English | EN 
LabTalks is an event full of inspirational stories from our colleagues. It’s a unique concept where personal stories are connected to technology. Will Thieme was one of the speakers and he invested his spare time to translate his story to a blog, enjoy!
In the contemporary digital era, Artificial Intelligence (AI) and Machine Learning (ML) have emerged as the two pillars of data science that are elusive to the common masses. Everyone is talking about AI and ML, but only a few understand what they actually mean. A bid to explore the roots of data science only leads to a complex jumble of theorems, mathematical calculations and obscure programming language that are apparently incomprehensible to the ‘not so intelligent people.’ Nevertheless, no matter how seemingly complex and expensive it appears to be, AI and ML are actually accessible to everyone, even those on a tight budget.
To trace the origin of AI and ML, let’s travel back in time to the good old days of the simple mathematical calculator. We needed an automated machine to do large, complex and repetitive calculations for us which we, the human, were either too lazy or too dumb to do. Then came a time when we needed robots to do more complex and multifaceted jobs with minimal human intervention. That’s how AI and ML came into existence. It’s nothing but a complex game of analytics, which in layman’s terms may be defined as a series of pattern recognition, algorithms, and result predictions.
But how does ML actually differ from traditional programing? In simple terms, traditional programming is a manual process, wherein a programmer creates the program and therefore designs the logic. ML, on the other hand, is a computer formulating its own logic. The programmer merely has to tell the computer what to do in a specific situation and the computer will learn how to best do exactly that. In other words, ML gives computers the ability to learn without being overtly programmed.
We all know the Tesla AutoPilot. This is a striking example of ML. The engineers at Tesla have trained a car to drive by itself and even predict crashes before they even happen.
Another example of ML is Amazon Alexa. Alexa seems to understand what you are saying to it. She even talks back to you. Alexa contains speech recognizing, natural language processing and speech-generating algorithms that are all generated using ML.
Another example of ML that might not be so obvious is the Apple iOS keyboard. The keyboard tries to predict what letters you are more likely to press next and will make the click boxes of those letters bigger so that the user will make less typing errors.
But are all these ML applications for the big players only? The answer is no. Let me show you. I set out to create an application that might benefit us one day. Imagine you are on holiday to a foreign location and you come across new currency coins that you are not familiar with, neither do you know their exact value. Wouldn’t it be great if you could simply scan the coin with your smartphone and it could classify and tell you which currency it was and what the exact value of the coin was in terms of Dollars or Euro? Basically, I set out to create an App that could recognize various foreign currencies for me. There are three challenges that I needed to overcome to make this application. First, I needed lots of data, secondly, I needed software that I could use and understand. Let me tell you how I tackled each of these problems and lastly, I needed hardware.
To get the data I needed I just looked at the coins in my wallet. I had some foreign coins that I could use. I used my phone to take about 1000 pictures and labeled them according to what coin was in the picture. Then I needed software to train my model. I based my model on an existing architecture called Xception and trained it using Keras. To get this up and running I needed no prior knowledge about the difficult mathematics theorems that go into ML. When I ran the training of this model on my own laptop it was so slow that the training would actually take days. Dedicated hardware is expensive but luckily our friends and Google made a new IDE called Google Colab which allows you to train your models on Google’s hardware absolutely free of cost. In the end, it took about half an hour to train this Model and about a day to get everything else ready.
This exactly portrays what the big players like Google, Apple, and Amazon are currently doing. They’re trying to make ML more accessible to everyone. The idea is, you do not have to be super smart or super-rich, neither do you have to have any formal education, training or experience in programming, math, and engineering to start as an AI or ML developer. AI is no longer limited to just the big players. It is accessible to all developers. A survey by Kaggle ML & DS revealed that only 30% of those working in AI and ML development have an academic background in machine learning and data science as part of their formal education. On the contrary, 66% of AI and ML developers are self-taught and 50% of overall AI and ML developers confessed that they used online resourced to train themselves and learn about this discipline.
The moral of the story is, it’s time that you stop thinking about ML and actually start building it.
Will Thieme is a practice lead at Sogeti. His main responsibility is ensuring that the mobile development and implementation projects at Sogeti are running smoothly from a technical point of view and are up to the quality standards our clients expect. As an expert in the mobile development domain, he is experienced in leading teams both onshore and offshore. His main role is translating our clients’ wishes into new features and improvements, and making sure these features and improvements get implemented correctly.
