In this video, we're going to define Data Science, and compare it to Machine Learning. We'll provide examples as well so by the end you will understand both the differences and the overlap, between Machine Learning and Data Science. There can be some confusion around these fields, mainly because they have so much in common. They both rely on data and statistics, and use a lot of the same techniques. They're separate but overlapping disciplines. Data science is an interdisciplinary field, often described as the intersection between statistics, programming, and domain knowledge. I like pointing out the importance of visualization and communication as well, because data science is about using math and human ingenuity, to extract insights from data. Remember, we said Machine Learning was about generalizing from data, this slight shift from extracting insights, to generalizing is significant. Like data science, Machine Learning is at the intersection between statistics knowledge and programming skills, and you definitely want domain expertise when you're applying Machine Learning to a specific problem. But Machine Learning is also a subdiscipline of AI. We want to build systems that guess well on examples they haven't seen. In other words, we want the computer to build its own question-answering machine, rather than having to answer the questions for every possible example ourselves. To highlight the differences between data science and Machine Learning, let's look at two approaches to solving the game of two player limit Texas hold 'em. This example is close to home for us at Amii because the University of Alberta has been actively researching poker for decades. The Computer Poker Research Group is older than Amii itself. A Data Science approach to building a poker playing bot, is to look at the archives of human expert games and use statistical tools to determine what those humans have done. You may uncover all kinds of interesting and previously unknown techniques by doing this data analysis, and then use those insights to define the behavior of your poker bot. Ideally, your bot will then imitate excellent human play, but it is unlikely to do a lot better and may not have any way of dealing with completely novel situations. In contrast, a Machine Learning approach is to build a bot that plays the game, maybe even poorly. It may analyze human records and pull patterns from there or it may just play billions of games against itself, but it's not restricted to copying what has been done before. Machine Learning techniques led to the development of Cepheus, the best limit Texas hold 'em bot, and Deep Stack the first no limit bot to achieve superhuman play. Machine Learning uses data, but isn't limited to copying the data. Machine Learning systems always create a question answering machine. So while a data scientist may use the insights they have found to create a question answering machine, Machine Learning systems always create one. While Machine Learning systems may provide illustrations or explanations of the insights they use, they don't have to, whereas Data Science is about providing those insights to humans. Machine Learning scientists use many of the same tools as data scientists, like regression and clustering, and data scientists turn to machine learning algorithms such as topic models, and dimensionality reduction techniques, but our goals are different. Data science is focused on using human expertise, to better understand the data you have, and Machine Learning is focused on building question answering machines, to make predictions on unseen data. So now, you know what to say the next time you hear, "Machine Learning is just data science plus hype or data science is just a subset of Machine Learning." They're complimentary fields, and when we work together, we supercharge our technology.
