A couple of days ago I had a quick chat with a coworker related to the concepts of Artificial Intelligence and Machine Learning. Looking deeper I noticed there was a bit of confusion and in today’s society we tend to combine the two topics. I know I was a bit confused when I first started learning about these topics.

To take a step back, the first time I got to see a interactive robotics application was during second year of my undergraduate degree. I got the chance to see a really cool Ted Talk (https://www.youtube.com/watch?v=w2itwFJCgFQ if you are interested in watching it) which focused on the dynamic response of drones and the various ways they could collaborate to complete a task. Spoiler alert, one of them was to send a ping pong back to the presenter. This was my first introduction into quadcopters ( which are now a passion of mine) and the type of human/machine interaction that was possible. At the time I had no idea of the complexities required to get a drone to hover, let alone figure out how it could dynamically adjust to changes in its environment. In a way, this was one of the first examples I had seen of advanced Artificial Intelligence. Now the term “advanced” is relative, what was considered advanced in 2013 may not be as advanced in 2021 or when you read this article. But for someone in second year of their undergraduate mechanical engineering degree this was very impactful, in a positive way.

What does this have to do with Artificial Intelligence you may ask. Well, the program that was designed for those drones, in my opinion, was an example of Artificial Intelligence. The drone used sensor information (most likely a 9 axis IMU and the position information from a motion capture system) to predict what it’s motion will be given a set of motor commands. To look at other places where Artificial Intelligence is used we also don’t need to get that fancy. A more simplistic application could be a door sensor reacting to someone entering a store. When I presented the door example to my coworker there was a sense of ” That can’t be Artificial Intelligence, it’s too simple” which got me thinking. In movies when we consider an Artificial Intelligence we think of sophisticated systems with almost human like levels of awareness and I feel this has distorted a bit what it means for a system to have a AI. I believe a system could have a simple AI like the door example, or something far more complex like what’s implemented in the drone. In both cases we now look at the architecture used to design these systems. I believe the architecture is what defines the level of complexity for each system. In the case of the drone a proposed architecture could include a combination of State Machines for scheduling tasks and handling safety critical actions, to using more control centric approaches like a Model Predictive Controller which analyses the dynamics of the drone and generates a prediction into the future of where the drone will be, to sensor filtering and attitude estimation in order for the drone to make sense of its’ attitude in space.

What about Machine Learning? Where does this come into play? When the concept of Machine Learning and data analytics was presented to me it was shown as a application of statistical modelling. It was presented as a set of tools used to teach machines how to extract information from data, such as the relation of sales trends and customer needs or like in my previous blog, trends in COVID research which could be applied to future pandemics. In other applications different ML architectures could be used in visual data to extract patterns such as identifying objects in an image, or convert videos which have a low frame rate into a high frame rate/slow motion version (https://www.youtube.com/watch?v=MjViy6kyiqs is a cool example by Nvidia). Notice how the title still calls for AI even though the process of generating the images is via Deep Neural Networks which falls under Machine Learning?

In the end, the thought that came to mind was not to compare Artificial Intelligence with Machine Learning. Instead it was to shine a bit of light in the constant equivocation between the two topics and add my own take on the topic. For me Machine Learning is a tool, just like State Machines and Model Predictive Controllers, which can be used to give a robotic system an increased level of Artificial Intelligence. As technology advances we are presented with ways to develop even more sophisticated tools. Who knows, maybe we will get closer to a robotic system which can one day reach human like levels of intelligence. I also believe the more tools we find the closer to we can get to understanding the most complex computer ever designed, the human brain.

Hope this quick blog sparked some interest and I’d like to know what you think!