Judging by the number of confusing posts, blogs and articles on this topic, anyone exploring the potential of what the IOT can deliver to their business/organisation can be forgiven for thinking that the IOT will need a new set of AI technologies to work correctly. Throw into the mix the hype that the IOT will need big data analytics & platforms to work and we have a very confusing IOT landscape to navigate and understand. Having been involved in delivering intelligent software solutions for over 30 years using ‘traditional’ AI, I feel well positioned to assess the need for new AI for the IOT.
Traditional AI has evolved continuously over the last 35 years and is now routinely embedded within many classes of software applications. There are two main manifestations of traditional AI:
- Rules automation/expert system technologies. These systems are designed to capture human expertise (decision making, risk/situation assessment, diagnostics/trouble-shooting, advising on products/services, asset performance monitoring etc.) and to automate policy rules and regulations.
- Machine learning. These systems can learn new patterns/rules from historic data. The learning can be either algorithmic (black box models such as neural networks) or symbolic rules/trees that are understandable to humans.
People often quote Apple Siri, IBM Watson or the Amazon recommendation engine as examples of the brave new AI world!! The reality is that these new AI technologies complement rather than replace traditional AI. The main limitation of traditional AI is that it operates on structured data (numeric values such as price, age, voltage etc. or a pre-defined set of discrete symbols (labels) such as colours, occupations, etc.). The new AI technologies are focussed on interpreting and learning from unstructured data such as free-format text, speech, images and videos. The concepts, patterns, features and attributes generated by the new AI from free-format text, speech, images and videos can be used as structured data to drive traditional AI.
Having clarified the distinction between traditional and new AI, I will further argue that the IOT does not need new AI to work correctly, but what it needs most is distributed traditional AI (intelligence) as outlined below:
- The growth of the IOT is being driven mainly by the availability of low value, low power, small sensors attached to objects & things ranging from street lamps to farm animals to home appliances to industrial plants to elderly patients. By definition these sensors generate structured numeric data which is easily processed by traditional AI (with the exception of data from microphones and digital CCTV cameras which need new AI to pre-process into patterns and features)
- Most IOT ecosystems involve a model of centralised intelligence whereby data from sensors (things) are uploaded to a central private/public cloud where is it processed by a cloud based AI engine before actions/alerts are notified back to devices/people at the edge of the IOT. Such a centralised model will not work for two reasons; firstly it is critically dependent on the internet connectivity and will lose all intelligence if the network is down, and secondly with over 30 billion things forecast to be connected to the IOT over the next 5 years, the amount of data being uploaded to the cloud will overwhelm the bandwidth of most internet networks. The solution is distributed intelligence with traditional AI/rules engine running everywhere on the IOT echo system (IOT edge hubs/devices, cloud and mobile devices)
In summary, the IOT needs not new AI but distributed traditional AI with engines that are scalable in terms of performance and footprint so that they can run anywhere from a Raspberry Pi at the IOT edge to a massive Azure Service Fabric cloud server to a smart phone. Distributed intelligence is the key to a resilient IOT with real time intelligence.Share this post via: