Distributed Intelligence

IN ORDER TO provide a focus for detailed analysis of IoT subject matter; we hereby provide a comprehensive definition for: Distributed Intelligence. This new concept of Distributed Intelligence is closely associated with the related concept of Situated Intelligence; and both should be considered twin sides of the same coin; namely constructive application of AI to the IoT problem space.

Distributed Intelligence is when we have embedded Intelligence and Connectivity across the environment in useful ways. In other words it is recognised by a spatial distribution of computing resources/elements. Whereby the problem-solving capability is supplied by several networked IoT objects/systems; and so benefits from expansive knowledge of the same problem-space, whilst typically engaging multiple computing elements (sensing/processing/active capacities) towards an extended geographical region.

Distributed Intelligence (plus Situated Intelligence) is what the IOT is all about—namely providing enhanced problem solving capabilities: including solution AUTONOMY (task replacement/managment) and solution VISIBILITY (task awareness) with respect to Smart Things—or instrumented Devices, Systems and People—located at disparate geographic LOCATIONS (within and across IoT environments).

Here in this section, we explain those particular advantages that may be supplied by a Distributed Intelligence: specifically in relation to the provision of an enhanced problem solving capability.

Problem Context

Distributed Intelligence is defined as the ability of an IoT Thing/System to provide the following functional qualities (problem solving facets):

  • MONITOR (ENVIRONMENTAL SENSING):  degree to which an IoT object/system usefully SENSES data within and/or across the IoT environment;
  • INTEGRATE (PEOPLE, THINGS , SERVICES): degree to which an IoT object/system COOPERATES usefully within and/or across the IOT environment;
  • ACT (MODEL, REPORT, CONTROL): degree to which an IoT object/system INTERACTS usefully within and/or across the IOT environment.

Note the important definition change for Distributed Intelligence, as opposed to the related concept of Situated Intelligence; whereby we have replaced the term: “within the Environment”; with the term: “within and/or across the Environment”.

As stated, Distributed Intelligence is when we have embedded Intelligence and Connectivity across the the environment in useful ways. The result is a real-world object/system that gains the ability to calculate, process and decide in relation to a specific class of problem-space; whereby aspects of the world suddenly ‘wakeup’ and begin looking-at and interacting with “extended and/or geographically remote” aspects of the environment in a highly useful manner.

Distributed Logic

Distributed Intelligence (Distributed Artificial Intelligence)—or Multi-Agent systems—refers to systems which employ “distributed logic”; whereby we separate the processing from a large problem space into multiple subsystems or extract certain processing functions from the main system and place them into separate machines and/or at geographically separated computing ‘nodes’.

Systems built to exploit and/or employ Distributed Intelligence consist of autonomous learning plus autonomous processing nodes (or agents). Said nodes are distributed widely across the environment, often at a very large scale (i.e. large numbers of nodes are deployed and/or we have these nodes separated by large distances). Whereby these nodes can act independently and partial solutions are provided and integrated by communication between nodes, often asynchronously.

By virtue of their scale, systems using Distributed Intelligence are robust and elastic, and by design, loosely coupled. Furthermore, DAI systems are built to be adaptive to changes in the problem definition (problem space) and/or built to be adaptive to changes (possibly structural) within underlying data sets (often due to the scale of deployment and/or difficulty in redeployment etc).

In summary, Distributed Intelligence provides:

  • Autonomous processing / learning
  • Enhanced coordination of problem / solution space(s)
  • Reduced central dependancies (no data bottlenecks)
  • Adaptive problem solving / flexible to problem re-definition
  • Independent and locally optimised actions
  • Robust / independent: data processing pipelines
  • Elastic / non-brittle: logic dependancies
  • Faster and localised responses to the environment
  • Live command and control sequences (updates)
  • Loosely coupled nodes / asynchronous computing

A key advantage of distributed systems is that they do not require all the relevant data to be aggregated to a single location (aka the Cloud / central server), in contrast to monolithic or centralised Artificial Intelligence systems which have tightly coupled and geographically close processing nodes.

