From Michael Nielsen’s book on neural nets

Neural networks as the architecture of human work

In the brain there are neurons that link. This firing together creates a connection. On the Internet there are servers and people that are linked in temporal interaction, sometimes as a result of being interested in the same topic, thus creating contextual interdependence. This short-term communication, firing together, sometimes leads to a longer-term relationship that increases the strength of the connection. No neuron links with all the other neurons at the same time. No server links with all the servers at the same time, and no one person interacts with all the other people at the same time. So all interaction is always “local” creating “events”, whether in the brain, in an organization, or on the Internet. However, local here does not mean spatially local. The nodes in local interaction can be physically located in different parts of the world.

All societies and all enterprises have headed, at least to some extent, towards specialism. The assumption has been that the further a division of labor is carried, the greater are the savings and the better the quality of the contributions. This has led management to focus on the efficiency of activities separated from other activities and organizational design has been seen as the planning and execution of a collection of activities forming the organizational system.

The mainstream management paradigm is accordingly based on the presupposition that activities are the independent, governing factors and the scheme of interaction conforms to the planned division of labor as a secondary, but equally important feature. The organizational structure, as a number of activities, comes first. Then an appropriate system of co-ordination and communication is put into effect. If, however, action and interaction are mutually dependent, it means that low-quality interaction leads to activities that are poorer than planned, just as enriching, high-quality interaction may lead to higher-value activities than planned.

The present ways of dividing labor have historically been based on a very different communications environment than the one we are living in at present. The earlier high cost of coordination and communication is the reason behind many of the organizational forms that are taken for granted and which we still experience. The digital world we live in today is totally different when it comes to the quality and costs associated with coordination, communication and contracting and allows us to experiment with totally new value creation architectures.

“The cost of communication is reduced by more than two orders of magnitude. In other words, it’s possible to send 100 to 1000 messages between services in the same amount of time as communicating and processing one message would take a decade ago.” (Adrian Cockroft)

Resource allocation has always been one of the main tasks of management: what is to be done by whom and when. In centralized systems and with homogeneous resources, this allocation can be performed top-down and in advance of action, separately from the people who act. When knowledge and creativity are the decisive factors of value creation and when work takes place in digital, global, decentralized environments, this top-down process is increasingly inefficient.

A manager cannot know who knows or where the most valuable contributions could come from. This is a problem because time to value is an increasingly important metric.

Because of the aforementioned growing needs in daily organizational life a new architecture of work is emerging. Ideas and practices from post-blockchain smart contracts, artificial intelligence, machine learning and neural-networks are being adapted not only to work, but also to human work.

The new event driven, layered architecture refers to a new, open production logic: people from the whole network can contribute pieces of their time, creativity and expertise to ongoing events according to their interests, availability and experience, working in a transparent environment. The black box of machine learning is turned into an open field. The hidden layers are not hidden anymore. Work is a movement in time where some contributions are followed up by others and and some are not, creating a storyline and a developing narrative. The patterns are not caused by traditional competitive selection, or independent choices made by powerful agents. Instead, what is happening happens in networks of peer interaction.

The product is the dual process of problem definition and problem solving.

This method has systemic advantages over traditional production hierarchies when the work in progress is mainly creative and the capital investment involved is distributed. For most knowledge-based contributions and creative services, this kind of production is the most efficient method from a resource allocation point of view.

The system is developed as much in a bottom-up manner as a top-down one. In a top-down system everything is provided by the organization to the worker. The worker has no or very little control over what tools, information and people are available to him. Instead of forcing people into predetermined groups in the way Enterprise Resource Planning (ERP) does, here Artificial Intelligence (AI) and Machine Learning (ML) technologies mimicking neural networks facilitate the natural formation of connections based on emergent, contextual need for interaction.

Work is exploration both what comes to defining problems and finding solutions.

We often think of individuals as independent and self-contained. The view suggested here sees individuals as nodes of the complex networks they form when interacting with others, co-creating themselves and the reality in which they participate.

The network is a complex system consisting of a large number of agents/nodes behaving according to their own principles of local, self-organizing interaction. No one agent or group of agents, determines how the system as a whole behaves. It is about self-organization.

Self-organization here means the agents interacting “locally”, following their own rules and intentions, without any steering from outside that particular event. All influence takes place in the layered, local events.

One of the biggest promises of Internet-based work is the way it potentially redefines local interaction. This is how the brain works. And this is how neural networks and machine learning work and this is how we may well work in the future!


Credits: Michael Nielsen, Doug Griffin and Yochai Benkler

Fred Wilson on decentralized, self-organizing systems

Serverless architectures such as AWS Lambda