What I think about Machine Learning

Machine Learning is a term that is on everyone’s tongue these days. A survey conducted by ServiceNow revealed that 89% of organizations have deployed Machine Learning or have plans to deploy it. Then there is the news of companies like Apple, Google, and Facebook, jumping onto the Machine Learning bandwagon. Many, I included, have been quite fascinated by the number of opportunities and benefits Machine Learning is said to bring to the table and there is enough been written about the ‘march of the machines’. So, in tune with the spirit of Machine Learning, here is a bit of my own learning. Here is what I feel this technology is all about.

Machine Learning – What is it?

Machine Learning and Artificial Intelligence are two technology terms that many use interchangeably at their convenience. However, Machine Learning seems to be a category of artificial intelligence, one that is being leveraged to relieve IT of the common problem of managing non-routine cognitive work.

As a specific subset of AI, Machine Learning focuses specifically on training machines to self-learn without explicit programming. Using Machine Learning, computer programs can easily learn, grow, change, and improve by themselves. So, Machine Learning consists of:

?   The computational algorithm for making core determinations

?   The variables and features constituting a decision

?   The base knowledge that trains the system

?   The experience generated during training which is used to predict result

The concept of Machine Learning can perhaps be traced back to Alan Turing’s famous WWII Enigma Machine. However, the ability to apply complex mathematical algorithms automatically, iteratively and fast has developed now, giving the technology the power to independently adapt to the new data.

Why is Machine Learning gaining prominence?

The interest in Machine Learning is driven by the growing volumes of data and the inevitable focus on faster analytics and growing computational power. We are moving deeper into the data economy and data sets are becoming increasingly complex. And the more complex the data sets, the greater is the potential to uncover relevant and impactful insights.

All organizations and all industries are now trying to leverage data to discover relationships between everything that impacts a business – be it pricing, business models, revenue, customer relationships, risk…the list goes on. It is only when organizations can analyze bigger and more complex data sets that they can identify new opportunities faster and avoid risks. This technology is transformational, to say the least! It removes the drudgery out of data and analytics and provides deep insights drawn from data patterns that would have otherwise gone unnoticed.

As the world becomes more data-rich, Machine Learning is starting to assume a very visible role. Google’s self-drive car, friend suggestions of Facebook, recommendations on Netflix, virtual assistants like Siri and Alexa, product recommendations on Amazon are all everyday examples of machine learning at work. Industries such as retail, banking, manufacturing, IT, and others are already leveraging Machine Learning to enable greater personalization, enable real-time decision making, identify new opportunities, and make the enterprise more secure and proactive to respond to change dynamically, especially in this age of digital transformation.

The Perils of Machine Learning

Machine Learning clearly has a lot of promise and is becoming an integral part of transforming the organization into a cognitive enterprise. However, we are still at a nascent stage of this technology-driven transformation of the enterprise and for Machine Learning to do its magic, we need data. And this is where I believe the challenge begins.

Obviously, Machine Learning only works when the “machine” gets data. This teaching data that is provided to the machine has to be clean, accurate and has to clearly drive towards the desired outcomes. However, enterprises have to navigate a massive challenge of getting the humongous volumes of clean, accurate and contextual data. 

Then comes the challenge of mitigating data bias. This is a real problem -research shows that Machine Learning algorithms not only learn to understand human language but also learn to replicate human bias. For eg. in an interesting effort, researchers from Boston University and Microsoft Research attempted to remove linguistic bias without altering the meaning of the words.

Organizations also have to work around operational bias. What happens if an incident occurs and is sent to a team who might not be the best suited to handle the problem? When the machine is fed data it will come up with a solution fast. However, this solution might not be the right one. Why? Simply because bad data will only lead to bad insights – only that these will be delivered faster.

The hype around Machine Learning also acts as a double-edged sword – on the one hand, it helps because everyone (including the leadership) is more receptive to it. But on the other hand, the expectations of what Machine Learning can do can be unrealistic and also unreasonable.

Then there is the talent challenge. To reap the potential benefits, organizations need the right people with the technical, cognitive, and creative ability to understand and implement this technology. Over time this problem will have to be addressed by the market forces as well as advanced automation.

Clearly, while there are multiple opportunities with Machine Learning, organizations must also deal with the clear and present challenges.

To recap, Machine Learning is here and is here to stay. However, for it to deliver on its promise huge volumes of data are needed and the hygiene of the data cannot be compromised. Conscious and unconscious biases must be isolated and removed from the data. And finally, we need to keep our expectations from machine learning real.

Being a technology enthusiast, I would like to believe that technology will help us deliver a better future. But it is also true that that will take some time, effort and data. Lots and lots of data.