In summary, distributed systems often operate (beneficially) on sub-samples or hashed impressions of very large datasets. Faster, more robust and elastic computing solutions are the case; whereby logical conundrums do not cause the whole system logic to halt. In addition, the source dataset may change and/or be updated during the course of the execution of the overall (distributed) data processing system.


A key objective of Distributed Intelligence is to solve the reasoning, planning, learning and perception problems of artificial intelligence. Whereby a distributed approach is especially useful when dealing with large data sets and/or big analogue data, and because the data-processing problem can be tackled efficiently by distributing the problem to autonomous processing nodes (agents).

Advantages of a Distributed System may include:

  • Robust and elastic computation: Data processing (system-wide) continues regardless of unreliable and failing resources due to loose coupling of nodes;
  • Enhanced coordination: Coordination of the actions and communication of the nodes—leading to optimal problem solutions that can solve the big-picture problem with utmost efficiency;
  • Multi-tasking: processing of many subsamples continue simultaneously on large data sets, leading to faster processing overall; plus on-line machine learning provides for sharing of intelligence amongst nodes.

In terms of any particular usage scenario; there can be many reasons for wanting to distribute intelligence; whereby the following may apply:

  • Parallel problem solving: deals with how classic artificial intelligence concepts can be modified, so that multi-processor systems and clusters of computers can be used to speed-up analysis / calculation and/or to render more effective (impactful) machine logic.
  • Distributed problem solving: refers to multi-agent systems that work together to solve a large-scale problem like collaborating insects, and/or to autonomous entities that can communicate to foster a coordinated approach to problem solution. This approach lends itself to problem abstraction whereby a central commanding node can rely on multiple sub-nodes to solve localised problems optimally in terms of the overall problem space.
  • Multi-agent simulation: a branch of computing that builds the foundation for simulations that need to analyze not only phenomena at macro-level but also at micro-level, as is the case in many social simulation scenarios.

Two classic approaches to distributed computing are as follows:

  • Multi-Agent Systems: coordinate their knowledge and activities and reason about the processes and results of coordination. Whereby agents are physical or virtual entities that can act, perceive and communicate with other agents. Each agent is semi-autonomous and has skills/knowledge to achieve certain goals (the overall focus here may be on solution of either local and/or global tasks—or both). Whereby each agent can adjust its programming and/or goals and/or implemented logic—as a result of information received from the other nodes present in the network (solution coordination).
  • Distributed Problem Solving: In a distributed problem solving computing system, the tasks are divided among nodes and the knowledge is shared (typically each node is concerned with only with a specific set of local tasks—whereby said local tasks are coordinated or set externally to the network of nodes by macro-level problem division). The main issues here relate to task decomposition and synthesis of the data-gathering, knowledge and solutions.

The upshot of both approaches is that we can apply a bottom-up approach to AI and problem solution, whereby a subsumption architecture can be applied to the problem space; as well as the traditional top-down approach of AI.

In addition, both methods can become a vehicle for emergence behaviours in which the problem space is tackled using an emergent type algorithm. In essence an emergent algorithm provides a set of simple building block behaviours that when combined exhibit more complex behaviours. One example of this is the implementation of fuzzy-logic motion controllers which are used to (for example) adapt robot movement in response to environmental obstacles and/or otherwise be applied to solve issues related to massive environmental complexity and/or constantly varying 3D problems and/or varied environments.

Coordination Age

The world is entering a new age of programmable and connected Things. Some experts have called this the Coordination Age, which is being driven by a growing need for resource efficiency—irrespective of service user/delivery location(s)—and enabled by next generation 5G networks, AI, SDNs and the IoT.

To release the benefits of the Coordination Age, all manner of intelligent “things” will need to be able to discover each other, communicate autonomously, collaborate, and then self-decide plus self-act to solve problems with new vigour. Accordingly, Industry 4.0 must dramatically improve the efficiency of resource utilisation, arising from a combination of developments in the demand and supply of services. 

