By José Ignacio Orlando, PhD, Assistant Researcher @ CONICET & Nicolás Moreira, Head of Engineering @ Arionkoder
Artificial intelligence (AI) is found more and more often in headlines and social media, with models that can accurately diagnose diseases from medical images, generate art or even make code for us. But what is exactly AI, and how does it differ from other closely related concepts like machine learning, deep learning, big data, or data science? In this article, we help you understand how these concepts differ from each other by giving you a comprehensive survival guide to surfing the hype wave.
AI is ubiquitous. From bots and assistants that help us write computer programs to models that make art out of a few sentences, this technology is revolutionizing markets and industries with an amazing impact on our lives. However, as it becomes increasingly more present, we witness the growth of its hype, from a tiny little ripple in the news up to a tsunami in which almost everyone has something to say.
How can we, then, discriminate between real, impactful AI technologies and simple computer programs that have nothing to do with it? Below, we’ll try to cover the most important aspects as well as some common questions and misunderstandings around AI.
What’s Artificial Intelligence in the first place? AI is the capability of computers to mimic human cognitive functions. While it might sound like a textbook definition, this simply means that an AI system can reproduce at least some of the actions that we can do as human beings: making decisions based on certain stimuli, e.g. to solve a specific problem; perceiving the environment by interpreting signals captured by task-specific sensors; or even understanding human communication.
Let’s name a few AI systems that we use on a daily basis. Google or Apple’s Photos apps use AI to filter out images based on keywords: a predictive model takes your text input from the search box, understands its meaning, relates it to the content of thousands of images that you’ve uploaded to the Cloud and returns the ones associated with that input. If you think about it, the AI models in the background are actually understanding our language and interpreting our images to decide on their own if they should be included on a list or not. On a slightly more technical note, there are no IF clauses involved here: just a computer analyzing patterns in data. Something similar happens with Siri, Cortana, Alexa or Google Assistant: we say something to them, and an AI analyzes our voice using signal processing techniques, maps the waves to a concept representing what we just said, and based on that automatically decides whether to turn on a lamp, play a song on Spotify or set up an alarm.
Is any standard computer program an AI? Not at all! AI systems are not explicitly programmed to do something, but rather have the ability to make decisions on their own without being directly instructed to do so. If a code establishes exactly what has to be done, then there’s no AI there.
And what’s Machine Learning, then? Machine learning (ML) is a subfield of AI that applies mathematical models to data to help computers learn without direct instructions. Hence, ML is a tool that we currently use for creating AI systems. This field relies on statistical learning frameworks to generate predictive models that identify patterns in the available training data. Then, the resulting algorithms are applied in test time to automatically predict outcomes from data by processing new input samples.
So… are AI and ML the same thing? Nope. They are not. We can have AI systems without applying ML techniques. Videogames and navigation programs frequently rely on pathfinding methods, for instance, are algorithms that bring you from one point to another by choosing an optimal way and are not trained using ML techniques. A long time ago, non-ML rule-based systems were used to create chatbots that did not learn from data but rather searched for an answer to a specific question based on a series of predefined clauses. Having said that, current AI systems are mostly built based on ML models.
What about Deep Learning? Deep learning (DL) is a subfield of ML in which the models are artificial neural networks: they resemble the way in which our brain works, and are based on multiple stacked layers of artificial neurons that are outstanding at finding patterns in data. Training these networks usually demands a huge set of annotated samples and sometimes dedicated hardware such as GPUs (yes, the same we use for playing videogames!). Deep neural networks are currently state-of-the-art for solving ML problems.
And what is Data Science? This name acts as a huge umbrella under which many tools stand, with uses that range from extracting meaning from data to informing decision-making and planning. In general, these are classic statistical methods that are widely available in frameworks like R, including regression analysis, graphical representation techniques and dashboards. Notice that these tools are not smart or intelligent in any way: they just let us make much better decisions based on evidence, but they are not making any decisions on their own. So data science is not AI.
A note on Big Data. You probably have seen the term Big Data entwined with all the concepts we’ve covered here, something like “the foam” of this hype wave. Big Data refers specifically to a family of techniques aimed at leveraging the humongous amount of data that we as humans produce on a daily basis. Big Data is also a culture and a philosophy, intended to extract valuable information from the massive datasets available in companies and businesses. We can build AI systems using ML models on top of this information, or apply Data Science techniques to improve our decision-making processes. In any of those cases, we’re using Big Data.
AI and ML can help you reach the business results you want. Leverage our teams of experts and discover everything we can achieve together! Contact our Head of Engineering, Nicolás Moreira, at nicolas@arionkoder.com