA, which generate the interactive WWW pages dynamically. Roulette wheel selection method is used for this purpose. In this way, the problem in which many people are interested will appear in the interfaces more frequently. The frequency of appear-ance of the particular problem in the interfaces and in dynamically generated WWW pages can be thought asa measure of attention the system pays to a particular problem.
Today we are in the cusp of distributed autonomous organizations yet so far not much discussion has been had on how to integrate evolutionary computation processes into distributed autonomous organizations. This article is my attempt to contribute to and help initiate that discussion.
Distributed Evolutionary Organizations
Distributed evolutionary organizations (DEOs) are a new breed of social organization that can leverage the trends such as Big Data, quantified self, Moore’s law, artificial intelligence, Reed’s law, and even the technological singularity if such an event should take place. These organizations could potentially symbiotically augment the human being through a co-evolutionary process. As humans provide more feedback to the distributed evolutionary organization through their participation the organization can intelligently evolve to fit the preferences of the human participants within the organization. This type of organization could also be described metaphorically as a distributed smart organization because like a the smart home the virtual organization would respond to the preferences of it’s inhabitants.
Distributed Evolutionary Organizations represent a new sort of digitized smart architecture which conforms to the wants and desires of whomever is using. This could for example mean that the user interface continuously adapts to how the user participant likes to interact with the system. Personalization of interface design could be achieved through a personal interactive genetic algorithm so that the interface component of the distributed evolutionary application/organization learns the favorite colours, patterns, textures, icon placement, widget placement, and more of the participant over time. The personalized interface would be continuously self generating until it approximates to fit the aesthetic of it’s owner.
A lot more can be achieved through a distributed interactive genetic algorithm where the sum of all users preferences become feedback for a process of artificial selection. Over a period of time the designs will become increasingly less wrong, less ugly, until it reaches an approximation of universal fitness based on democratically selected search criteria. So in essence design creation which is usually considered a creative process can actually become an evolutionary algorithmic process and this algorithmic process can even be used on the design of the entire distributed evolutionary organization itself.
One of the fears many would have is the run-away AI danger. In order to resolve these fears it is important to note that the human participants and their preferences remain the driving factor in the evolutionary process. Human beings will be able to vote, rank, rate, discuss, every single component within the distributed evolutionary organization. Humans can rely on reputation, prestige, accountability, and other criteria to give different weight to the feedback from different humans so that overall the best human feedback from the most trusted and qualified rises to the top. If this is in place along with making sure every component is open source then what you can have is an organization which evolves within specific constraints.
Learning Conditional Preference Networks and the power of feedback
Feedback is one of the key mechanisms in a distributed evolutionary organization or virtual evolutionary organization. In a distributed evolutionary organization (DEO) the feedback is acquired from the software components. This requires that the design of the DEO be as modular as possible so that any component can be switched in or out. Every component must also be capable of collecting feedback from participants interacting with in such as a simple feature rating system, an algorithm rating system, or abstract design rating system. Feedback can also become an automatic process through software agents combined with learning conditional preference networks as a way around the user fatigue problem.
Automated voting and learning conditional preference networking
The Amazon Echo or Siri was used as an example because it’s the only practical popular examples of autonomous agent AI. Autonomous agents can go much further than Amazon Echo/Siri and if preferences can be known by an AI then the AI could act as a delegate to the human participant. The autonomous agent could vote on behalf of the human according to preferences the human inputs into the AI or the AI could attempt to learn the “self interest” of the participant using the continuous stream of data the participant provides. It is currently well known that algorithms know us better than we know ourselves and if the trend of Big Data continues it will be likely that our autonomous agents could know our self interest better than we do for ourselves. Why not vote by algorithm if this is the case?
Interactive Genetic Algorithms can be used for preference guided self optimizing components
In a proper distributed evolutionary organization every component can receive a rating. This would allow voting on every component, every design, every algorithm, even the source code, so that humans are continuously auditing, rating, judging, reviewing, every part of the distributed evolutionary organization. The role of the interactive genetic algorithm is to use AI to augment human creativity in certain processes such as algorithm design, aesthetics, and more. The genetic algorithm will generate candidate solutions which the human participant would interact with by simply rating each solution. These preferences could then be recorded to a blockchain or distributed ledger anonymously as a number pattern which represents the unique preferences of that participant to the personal interactive genetic algorithm or to a distributed interactive genetic algorithm.
Evolutionary Distributed Self Optimizing Organizations/Applications
Each and every component within a self optimizing organization is in a continuously redesign state. As participants within the system use the system the system is trained to evolve toward the preferences and use patterns of the participants. Components which don’t get used very much do not get trained and as a result will lose design priority but if a component is used a lot the system would know from this feedback that it is a highly valued/prioritized component according to the human participants.
If for sake of example the system is a distributed secure communications forum then as people use each feature they’ll have the ability to rate every feature. This rating would give a numerical value to every feature in the system. This rating provides feedback and to avoid user fatigue these ratings do not have to be manual but can be deduced by anonymous usage patterns which create models for different kinds of participants in the system. This information would allow the human developers of the distributed self optimizing application to have a continuous stream of anonymous feedback, at the same time the system itself could be autonomous enough to allocate upgrade funds according to actual usage patterns.
Learning conditional preference networks are a technical mechanism for producing ordinal representations of participant preferences. The learning component allows for a process of continuous improvement so that by algorithm the AI can learn what the participant likes and or wants. It is in the learning conditional preference network component that the human and AI interact closely with the AI functioning similar to Amazon Echo or Siri as a smart autonomous agent for the preferences of the participant. This learning conditional preference network can if it becomes smart enough help to relieve user fatigue and also may even provide for a democratic process which takes place in the background between the participant’s conditional preference networks.
Code.google.com,. ‘Deap – Distributed Evolutionary Algorithms In Python – Google Project Hosting’. N.p., 2015. Web. 5 Mar. 2015.
Kosorukoff, A., & Goldberg, D. E. (2002, July). Evolutionary Computation As A Form Of Organization. In GECCO (Vol. 2002, pp. 965-972).
Wikipedia,. ‘Collaborative Filtering’. N.p., 2015. Web. 5 Mar. 2015.