Machine Learning is not 'Pixie Dust'

Google Cloud Next London

Today, I had the opportunity to attend Google Cloud Next in London. Even though the event focuses on their Cloud Platform offerings, there were various sessions about Big Data, Machine Learning, Tensorflow, GPUs and how businesses are using the Google Cloud Platform + ML to enhance their business processes.

It is exciting to see and understand how different industries (finance, travel, retail, e-commerce) use Machine Learning to understand their customers better or just automate tasks that a computer wasn’t able to do before. For example, in Retail, ML can be used to recommend users a product based on their behaviour or find the right product based on their search, instead of “hardcoding” rules.

I want to highlight something that Ram Ramanathan - Product Manager @ Google, said during his presentation today

Machine Learning is not Pixie dust. Is not something that you sprinkle over your business and it starts doing magical things.

Lately, there is quite some hype around Machine Learning / Deep Learning, and there are companies that feel the need of implementing something using ML just because is considered trendy. It is true that some of the problems that ML has helped solve lately seem to be magical, but they’ve solved a real business issue/need.

The goal shouldn’t be -> We need to implement something using ML because everyone is doing it.

The goal should be -> The business has a problem/need. How can we improve, change the business? And ask ourselves if ML is a tool that can help us achieve it.

In the end, the problem we are solving should benefit the business. This could be by creating a financial benefit, simplifying their processes, drive customer engagement, etc.

Google Natural Language API

There are easy ways of enhancing a business using Machine Learning, that doesn’t even require huge amounts of data, creating or training a model. It could be as simple as extracting what customers are saying about your business/product in Social Media and do sentiment analysis using one of the publicly available Machine Learning APIs, e.g. Google’s Natural Language API.

Written on May 4, 2017