Evidently – problem and solution space – coordination is the key job that needs to be done across many market areas. People, Things and Services need to be brought together at the right time and in the right place to deliver the desired outcome.

Forces driving coordination include:

  • SMART HOME: devices, sensors, appliances and applications created by many different companies need to be coordinated into an easy-to-manage solution for consumers (key requirements: enhanced solution INTEROPERABILITY and high SYSTEM RESPONSIVENESS enabled by the plug-and-play IoT plus situated intelligence capabilities);
  • SMART ENERGY: to manage the generation, storage and delivery of power (e.g solar/wind energy) across highly complex, disparate and sometimes international supply chains (key requirements: optimised solution OPERATIONAL FACETS provided by the efficient real-time optimisation of network operational and maintenance tasks: requires tailored and aggregated generation/storage/delivery facets and efficient use of communication and power transportation/delivery networks [M2M, M2H and H2M systems] – and all provided by distributed / situated intelligence);
  • SMART HEALTHCARE: clinicians, patients, treatments, resources and information need to be coordinated for successful healthcare outcomes (key requirements: highest service RELIABILITY/RESPONSIVENESS plus comprehensive SECURITY FRAMEWORK delivered with full solution INTEGRATION (distributed intelligence) and real-time sharing of resources [i.e. intelligence + command/control aspects are plug-and-play also]);
  • SMART TRANSPORT: manage transport flows for both public and private transportation, to ensure the best use of available resources and where to direct investment most effectively (key requirements: DISTRIBUTED LOGIC SOLUTION providing comprehensive transportation management capability provided by macro-level understanding of micro-level happenings (distributed intelligence): large scale sensor networks adapt themselves to local problem-spaces and report using real-time service data aggregation at the correct granularity);
  • SMART LOGISTICS: to manage the distribution and delivery of stock and produced goods across highly complex, international supply chains (key requirements: optimised solution OPERATIONAL FACETS provided by the efficient real-time tracking of consignments and tailored packaging/transportation and just-in-time networks [M2M, M2H and H2M navigation systems] – and all provided by distributed / situated intelligence);
  • SMART INDUSTRY: to ensure that manufacturing and supply-chain processes deliver, assemble and process goods and materials efficiently (key requirements: optimised solution AUTOMATION by enabling accurate, precise and cost-effective matching of problem and solution spaces – at the correct scale and place/time – whereby macro and micro problem-solution space views are closely matched, henceforth the solution employs a combination of situated and distributed intelligence).

The upshot is that we need new kinds of advanced plug-and-play IoT systems – and each one carefully designed and hence tailored specifically to its own specific purpose(s) and/or usage-scenario(s). Whereby we apply advanced Artificial Intelligence (situated and/or distributed intelligence) to automate, manage and accelerate solution delivery. Put simply, the Coordination Age is about improving what people and companies get for their time, money, effort and attention. 

But there is a problem (or open secret) that limits the development of the Coordination Age; whereby there is something inherently wrong with the very idea of the IoT.

What’s Wrong with the Internet of Things?

There’s a problem with the current concept of the ‘Internet of Things’; put simply: it isn’t an internet. The IoT isn’t even a continuous network, whereby it is severely limited in its capacity to operate efficiently, grow, evolve in intelligence and capability, and deliver the benefits that have been envisaged for it. Most current applications are in reality closed – and private – command and control solutions using standalone technology that is applied to limited application areas.

Oftentimes, a private network of IoT Things is the right solution – and especially to protect data privacy, security and to provide speed, low latency, and service reliability etc. However in order to make some of the most complex and dynamic applications work, specific sets of “things” (present within the problem/solution space), including not just sensors but also IT systems, will need to be able to find and communicate with each other autonomously (rapidly, efficiently and without low-level design work etc).

