V-Squared Data Strategy Consulting

At IBM Connect I saw a company faced with a challenge common to many software vendors and to machine learning as a whole. Get customers who are deeply rooted in the old software model to join them in moving towards this new era of machine learning solutions. From startups to established players, the push into mainstream enterprise adoption has been an uphill, sometime vertical, climb.

IBM’s description of machine learning software, what they’re calling cognitive, is rooted in the traditional software paradigm. That paradigm features programs and apps that do static things repeatedly and reliably. That paradigm reinforces the concept that it should always work and any deviation from expected results is a defect. That paradigm showcases certainty in the known, repeatable, and highly stable.

None of that applies to machine learning based software. That fact was on display in a number of presentations I went to. When a developer using Watson for NLP ran into issues with the API detecting a French name and assuming the language was French, he stopped using the API altogether. During the keynote, a speaker used Google Home’s speech to text interface and the first attempt failed. I can’t tell you how many times that was brought up in a derisive tone. The expectation for machine learning is for it to work in the traditional software paradigm. When that expectation isn’t met, customers dismiss it as either hype or just not ready.

All non-machine learning based software is now legacy software.
The problem at conferences like Connect, is that companies are perpetuating a standard which doesn’t work with machine learning based software. They’re acting as if the fundamentals of software functionality haven’t changed. We’re selling machine learning as an incremental change to software. The reality is, all software that doesn’t leverage machine learning is now legacy software.

The emergence of machine learning in software is a fork in the road. The traditional software model’s roadmap leads to a rapidly approaching dead end, obsolescence. It cannot meet modern users’ demands for personalization, scale, autonomous function, continuous improvement, or several others. However, many vendors don’t paint that picture for their customers. Machine learning is just another type of software.

The result is uncertainty on the part of many businesses, especially among IBM’s (and many software vendors') core customers. They have been told to expect certainty, stability, and consistency so when they don’t see it, they have good reason to doubt the technology. One the other side they’re also seeing practical business cases from companies succeeding with machine learning and data science. Hence the confusion. They want in on the success but the vendors they’ve come to trust with delivering software solutions don’t seem to offer anything that’s ready for prime time.

Time to rename Watson to more accurately reflect the new paradigm.
What IBM needs to do, is rename Watson. Watson is Sherlock Holms’ trusted, reliable, predictable, and not at all threatening to the genius that he assists partner (also a very famous IBMer so a powerful dual meaning). It is a great name choice for the traditional software paradigm message but again, we’re not in that era anymore. They need to start calling their cognitive offerings, Sheldon. Yes, that Sheldon from the Big Bang Theory.

If you’re not familiar with him, his character is a brilliant physicist. However, he has several, very pronounced quirks. Do anything outside of his expected norms and his reactions can be unexpected. He needs a lot of training to handle new situations and until he learns his lesson, the outcomes are not what you’d expect from most people. He’s socially inept and has a tough time understanding nuanced speech or emotion.

That’s an accurate description of how machine learning software operates now. Out of the box, machine learning is like a new employee. Depending on the amount of initial training data and sophistication of the model(s), that employee could behave like an intern all the way to a senior team member. As machine learning spends time learning on the job, the software improves dramatically. It moves from knowing how to accomplish intended tasks in a generic way to understanding how it’s done at a specific business or even for specific users.

Sheldon’s come a long way since the first season. He’s still rough around the edges but his friends have helped him handle a lot of real world situations on his own. Most businesses are waiting for that Sheldon, or machine learning, to show up before they start to adopt. The problem is, two years from now, it’ll still be season 1 Sheldon that they get. There will be advancements over the next two years but nothing that will change the paradigm of machine learning. The software will still need training and adaptation to become the more mature product businesses want.

Machine learning at the tipping point.
The enterprise is at the tipping point of adoption. The drive of early adopters was to be the first to figure out how to apply a new technology to their market. That’s still important, though with machine learning, there’s another dimension to consider. The software needs time to learn. It needs on the job training like any other new employee.

This new employee is different than software that you’ve hired before. Most software comes with a degree in accounting and five years of experience or a degree in HR with a focus on compliance. Sheldon has a degree in everything and very little real world experience. Rather than playing a single role within the business, it can contribute to any function or team. It’s a raw talent with great potential. It has a voracious appetite to learn and picks up complex tasks very quickly. It never tires and will amaze you with the connections it uncovers.

Sheldon’s ready to work and provide value on day 1 though there will be some missteps. Like other employees, as it becomes more experienced, that value rises dramatically. To get that value requires leadership and guided training. As machine learning strategists, we need to get businesses to think of machine learning in the same way they think of any new hire. We need to talk to businesses about how to read the machine learning resumes from TensorFlow, Watson, Azure for ML, AWS for ML and many others for fit within their organization.

Unlike a promising young talent, businesses can’t poach them from another company after they’ve had time to grow and mature. That’s because the knowledge and capabilities you train into Sheldon don’t leave the company for a better offer. Sheldon doesn’t come with the downside of turnover or attrition. Competitors won’t sell you Sheldon because it becomes a core component of their competitive advantage.

That’s the tipping point. Businesses need to be educated about the new paradigm for software or they’ll be left behind. Being left behind means higher operational costs and an inability to satisfy customers’ expectations of technology. Those disadvantages are significant. They’re not crippling but they are the kinds of issues that put businesses on a 3-5 year downturn with a costly, multi-year restructuring and catch-up phase. Our role, whether we’re machine learning strategists or vendors, is to educate businesses about the new paradigm and how to be ready for it. While we’re starting to hear more about machine learning for the enterprise, what will get us past the tipping point is talking about what’s different this time and how to manage the change.

Your Name: *
Your E-mail: *
Share E-mail: *
Message: *

Thank you for sharing this post.

Your recipient
will get your email shortly.