So much is being said about artificial intelligence currently. While there are sensible voices, much of what is being said is, in my opinion, hyped up for the sake of an appealing news, confused with actual sentience or just plain fear mongering. Which is why we prefer the term ‘machine learning’.

Machine learning (ML) is a type of artificial intelligence that involves developing algorithms and statistical models that enable computers to learn from data without being explicitly programmed. The goal of machine learning is to enable computers to automatically identify patterns and make predictions or decisions based on those patterns. It involves using a large dataset to train a computer algorithm or model to recognize patterns, make predictions, and learn from mistakes.

ChatGPT wrote that last paragraph. Good isn’t it? It’s a bit dry, but we could have asked it to “Define Machine Learning in the style of Robin Williams” and it would have a fair attempt at it based on what it has been trained about both ‘Machine Learning’ and Robin Williams. If you haven’t already try it for yourself. ChatGPT ‘learned’ to write like a human via what is called a large language model. This technology is also responsible for the autocomplete when you’re writing a message on your phone. 

Machine learning is all about pattern recognition

Machine learning is all about pattern recognition. Whether recognising something visually or recognising what word should follow the previous word when describing something or what visual elements have been used previously to convey something in a picture.

Here is a visualisation of a neural network doing character recognition. It’s quite mesmerising. This is a multilayer convolutional neural network, it employs relatively simple mathematics repeated an incredible number of iterations to detect ‘features’ by creating a statistical model that is supervised by human input. ChatGPT probably could have defined that better, but then I would feel like I was cheating you. The short of it is: when it comes to AI, the science is based on emulating how the human brain works - in terms of neural networks - but really this is just a case of clever data processing. Creating a sentient machine that has the cognitive ability to decide it loves you or it wants to murder humanity is something completely different. So let’s move on.

It is enough to know what each form of ML is for and for the purposes of this article we don’t need to understand what goes on under the hood. What we really want to understand is how Cadasio could make use of the various types of machine learning?

In my last article I looked at how Cadasio could leverage augmented reality to improve the user experience in situ. But to truly make the most of AR we would have to integrate some form of object recognition. We’ve previously imagined a user who is wearing AR glasses being presented with, in their field of vision, a Cadasio powered manual and having the components of whatever they are making or fixing being overlaid with information in their AR view. 

There are methods of object recognition that do not require a machine learning approach. Geometry can be used to match a real object to a virtual version, but this approach isn’t good enough when it comes to many small and complex things, which is a big issue for us. 

I think a good approach would be to implement a convolutional neural network (CNN). The CAD data uploaded by our users could be used to train a CNN so that it could then be used to identify real life objects for the end user. The downsides to this approach is that it takes a significant amount of processing power to train a CNN and for CAD data with many components it could take some time to process. But once it’s done it's an incredibly accurate method of object recognition. CNNs used to detect cancer have achieved over 98% success rates

What sorts of use cases could Cadasio support with object recognition? 

Imagine you’re about to begin building the Lego Delorean featured in our showcases. You’ve got all the little pieces laid out in front of you and lots of them are really similar and you want to be sure you’re using the right ones. Your AR glasses pass the image of what you’re looking at to a neural network trained with the CAD data for the Lego asset and it overlays what you're seeing with useful information. It indicates all the pieces you need at each stage in the process, and in which order they are needed. I think this could be useful, especially for instruction sets that contain many small, intricate and similar parts.

Cadasio can be used to create a wide range of instructional material and one other potential application for AR/object recognition is in the servicing of equipment on location. An engineer servicing machinery can be guided through the process step by step with the parts they need to interact with being highlighted in their AR glasses. The Cadasio instructional content of what exactly they need to do would be overlaid in a transparent animation. This opens up many possibilities for speeding up maintenance, but also opens up possibilities for non technical people to service their own home appliances, cars etc.

Returning to ChatGPT, it is already the fastest growing consumer application to date, and even some of our most experienced tech contacts - who are usually quite cynical about hyped up tech - describe it as ‘a game changer’. 

Could an AI write all our documentation in the future?

One thing that ChatGPT is able to do really well is take a natural language input and translate that into technical output. This is coming to the attention of the creators of development software such as Unity and Unreal. These are environments, commonly used for the creation of games, but they can be used by architectural and engineering focused businesses to develop virtual spaces. Historically these types of software have required the knowledge of a programming language like C# or C++. But by allowing the user to express themselves as if they were talking to someone, entities like ChatGPT can translate this input into code - programming languages are still languages and can be integrated into their language models.

The potential for this is kind of mind blowing. The user can describe what they want to happen in the software and ChatGPT will implement it. For example “Place a sphere in the centre of the scene and have it slowly, randomly cycle through the colours of the rainbow” would be trivial for ChatGPT but practically impossible for anyone without any familiarity to programming.

Github, the worlds most popular development and version control service for software recently released Copilot, which offers autocomplete style suggestions, from either code that you start to write or by writing natural language describing what you want the code to do. With such a big player in the coding world being invested in the technology it is only going to get better and more prevalent.

Whilst I think that the content creation UI in Cadasio is intuitive, we might have to consider at some point if it’s worth leveraging the power of natural language processing to deliver the benefits of being able to talk Cadasio through the creation of your instructional content. We think this is why ChatGPT is going to become a game changer - not because it can write your documentation or paragraphs in articles, but how it will interface with the rest of the world.

One other take away from what we have just described is that it can write code. You can try it yourself right now. You can even show it a website design (via a url to a picture of the design) and ask it to create the HTML and it will look at it and churn out all you need to get that webpage up and running.

We at Cadasio think we’re pretty good software engineers, you can judge us by our work, but we could in the future use ChatGPT to ensure our code is bug free and optimised and we can ask it for suggestions about how to go about things better. Also, as technology develops - software libraries getting updated for example - we need to refractor our code to make the most of new features. ChatGPT could be very useful in these endeavours.

One thing for certain is that machine learning is here to stay and it is gaining traction in more and more consumer facing applications. We’re moving away from those annoying chatbots and recommendation algorithms, but we’re moving into unknown and unexplored territory - and this is profoundly exciting and sometimes terrifying. There are new opportunities to allow creative but non technical people to make amazing things, but there are also implications for potentially massive amounts of job losses. We think this is the thing to be scared of, not the uprising of the machines, but the endless appetite big corporations have for cost savings and profit. But as ChatGPT can now score in the top 10% of bar exam passes, maybe it will be such a threat to lawyers that they’ll protect us all with new laws restricting the use of AI. Failing that, we at Cadasio would just like to make it clear that we welcome our new AI overlords - and we always remember our Ps & Qs on ChatGPT.