I accept cookies
Continue
We use cookies to ensure that we give you the best experience on our website and to show you advertising that is relevant to you. Read more about "Cookie Policy".
By continuing to surf on the website, you explicitly agree on the use of cookies on this website.
Mind+machine, Analytics, Artificial Intelligence, General Interest

Don’t Get your Basics Wrong – AI, Machine Learning, or Big Data

In recent times, my conversations with general managers have inevitably veered to the topic of Artificial Intelligence (AI) and Machine Learning (ML). What puzzled me was the lack of true understanding of these terms and the imprecision with which they were being used. Some general managers stated their need for better understanding, while others persisted with these terms, trying to impress the audience with half-knowledge or hoping that the other participants of the meeting don’t know the details either.

I realized that the key challenge, when talking about AI and ML, is determining what we really mean and what such technologies can really deliver at this point in time. It is vital that managers who decide on resource allocation for ML and AI projects are able to clearly state the benefits that products containing ‘smart’ technologies will deliver to their clients.

Overstating the market promise or misallocating resources can have a devastating effect on organizations. In the last 12 months, I have come across two serious cases of organizations having to write off USD 60 million and USD 30 million respectively in misguided AI / ML projects. Even worse, one of the organizations (operating in the healthcare domain) and the corresponding AI vendor suffered negative reputational effects.

As I have already mentioned, in most of my conversations I notice these terms being used interchangeably, and sometimes in very fuzzy or plainly erroneous ways. Moreover, these technologies’ ability to deliver is often overstated by enthusiastic program managers or semi-knowledgeable general managers and sales people.

One of the reasons for such over-enthusiasm might be the current hype around these topics, and the very public progress made by companies such as Google with TensorFlow and AlphaGo, Amazon, Microsoft, and IBM Watson.

The result? Everything, even a tool with a few basic rules programmed in a deterministic fashion, is being pushed as AI / ML. On top of that, throw in a few complex terms, such as deep neural networks, and you will emerge the winner in party talk.

All in all, there is a lack understanding of the difference between ‘true’ cognitive AI, in which the machine comes up with the rules or findings by itself (e.g., with unassisted deep learning); smart behavior trained by humans (e.g., with assisted deep learning); and rule-based expert systems in which a human programs deterministic rule sets that are run over large data sets. Often the terms AI and ML, and also Big Data are mentioned in one breath, although they are very different concepts. (Do you still remember the term Big Data? Not many of our clients refer to it anymore.)

Just because a product uses AI / ML in its marketing brochure, it does not mean that the product will solve the use case at hand or generate RoI. In fact, current developments suggest that specialized application of AI and ML, trained for specific and narrow use cases, deliver much better results. Managers should be highly cautious with statements such as ‘We have AI and ML in our product.’

In any business discussion involving AI, ML, or Big Data, it is important that these terms are used correctly, so that clients know exactly what outcome and performance to expect from projects that use these technologies. AI refers to ‘the broader concept of machines being able to carry out tasks in a way that we would consider “smart”.’ ML can be seen as an application of AI, where machines learn and improve, based on certain rule sets or patterns, independent of their ‘smartness’ at the start of a project.

A common misreading is that AI and ML products always need Big Data for training. While it is true that several ML applications require Big Data sets, e.g., in image recognition, several other applications don’t. They can live very well with ‘small data’, e.g., when a machine learns to optimize its user interface for a specific customer. Keep in mind that simple ML with ‘small data’ may provide very high RoI.

Of course, this barely scratches the surface, but the purpose of this blog post is to highlight the need for better understanding of underlying concepts and their ability to deliver results among the general management community.

Speak your Mind

Your email address will not be published. Required fields are marked *

  1. Rocky M says:

    The article is just like an Evalueserve Deliverable – Shallow information, Circular reasoning and a Caveat claiming no responsibility on the information quality at the end.

    1. Evalueserve says:

      Thank you for you honest feedback! We are sorry that you have a negative opinion about some of our deliverables and we would like to understand in more detail on what you base your feedback on. Please leave us a direct message at marketing@evalueserve.com, so we can look into it. Thank you!

  2. Ankit Kumar says:

    How do you classify Evalueserve’s SNAP product i.e. Machine Learning or Deep Learning?

    1. Evalueserve says:

      Currently we use Multinomial Naïve Bayesian Model which comes under Machine Learning. We are also looking at specific features that can be implemented with Deep Learning.