The upshot is that the world has an urgent requirement for  a true Internet of Things (IoT) platform; which would provide: a combined SITUATIONAL INTELLIGENCE and DISTRIBUTED INTELLIGENCE that is defined as the ability of an IoT Thing/System to provide the following functional qualities (problem solving facets) – whereby these are supplied in a plug-and-play fashion (often on a private and/or secured network):

  • MONITOR (ENVIRONMENTAL SENSING):  degree to which an IoT object/system usefully SENSES data within/across the environment;
  • INTEGRATE (PEOPLE, THINGS , SERVICES): degree to which an IoT object/system COOPERATES usefully within/across the environment;
  • ACT (MODEL, REPORT, CONTROL): degree to which an IoT object/system INTERACTS usefully within/across the environment.

Advantages of an “Internet for Things” (I4T)

Another way of stating the problem is to say that we need an Internet For Things (I4T); or in other words a broadly-applicable connectivity solution that is designed to allow Things to be connected and to interoperate with Plug-And-Play functionality. An Internet for Things would be a digital enabling fabric for wholly new levels of functionality, the same being of potentially great benefit to individuals, enterprises and our environment.

Potential advantages of I4T (sample):

    • COMBINATIONAL INTELLIGENCE: An Internet for Things would allow data to be combined and enriched in previously inconceivable ways; by means of enhanced collaboration between seemingly unrelated IoT Systems – mashing up intelligence from different and seemingly unconnected sources for informational, security and commercial purposes.
    • CONTEXTUALISED INTELLIGENCE: It would enable more meaningful machine to machine conversations. One device might offer enhanced functionality by deriving important contextual information from other communicable entities and/or devices in its environment.
    • SITUATED INTELLIGENCE: A truly intelligent objects integrates with its whole environment  – or its purpose – in a seamless manner (including all factors of relevance). For example, an in-building climate controller could offer more accurate control based on combined data from sources within its network, such as security devices and thermostats, plus external sources such as personal smartphones and smart watches, plus weather channels etc.
    • AUGMENTED INTELLIGENCE: Intelligence should be (potentially at least) gathered and combined from everywhere and anywhere. It would trigger a quantum leap in the volume and quality of intelligence available to IoT Systems, Things, individuals and organisations. All kinds of “things” – buildings, vehicles, infrastructure elements, people – become data points and data sources, some static, some mobile, all contributing to a vast, accessible pool of crowd-sourced information (data sets may be public or private to a specific system/user).
    • DISTRIBUTED INTELLIGENCE: It is the potential for the automatic collaboration via distributed intelligence capabilities – between Things when solving problems that makes the I4T a real game-changer. The potential of the Internet for Things – and related intelligence related coordination factors – are emerging just as the world is facing massive challenges in terms of the use of its resources.
    • COLLECTIVE INTELLIGENCE: is shared or group intelligence that emerges from the collaboration, collective efforts, and competition of many individuals and appears in consensus decision making. The related concept of Cognitive City is a term which expands the concept of the Smart City with the aspect of cognition or refers to a virtual environment where goal-driven communities gather to share knowledge. A physical cognitive city differs from conventional cities and smart cities in the fact that it is steadily learning through constant interaction with its citizens through advanced information and communication technologies (see work of E. Portmann). 


In this section we have examined salient features of IoT system design. Whereby we have demonstrated how a particular set of IoT problem space(s) can be optimally addressed – or closely matched (each one)  –  to its specific IoT solution space using the new concept(s) of Combinational, Contextualised, Situated, and Augmented Intelligence(s); plus Distributed and Collective Intelligence.

In sum, it is hoped that this new analytical framework can be a game-changer; in terms of the development of IoT systems that engender far more detailed, effective and tailored relations between IoT problem and solution space(s). The inevitable result will be IoT Systems that perform optimally and in predictable ways. All in all,  we have demonstrated how a set of generalised intelligence provision factors can be used to ensure that IoT Systems are designed and built to meet operational requirements in full.