The book Design, Launch, and Scale IoT Services classifies the components of IOT services into technical modules. One of the most important of these is Artificial Intelligence. This article is intended to supplement the book providing an insight to AI and its application for IOT.
After many years in the wilderness, AI is back on the hype curve and will change the world again. Or, will it? AI is always cool, but what has changed to justify the current hype?
There are several contributing factors. The volumes of data that will be produced by many IOT services suggest that this data cannot be managed by humans with traditional analytics tools. Therefore, AI can offer opportunities for IOT services to extract maximum value from the data. IOT cloud platforms are now offering AI services via APIs and application development tools, making AI more accessible for many IOT services. Now, AI can be incorporated without requiring extensive development and excessive costs.
There is a cost associated with every feature in a IOT service, and therefore business owners should ask themselves, “Why introduce AI?” Understanding the end-goal is the starting point. It is not suitable for all services and requires evaluation to understand when and how it should be introduced.
The following questions can provide a useful starting point for evaluating the introduction of AI:
The majority of IOT services include (or claim to include) some aspects of AI in their solution. This is due to a wide diversity in AI definitions (supervised/unsupervised, reinforced/deep learning) and the hype surrounding AI. (Note: All IOT services should take advantage of this hype while it lasts.)
Let’s look at the most common AI features and IOT industries to consider how IOT service owners can best evaluate AI and answer the questions above.
IOT cloud platform providers are offering powerful AI visual recognition APIs. For example, developing a human visual recognition tool has now become a trivial exercise for developers, and the cost for using visual recognition in IOT services has reduced drastically. These tools are best used for use cases recognizing humans and objects, but may not be useful for very precise recognition use cases. Developing specific visual recognition capabilities proves too expensive for most services, but it does make the service more attractive for end users.
Robotics is a branch of AI that, for many, implies a 2-armed, 2-legged machine that communicates with humans using visual or voice recognition. However, the most important use cases for IOT robotics involve the collection of data from sensors or extracted from robot programs. This data can be used by IOT services as input for AI machine learning algorithms to increase the robot efficiency, implementing features such as predictive fault management or adaptive positioning. AI can be used to increase productivity with robotic systems as part of Industrial IOT services that will become vital for many Industry 4.0 use cases.
These features have become widely available in mobile phones and CRM (customer relationship management) systems. They can be implemented via IOT cloud service APIs, i.e. it will be an option for many IOT services without requiring significant investment. It will make most services more attractive, implying more sales (devices.) However, we are probably quite far off from the stage where it is fundamental for IOT services. It is available on many mobile apps, but most users still prefer to use the touch screen. The main use cases for voice control systems will most likely involve voice to text transcription for operational or CRM activities to reduce cost, but may increase frustration for end users. (Note: Cloud providers are also introducing AI audio recognition APIs for fault detection that can be used to replace or augment visual recognition features.)
Smart factories offer numerous opportunities for implementing use cases that can increase efficiency via visual inspection, checking for faulty components or assembly processes errors. The analysis required should include the cost vs. the benefits. If visual inspection slows the production process, it may be counterproductive to introduce it in a manufacturing process that has a low fault rate. For example, let’s say that a smart factory is creating 5,000 components per day averaging 50 faulty components per day. The introduction of visual inspection may reduce it to 0 faults, but if it slows the manufacturing process to produce only 4,000 components per day. Is it worthwhile? The process owner will have to calculate if the reduction in throughput outweighs the benefits of a reduction in faulty components. This is an example of real-time fault detection that can useful for industrial IOT services. (Note: Many of the IOT Cloud platform providers offer the possibility to implement AI on edge devices, thus increasing the number use cases for real-time AI.)
Many industrial IOT solutions suggest that visual recognition will be used to determine the current health and emotional status of machine operators. This would require quite advanced features to be beneficial and therefore it is unlikely it will be relevant for most IOT services.
Visual inspection is showing great promise in detecting cancer and other ailments using advanced AI techniques, and it is improving the accuracy of diagnosis in many IOT health use cases. Very often it requires large volumes of sample cases and training sets to ensure the performance is acceptable. We have Genome technology generating billions of data items mapping our DNA that cannot be handled by humans and analytics tools. The introduction of AI offers the possibility to predict future health issues. Using data volumes of this magnitude requires unsupervised learning techniques, such as clustering. This may prove too complex and expensive for the majority of IOT use cases. Again, the cloud service providers provide options facilitating the management of training models and data is with tools such as Google Cloud AutoML. However, it is likely this will only be cost effective for limited IOT services.
It surprising that we haven’t seen a widespread deployment of AI in the management of intelligent hospitals. As with any complex logistical processes, AI can create significant efficiencies with relatively low investment.
Many smart home IOT services will implement voice recognition that connect with smart speakers. These are widely available from providers such as Amazon, Google, and Apple and can communicate with most smart home devices without significant complexity. It is likely that it will be an add-on for the majority of IOT services; nice to have, but not fundamental. Therefore, in most cases, the IOT business owner may have to budget for this as a premium service.
The potential of AI in transportation is very exciting (i.e. driverless cars.) There will be a lot of innovation with AI for drivers, but new IOT service owners will have to carve out a niche in this market. Although the technology is available, we may still be quite far off from many use cases being acceptable for drivers. We can imagine all the cars on the road communicating with each other and learning from each other as they are driving. Example: Car A detects ice on the road, informs other cars, and they all proceed to automatically adjust speed and brakes based on performance data from the other cars. This may seem futuristic, but the technology is currently available and AI offers the possibility of increased performance and decision making.
Analytics is closely interlinked with AI. If we are using AI, many ask, “Do we need analytics tools? Will analytics be dead with the implementation of AI?” Not quite. Most IOT services employ analytics, and therefore the data required by AI will already be available. AI should be able to replace a lot of the activities performed by humans using analytics tools. Or, the output of analytics can be the starting point of AI’s introduction in many IOT services. That does not imply analytics are a prerequisite; if the data is available, expert systems can be developed without analytics. Now, we are starting to see augmented analytics, where AI is assisting analytics with intelligent searching and other tasks. This may not be necessary for most IOT services, but we can be sure that it is being used by the Facebooks and Googles of this world. Most IOT services will not generate enough data for it to be cost effective to introduce.
Analytics, statistics, and lies are often interchangeable, and this will not be solved by AI. One challenge for many IOT services is that neural networks and deep learning AI techniques cannot explain why they are making decisions. This can reduce customer confidence and will be unsuitable for IOT services where a clear understanding of a decision-making process is important.
Barry Haughian was the Head of IoT Accelerator Program initiated to develop new IOT Services for Ericsson. Barry created a global organisation with 150 senior resources from Ericsson and external IOT partners to develop the IOT Services from Sales, Business, Technical and Operational perspective. He has developed Services in multiple IOT Industry verticals including, Industry 4.0, Smart Transport, Smart Energy, Smart City, Connected Home. These Services include GTM strategies, Operational models, Commercial models and have been developed in collaboration with Fortune 500 companies, IOT Startups, IOT Service partners and IOT Service competitors. With over 20 years of IT experience, Barry is now the Head of IoT at MTEK Consulting Services and virtuGrp developing Global IOT Services.
This article was contributed by Barry Haughian, author of Design, Launch, and Scale IoT Services.