What is Machine Learning?
In today’s world, there is a lot of virtual activity on various websites, from shopping online to entertainment and social media. A lot of our daily activity takes place on the internet, and when we interact with these sites we generate data. When all of us use these sites together, we generate a gargantuan amount of data. In fact, in a single minute, we collectively send over 16 million text messages, ask for 18 million weather requests from weather.com, and in one year we will create 2.5 quintillion bits of new data (Marr). All of this data is largely abstract and humans alone can’t do too much with the data we create. However, with the help of computers and machine learning, we are able to process our data and create some amazing tools to help us in small and large daily activities. Tools such as news or shopping recommendations, autocomplete suggestions, and even fully autonomous driving! Often, we use these tools without realizing it but it begs the question, how does a machine look through data and decide that I might be interested in watching a new drama on Netflix? The best way to answer that question is to understand what machine learning is.
Machine learning can be oversimplified to a computer looking at all the data given to it and then making a guess as to what patterns it can find in the data. For example, if a machine learning algorithm is given a dataset on television viewing habits, it might see that most people who watch American History Documentaries also watch British History Documentaries. Then the machine learning algorithm would take note of this pattern and predict that if someone watches American History they would also watch British History. In a nutshell, that is how a new show might be recommended to you.
More mechanically, this type of machine learning process goes under the branch of Supervised Learning. This kind of machine learning takes in input data that has the correct output data and then notes this relationship into a model. This model will then take in new input data that has not been seen before and predict the associated outcome (Nichols, Herbert, & Baker). These supervised machine learning algorithms can be used for most predictive tasks such as recommending a new show on a streaming site or deciding whether an image is a stop sign for a self-driving car (Brown). Conceptually, machine learning isn’t as daunting as it might seem but it is still deep and complex with many more aspects to it. More complicated machine learning branches off into Natural Language Processing, Neural Networks, and Deep Learning. Each topic may be different in how they solve problems but they all attempt to achieve the same general goal of looking at data and then deciphering what patterns within the data might be useful.
Although the various machine learning techniques can be extremely effective at times and seem almost flawless, they are anything but perfect. Like humans, machine learning algorithms can be inaccurate, make mistakes, and have biases. For instance, an algorithm might pick up on a pattern within data too strongly. This might make the algorithm suggest only one output no matter the input data, known as model overfitting (Nichols, Herbert, & Baker). This overfitting of data could lead to the machine learning model being inaccurate at classification tasks such as grouping patients by whether they have a disease or not. For example, if a dataset contains an overwhelming majority of patients without heart disease then the algorithm might only predict all patients to not have heart disease simply because it has not seen enough cases of heart disease to detect a pattern in the data.
In addition, machine learning algorithms are only as good as the data they are given and the quantity they are given. They require the proper balance of quality and quantity to be effective. If they are given too much data then the algorithm runs the risk of overfitting. If they receive too little data, the algorithm becomes more inaccurate. Also, if an algorithm is given bad data, then it will not detect patterns correctly and will, in turn, not predict accurately. Bad data could be anything from irrelevant data, data with errors, or even biased data. In the case of biased data, the algorithm could pick up on the bias and output a biased prediction as a result. For instance, Facebook utilized a machine-learning algorithm to increase engagement with their ads but because the data they used involved people engaging with extreme content, the algorithm picked up on this pattern and pushed ads that were more extreme (Brown). Overall, it is important to note that machine learning is far from a perfect tool. Used properly, machine learning is very effective and useful but it must be used carefully and responsibly in order to achieve proper and accurate results.
In summary, Machine Learning is the use of computers and mathematics to detect patterns within large amounts of data to come to a useful conclusion about the data. Largely, machine learning is something that we do for ourselves every day. When we drive or cook, for instance, we are taking in data and making decisions based on the patterns we have noticed from the data we have seen. This process of taking in input data and using it to produce a decision or output is the essence of what machine learning algorithms do. The only difference between us and the algorithm is that the algorithm processes a lot more data in one instant than we do. Also just like us, machine learning has its flaws that can lead to inaccurate or undesired results that could possibly be damaging depending on the context. Machine learning is a relatively new technology but is also revolutionary in the things it can do and because of this, we must be responsible in using it and making sure it has the desired effects before implementing the algorithm into real-world applications. Conversely, it is exciting to see what else machine learning can do that we have not yet improved or discovered!
Works Cited:
Brown, Sara. April 21, 2021. Machine Learning Explained. MITSloan. https://mitsloan.mit.edu/ideas-made-to-matter/machine-learning-explained.
Nichols, J. A., Herbert Chan, H. W., & Baker, M. (2019). Machine learning: applications of artificial intelligence to imaging and diagnosis. Biophysical reviews, 11(1), 111–118. https://doi.org/10.1007/s12551-018-0449-9
Marr, Bernard. May 21, 2018. How Much Data Do We Create Every Day? The Mind-Blowing Stats Everyone Should Read. Forbes. https://www.forbes.com/sites/bernardmarr/2018/05/21/how-much-data-do-we-create-every-day-the-mind-blowing-stats-everyone-should-read/?sh=37a23df360ba.
