Learn AI

Unsupervised vs Supervised AI: Not all AI is created equal

But how can you figure out which functions within your business can actually be transformed by AI? What are the quality limitations? How can you evaluate which business service companies are using AI effectively while others could be selling hyped up linear algebra? The best way to know if an AI product is right for your business is by asking the right questions.

Today’s digital world is flooded with more data than ever. The cliché example for big data is “It’s like finding a needle in a haystack,” but digital world problems are on a scale that makes this analogy seem silly. After all, there are many simple solutions to find a needle in a haystack that are readily available. For instance, one could use a magnet or dump the haystack into a pool of water and watch the needle sink. Big data problems are much harder, they are more like finding one specific needle in a giant stack of similar needles. There are no simple solutions to big data problems.

There are many ways to tackle big data problems. We've all heard the buzzwords: Deep Analytics, Data Visualization, Machine Learning, Artificial Intelligence, Neural Networks, Deep Space Force Calculus (The last one isn't real, but who knows what the future holds?) Significant business investments are being made into products that claim to use these technologies. Before businesses invest, it's important to understand the capabilities and limitations of the underlying technology.

Consider artificial intelligence, any business thought leader will tell you it's the wave of the future. But how can you figure out which functions within your business can actually be transformed by AI? What are the quality limitations? How can you evaluate which business service companies are using AI effectively while others could be selling hyped up linear algebra? The best way to know if an AI product is right for your business is by asking the right questions.

Question 1: Is your AI Unsupervised or Supervised?

Artificial intelligence, is like human intelligence. There is wide variation in types, skill-sets, and ability. To understand the capabilities of a product, first segment AI products into supervised or unsupervised.

Unsupervised AI

Unsupervised AI Key points:

  • Sort data to find a pattern.
  • Flexible on the outcome of analysis
  • Can find insights that humans might not expect.

Unsupervised AI allows the AI to form its own conclusions based on the data set. Machine brains are highly effective at finding common features and sorting based on a categorical filter.

Credit: https://towardsdatascience.com/supervised-vs-unsupervised-learning-14f68e32ea8d

Unsupervised AI is great for analyzing large quantities of categorical data, in the image above unsupervised AI could easily sort by other filters like eye color, clothing, species etc. It can even figure out different filters that it's programmers don't explicitly state like "male ducks with no shoes."

Unsupervised learning is helpful for people trying to make sense of massive data sets. Unsupervised learning's best use cases are when people are flexible on the outcome they would like at the end of their workflow. It can provide insights that humans might not anticipate. For instance, based on the analysis of over 1.6 million transactions on iseecars.com the best used cars to buy are orange cars. Exactly why orange cars maintain their value better than other colored vehicles is unclear, but the AI's statistical argument based on the provided data set is valid.

Credit: https://www.iseecars.com/car-colors-matter-study

Supervised AI

Supervised AI Key points:

  • Human expected outcomes shape AI solution
  • Humans teach AI how to solve problem, then AI mimics human.
  • Supervised AI is more suitable for highly customizable tasks than unsupervised AI.

Supervised AI uses human expected outcomes for a given workflow to shape how the AI produces the optimal solution. Supervised AI is taught how to make decisions by the humans managing the workflow. Think of a teacher teaching a student the correct answers to questions in a classroom.

Over time, the student will begin to recognize the correct answers on their own. If for some reason the correct answer changes, the teacher can go back and reteach the student the new correct answer. This feature makes supervised AI more customizable than unsupervised AI. With the right data set, supervised AI can actually become better at predicting right answers than its teacher.

Credit: https://www.lexalytics.com/technology/machine-learning

What if you could customize your own AI? An AI that mimics your skills, instincts, and experience then applies them to the tasks that you don’t want to do.

What tasks would you choose to automate through AI? What are the business risks associated with using supervised vs unsupervised AI? Imagine a world where you could spend more time and energy on the parts of your job that you love. This potential world is the promise of AI in the workplace.

Latest Articles

View All

The use cases for AI you didn't know existed

You've probably read about how AI is being used in data security, financial trading, healthcare, fraud detection, or elsewhere. AI has taken hold in these industries, in part, because there is “big data” associated with each and ready-made AI solutions. But, what about the industries or use cases that don’t involve mountains of numerical data?

February 3, 2021

Talent classification cited as a new way for organizations to improve internal talent mobility

In the December 7th issue of First Analysis Quarterly Insights, managing director Corey Greendale concludes that there’s a significant information gap that makes it difficult for organizations to optimize internal talent mobility.

February 3, 2021

What it’s like to be intern at Vettd.ai

Vettd.ai college interns are gaining a unique perspective about the difference between classroom learning and real-world experience. They are being challenged to use open-source and low-cost software components to create prototype applications to compliment or compete with alternative approaches to natural language processing (NLP) solutions.

February 3, 2021