Utilizing Machine Learning requires restating your problem differently

Vaibhav Gadodia
Jul 21, 2017 · 3 min read

Humans are continuously inventing better ways to get work done. The Industrial Revolution was when we invented ways to produce more efficiently. The micro computer revolution made computation more efficient. The internet revolution improved efficiency and scale of communication and information search. We are in the middle of the Machine Learning revolution.

But what does the Machine Learning revolution make more efficient? What does it scale?

Machine learning scales prediction.

Humans are very good at prediction — which is the ability to make an observation or statement about a certain event, usually based on experience or knowledge. Doctors predict — when they give a diagnosis. A human driver predicts — when she notices a ball bouncing across the road. Parents predict — when they don’t hear too much noise from their young children’s rooms.

With advances in the machine intelligence field, it is becoming cheaper and easier to get computers to predict. Computers can be fed millions of structured and unstructured data points and they can then predict based on what the computers ‘learn’ by analyzing these data points.

In order to use machine learning to solve a problem, we must learn to pose our problem as a prediction problem

There are problem-sets which very naturally lend themselves to be solved through machine learning. Predicting disease based on historical test data and subsequent diagnosis is one such problem. Or fraud detection in financial systems. But we can utilize machine learning to solve challenges which don’t look like a prediction problem initially.

The clichéd example of one such case is autonomous driving ( warning: over-simplification ahead). One might think that autonomous driving can be solved by adding enough if-then-else logic to the driving algorithm.

if (object_distance < 100) then slow_down() else keep_driving();

To make this a prediction problem, companies like Tesla are feeding millions of hours (or miles) of driving data from cars being driven by humans to machine learning algorithms. The machines can now look at the various situations that these humans find themselves in and can predict what a human would do in a given situation. This prediction capability can then be applied to the autonomous driving algorithm.

Here’s a hypothetical example (maybe) — in large cities, where should law enforcement be deployed? Let’s say enough data is collected from cameras, sensors, time of day, historical crime. A machine could learn that whenever there are higher late-night visitors in an area which has many bars, there is a higher chance of crime. A machine could learn to predict potential hotspots in real time and deploy law enforcement resources accordingly.

If we can re-word our problems as prediction problems, the possibilities of machine learning applications are endless.

As a parent, one of the most difficult decisions for me is to figure out how to guide my children in terms of education choices. What will they be naturally good at? At work, one of the challenges is how to pick people who will likely fit the best in our company’s culture. Challenges around learning, education, recruitment, and career progression can be re-worded as prediction problems. There are already many data points that can be used for machine learning in this space. But we can start experimenting with other data points such as scanning our brains! The ultimate machine learning technology (although currently fictional) for this is the Sorting Hat.

image credit: Jerome K Moore

Arthur C. Clarke said, “any sufficiently advanced technology is indistinguishable from magic.” So, though it seems that the Sorting Hat derives the capabilities it has from magic, it is potentially a piece of advanced technology which can scan the student’s brain, tap into the profiling data that Hogwarts certainly has about the student, and also tap in to the machine intelligence it has acquired by sorting thousands of students to various houses in the past — it uses all this to predict where the student has the best chance of succeeding.

Let’s take a step back, look at our current challenges, and then try to think of them as prediction problems.

)

Vaibhav Gadodia

Written by

My topics of interest: technology, nature, travel, parenting, design, Nagarro.

Vaibhav’s musings

Articles and opinions mostly on technology, ways of working, and the occasional social commentary.

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade