Bringing Artificial Intelligence to the Internet of Things – with the help of the Cloud

We have already touched on the role that machine learning and artificial intelligence plays in building smart systems that provide value to customers and companies. From predicting when elevators will require maintenance and service to the advanced voice control functions used in iPhones, artificial intelligence is becoming increasingly ubiquitous.

This blog however has not yet done justice to the extent to which the entry bar has been lowered for advanced analytics, artificial intelligence and machine learning to be incorporated into a smart system, be it a fully online one or a system that incorporates physical things. There is an enormous range of online cloud-based platforms for carrying out data analytics, most of them accessible programmatically (and hence automatically) via APIs over the Internet, adopting pay-as-you-use business models. The classification and prediction algorithms that were once the preserve of the Internet Giants (think of NetFlix’s recommendation engine) and a handful of computer science departments are now available to the newest of startups.

Here we explore the main types of artificial intelligence and machine learning platforms and touch on the opportunities they provide

Natural Language Processing

Social Monitoring

DatumBox and Alchemy both offer sophisticated natural language processing capabilities. This involves analysing textual information and automatically assessing the sentiment (i.e. is it generally positive, negative or neutral), determine if it is subjective, pick up the key themes and topics, the languages used and the type of text – in other words, is it a technical document, a conversational piece or has it got primarily a commercial function. For example, the Alchemy API assessed this site’s review of CES as being overall positive in tone, and was specifically positive about Samsung, Apple and Intel, though mixed about Google and Qualcomm. For some unknown reason, it however classified the article as being about pasta and rice! The demo is quite interesting and worth at try – it can be found here.

Spoken Voice and Speech Recognition

Related to natural language processing are APIs that understand the spoken word. For example, MindMeld provides an API that creates a knowledge graph of a given website – (think of it as a map of the site, showing how different pages are related to each other), which can then be used as a basis for navigating using spoken commands, very much like Siri. By way of example, the MindMeld site showcases a number of examples to how the technology could be used – including selecting a recipe from a cooking site and searching for items on an online retailer. While these examples are all web-based, there is no reason why the technology cannot be accessed from a physical device, such as a TV, HiFi unit or a dedicated assistance device like Amazon’s Echo.

Spoken Voice and Speech Recognition

Related to natural language processing are APIs that understand the spoken word. For example, MindMeld provides an API that creates a knowledge graph of a given website – (think of it as a map of the site, showing how different pages are related to each other), which can then be used as a basis for navigating using spoken commands, very much like Apple’s Siri. In the meantime, Siri’s founders, apparently frustrated at the limitations being placed at Apple, have gone on to start another company, called Viv, which will apparently take artificial intelligence to the next step, creating what they call a “global brain”. Their vision is to create the world’s most effective intelligence engine, embedded in all sorts of apps, devices and websites.

Image and Pattern Recognition

Image recognition systems are crucial for allowing robots, drones, traffic monitors or other objects with cameras to interact with the physical world. However this poses real-time requirements which often rules out cloud-based services due to the time taken for messages to go to and from network. Instead these solutions are typically provided are in the form of software development toolkits for embedding in the device itself. One high-profile example is Intel’s RealSense 3D camera which automatically recognises objects, gestures and the general environment in three dimensions. This was showcased by integrating into a drone which automatically avoided obstacles.

Machine Learning and Predictions

Probably the most useful category of APIs in the context of the Internet of Things, there are a number of providers who offer generic though sophisticated machine learning algorithms that can be applied to a vast range of analytics or predictive applications. The key power of machine learning is to exploit very large data sets to discover patterns and meaning hidden in data, which would be otherwise remain hidden.

The most comprehensive of these is the Microsoft Azure Machine Learning platform, which combines a ‘drag and drop’ interface with the ability to create custom machine learning APIs and integrate with existing data processing code, using industry-standard R and Python languages. This enterprise-grade platform is already being used extensively in industrial applications. For example, an energy company in Norway is using it to manage a smart grid, predicting capacity problems and automatically managing the load in different buildings automatically. A research content provider uses the platform to customise the content offering to new customers, in the same way Amazon recommends new books. Crucially (at least according to Microsoft), this off-the-shelf platform significantly out-performed the custom in-house solution used previously. The Microsoft platform was created partly to exploit Microsoft’s AI knowledge and create a distinct differentiator over Amazon’s AWS cloud service, but also as a response to IBM’s own AI platform – called Watson after its founder. IBM Watson offers the whole gamut of artificial intelligence tools, including text analysis, speech recognition, visual recognition and predictive analytics, though is apparently rather more complex to use.

Going Forward

In this article, I have only scratched the surface of the range of artificial intelligence platforms, ranging from the big enterprise players to venture capital-funded startups. However, this should give a flavour of the sophisticated tools available to the creator of Internet of Things solutions, be it a healthcare or a refuse collection service, and the extent to which these platforms will be used to underpin these smart services.

Leave a comment