When most people think of machine learning or AI, they have a negative view of the term. They either think of robots that are taking jobs away from humans or they think of Skynet or HAL 9000, sinister sentinel beings that are poised to take over the world. In reality, we use machine learning every day with our phones and their smart assistants Siri and Cortana, or smart cars that analyze surroundings. Even while we browse the web we are hit with advertising based on our shopping habits. Machine learning is all around us and is the driving force of AI. But to fully understand machine’s learning impact on society, we need to look at the past, present, and future.
What is machine learning?
Machine learning is the scientific study of algorithms and statistical models that computer systems use to perform a certain task without relying on specific instruction, but on patterns and inference instead. The term machine learning was first coined by Arthur Samuel in 1959 while he was working at IBM. His ultimate goal was to create artificial intelligence (AI) and machine learning became a subset of his quest. After his discovery, there was a rift between theories and it wasn’t until the 1990s that machine learning really took off. This period was known as the AI Winter. The field changed its focus from achieving AI to solving problems of a more practical nature by borrowing models from statistics and probability theory. Machine learning also benefited massively from the invention of the world wide web as it gained access to a mass amount of data.
Different Types of Machine Learning
In order for machine learning to work, a lot of data is required to predict future outcomes or to train a machine to perform tasks. While machine learning started with a mass amount of unchanged static data, it has shifted to learning models that process data in real-time.
There are four main approaches to machine learning:
- Supervised learning: training data to perform a certain output
- Unsupervised learning: training data with no clear output
- Semi-supervised learning: training data with few desired outputs
- Reinforcement learning: rewarding the artificial agent based on what it does
Who is using machine learning?
You probably don’t even notice the machine learning that is going on in your everyday lives. That spam filter in your inbox? That’s machine learning. Machine learning is also already in most of our homes in the form of smart speakers, smart plugs, and smart thermostats. In the US alone, there are 47.3 million adults that have access to a smart speaker. However, machine learning has a big impact on certain industries of our world.
Machine learning can streamline internal processes very easily within the finance sector because of the large amount of data that this sector collects. Banks and financial institutions use machine learning to detect fraud with data mining (an unsupervised type of learning used to discover data patterns). A recent report indicated that in the next 15 years, robots will be able to perform 75% of financial services jobs with machine learning.
A recent article from Deloitte states, “Governments collect vast amounts of data on everything from health care, housing, and education to domestic and national security—both directly and through nonprofits that they support. Governments also produce data, such as census data, labor information, financial market information, weather data, and global positioning system (GPS) data.” With this data, government agencies in areas such as utilities, military, and infrastructure use machine learning to increase efficiency, detect fraud, and save money. Some infrastructure agencies use this data to predict when potholes will form on city streets, while the military uses machine learning to predict mechanical failures on tanks.
Wearable devices and sensors make machine learning a fast-growing trend in the healthcare industry. These devices help medical professionals identify any trends that might help diagnose a patient, which results in better treatment options. With large image databases in radiology, machine learning can quickly assess imagery that would take longer for medical experts to process. For example, with machine learning, Google trained computers to detect cancer in patients with 89% accuracy. Stanford is using a deep learning algorithm to predict skin cancer and when patients will die to give better hospice care.
Not only is machine learning used for autonomous vehicles, but it also has a huge impact on how your packages are delivered. Machine learning analyzes drivers’ GPS, where the package is being delivered when the customer receives the package, their speed, and the weather to find the most efficient route possible for the delivery driver.
Criticisms of Machine Learning
With all the outstanding accomplishments of machine learning, there has to be some criticism of the application. According to Dr. Allen from Rice University, machine learning is creating a crisis in science. Rice’s argument is that a growing amount of scientific research is using machine learning on data that has already been collected. She argues that the solutions that these scientists come up with are wrong because it is only applicable in that data set, not the real world. Dr. Allen states, “There is a general recognition of a reproducibility crisis in science right now. I would venture to argue that a huge part of that does come from the use of machine learning techniques in science.” She calls this a reproducibility crisis.
Dr. Rice isn’t the only critic of machine learning. Gary Marcus, a psychology professor at New York University has a long list of grievances against deep learning. They include a heavy reliance on large data sets to its susceptibility to machine bias and its inability to handle abstract reasoning. Marcus stated that his greatest fear is that AI will get pigeonholed as a “local minimum, focusing too much on the detailed exploration of a particular class of accessible but limited models,” further concluding that, “one of the biggest risks in the current overhyping of AI is another AI winter”.
What’s the next step in machine learning?
The more data we create and the more we study machine learning and its impact on technology it becomes clear that there are also ethical responsibilities we must uphold as a society. Right now, only a few tech companies are fully invested in machine learning and data collection. What happens if this technology becomes a monopoly? Will this technology only be accessible by those with wealth? What if the critics of machine learning are right and another AI winter will come to fruition? It will be hard to predict what the future knows.