Three common machine learning misconceptions
Machine learning (ML) is an area of computer science that uses data to extract algorithms and learning models and apply "learned" generalizations to new situations, including perform tasks without direct human programming. With increasing use cases in modern data analytics, it’s helpful to demystify common machine learning misconceptions to understand how we can take advantage of machines’ powerful potential.
Misconception: Machine learning and artificial intelligence will replace humans
Machines aren't coming for your job. In fact, machine learning and artificial intelligence (AI) are more likely to help you do your job much better, and allow you a greater focus on the fulfilling and critically human elements of your role—including creativity and strategy.
While ML and AI are useful in many applications, there are plenty of areas in which these technologies are not particularly strong and require human influence, intervention, or oversight. This includes:
- Long-term planning
- Abstract or creative thinking
- Understanding cause and effect
- Making decisions that require domain knowledge or context
Human judgment is also necessary to fight inherent bias in ML algorithms. Even as technological advancements occur, we may never see perfect, completely algorithmic solutions. Explainability and transparency are important for humans to trust the outputs and recommendations of machines; we have to understand what's going on inside the "black box" of their models to fully trust and integrate machine-influenced decision-making in our businesses and lives.
Misconception: Machines learn from experiences
Contrary to popular belief, machine learning isn't dependent on experiences, but rather on data. You can’t just turn a computer loose to attempt to solve a problem—machines need data to learn from and create algorithms to apply to future situations, which includes:
- A method to classify or represent the components of the data set
- Metrics to score or evaluate success
- Optimizing the model parameters to the data
This works by extracting a generalized explanation, like an abstract story, from the data set, which can involve complex patterns or hidden regularities that a human might have trouble identifying. In financial institutions today, ML is used to analyze transactional data to detect and flag irregularities that may be fraudulent charges, or assess risks and make recommendations for lending.
A simpler example would be to introduce photographs of donuts to a machine so it could determine whether or not a new photograph contains a donut. First, we'd provide photographs with and without donuts, and tell the image classifier which are which. This provides the data with which the machine will build a model to make a deterministic prediction to distinguish “donut” from “non-donut.” Then, we can introduce a new image and the machine would apply it's algorithmic model and make a decision—is there a donut in this photo or not?
Misconception: Machine learning is the same as artificial intelligence
Artificial intelligence and machine learning are different, but related concepts. One way to think about the relationship of AI and ML is that the former is a problem while the latter is one solution attempting to solve it. If the end goal is that a computer can solve a problem with the cognitive abilities of (human) intelligence, the process of algorithms through data to apply to new, and larger situations is one method of getting there.
To help with the distinction, consider AI as tackling problems that are easy for humans and difficult for machines, like computer vision. To expand on our donut example, let's introduce a new challenge: to teach a computer to discern bagels from donuts. This is something much simpler for a human, but more challenging for a computer. Here, AI is the machine's successful ability to recognize bagels vs. donuts (the problem), whereas ML is the way the computer may learn to reach a conclusion when shown a new photograph (the solution).
Conversely, ML shines and is often employed in situations that are easier for machines than humans—like executing complex, mathematical algorithms or using probabilistic calculations. The computational power of machines helps quickly execute more challenging tasks, or uncover patterns that might be missed by a human.
Machine learning use cases in modern analytics
Many organizations are bringing ML into their enterprise data analytics practices to help identify hidden insights and make smarter recommendations to inform business decisions. This is especially helpful in big data analytics, handle increasingly large and complex data sets. Machine learning can also identify behavioral trends within an organization to make suggestions to users that appear similar to others—like which data sources to use in data prep or analysis, or which analytical content is the most relevant to help them answer a particular question. Other areas of continued development include advanced and predictive analytics. Machine learning can help automate advanced statistical analyses and automatically apply models with the highest confidence, as is the case with augmented analytics, allowing less advanced users to take advantage of complex models. More advanced users can explore and modify calculations, which not only addresses trust and transparency, but also allows testing of different what-if scenarios. Machine learning is also being used in analytics to help users query their data with natural language. This essentially means learning to interpret human intent and semantics behind questions, and translating requests into a structured query language. With advances in natural language and other smart analytics capabilities powered by AI and ML, people without traditional data skills will be able to work with data in new and exciting ways to get new insights.