Artificial intelligence (AI) surrounds us all. AI, or more correctly termed machine learning, can be found in many different applications in your home, car, and workplace. Our relationship with machine learning applications started when we began to use voice-operated smartphone assistants like Apple’s Siri or Google’s Voice but voice recognition is just one example of a machine learning application; others include vision processing and object recognition.

Okay, Google

With the wide range of industrial and electronics sector clients Publitek has, machine learning is a frequently covered topic. Over the past ten or so years, we’ve seen the use of machine learning (ML) neural networks, once restricted to compute-intensive data centers, emerge to bring intelligent control to the edge of industrial applications. It isn’t always apparent, but many of the voice recognition applications we interact with rely on constant cloud connectivity to process our speech. You have probably encountered that when asking your smartphone assistant a question when you’re in an area of poor cellular reception. Voice-operated intelligent home devices need to constantly listen to your room’s ambient sounds for a command trigger word or phrase, for example, ‘Okay, Google.’ Once detected, the device records or streams an audio file to the cloud data center to process and identify what action you desire. The command word or phrase is the only part of the voice processing that is inferred locally.

Machine learning industrial applications

Regular technical content topics for our semiconductor clients cover inference – the term used to describe how ML determines the likely word or object – at the edge and TinyML. TinyML perfectly describes how voice, event, or object recognition is achieved on low power, low compute capability microcontrollers without internet connectivity.

At Publitek, we often create content about many aspects of machine learning developments. For example, a semiconductor client may offer a low-power microcontroller that, coupled with a vibration sensor such as an accelerometer, creates a vibration detection sensor for use with an electric motor supplied by one of our industrial clients. A sensor like this can detect a change, either sudden or gradual, in the motor’s vibration signature. It could indicate a warning to maintenance staff that an investigation and corrective action are required. This is a perfect example of the application of TinyML. No high bandwidth, always-on cloud connectivity is required, and the sensor can be battery-powered.

Can AI be used to write content?

The increasing abilities of AI and machine learning started to come to the attention of media companies several years ago. Keen to benefit from the economies that AI offers, media companies, particularly newspapers and news websites, have made their own headlines by announcing that they are replacing journalists with AI.

So, can AI write? The answer is definitely yes, but the real question is whether it can write well and whether the written copy makes sense and satisfies the reader? The answer to the supplementary question is not as convincing.

The OpenAI initiative has been working on several text creation projects since 2015 and has recently announced the latest version of its natural language generation (NLG) model GPT2. GPT2 and the associated NLG techniques are already widely used to generate automated text for dashboards, social media posts, and client portfolio updates. Companies such as the Associated Press and the Press Association use AI to write tens of thousands of reports each year.

Writing a short sports report is however totally different from writing a technical article or this blog. Several websites are now offering the opportunity to trial AI writing, for example, AI-Writer.

The reality of AI-based content generation

As it stands today, the results achieved from these AI-driven sites lack a professional writer’s structure and finesse. To experiment I tried three separate topics:

  1. Can AI write an article?
  2. How to conserve power in an embedded microcontroller application
  3. How do I cook an omelet?

As you will see from the examples, the generated text is composed of largely unconnected paragraphs that have been researched from the web. The embedded microcontroller example is probably unfair, but it highlights most of the key methods of power conservation. Unfortunately, they are not structured as a technical writer, writing for an engineering audience would. The sources used include examples not referenced anywhere else in the text – BikeComputer, for example. That said, it researched and wrote this in a matter of minutes and provided the reference sources. AI-Writer highlights that the text isn’t as good as a human writer can produce but also highlights the time-saving nature of idea generation.

Given the large numbers of organizations now investing in NLG applications, the likelihood is AI-written content will feature increasingly in our lives. After all, there are many content generation tasks that are routine, straightforward, and have a limited number of variables.

In the meantime, if you’d like to speak to a human being about content creation – technical or otherwise – then please contact Publitek’s Strategic Content Team.