Predicting our Future with Machine Models

The big data revolution is upon us. Companies are struggling to transform their digital capabilities while keeping their core business alive. Alas, there is no silver bullet but only "blood, toil, tears and sweat" as Winston Churchill once said in 1940 during a cabinet meeting. This context is different but the efforts are the same - companies around the world are feeling it. The struggle is real.

In Singapore, people are working on it. Looking for new ways to work with technology to lower human hours, increase efficiencies, stretch bottom lines and so on. In the end, revenue generation remains at top priority in the business world. Unfortunately there are no shortcuts in this digital transformation.

In the past year, I have been involved in developing an AI lab in professional services where I see a spectrum of new technologies go deep into systems. I also saw how people behave when they interact with those who use these technologies and those who create them.

From examining various machine models to predicting demand and attrition, to automating key processes to optimize human resources, I found that being intelligent is about being predictive. It would be a mistake to say we are smarter than robots. But it will be safe to say we can feel and connect worldly dimensions better than robots. And yes, they are catching up in every sense. 

Picking our Battles

What to do with machine models? Deploy them, one would say. Improve our own well being. Bring it on.

Humans size up a person much faster than a machine model. It doesn't require huge amounts of data. Surely these machine models can create outcomes at scale. There is also much more objectivity as opposed to what we (as humans) can't see.

For example, if employers look deeper on why people leave their companies, one might say that it's a tough journey. Many take mental short-cuts with long explanations. In closed door sessions, they discuss matters on very subjective matters. Some issues will most likely come up: Long working hours, lousy boss, horrible colleagues and a crazy work load. That's a fair assumption. Yet if we look deeper into the issue, it could be aspirational, a mid-life career change or simply adjusting to family needs. Regression models can tell us more. In any event, take note that predictions is not causality. When that becomes the case, problems arise, many of which are created by human mental models.

Here are some examples of prediction problems in a business:

• Making personalized recommendations for customers without looking at objective data sets. Didn't manage to go deeper than expected.
• Rating credit risks of loan applicants and forming up "Current Estimated Potential" or CEP too quickly - this means anticipating the future performance of employees
• Manipulating data sets to form stereotypes to suit outcomes. In other words, playing the game to suit needs.

Making Sense of the World

Before even looking at outcomes and solutions, there's a need to separate the signal from the noise. There are needs, wants and preferences, put simply. People tend to lump of this together. The ability to predict does not just come from machine models but a careful calculation of trends, history and an understanding of how humans, in their most basic terms, would behave.

Putting predictive thinking into another context, say Brexit or the strong populist movement happening around the world today. It's quite easy to make hypothetical assumptions that the world is getting more dangerous today. But the fact is: it is not. The world is safer than ever before since the World Wars. But it has definitely become more unpredictable. Results change unexpectedly. People vote differently based on frustration and resentment. Events happen not solely based on trends. Many of issues that come up today are not at all logical, in any sense.

To test any model, scientists would try cross-validation, that is to be sure of testing whether a model is accurate from the given sample size. When using a model or any type of methodology, you want to know you can use it reliably. Now with quick entry of deep learning capabilities, decision-makers can use both technical skill, instinct and cognitive abilities captured from a robot to test outcomes.

Bringing it Together

Already, companies have started to collect, analyze and act on data. We all know the power of the app and anatomy of our competitors. The future ahead depends on differentiation and away from product-driven approaches. And thanks to cheaper offshore labour, costs would work our differently these days. With analytics, you can know what your customer wants, how much they are willing to pay and the price to prevent problems.

With a good mix of human judgment, goodwill and partnership, hopefully we can all see how we can fit into this exciting world of science and technology better. In the next few months, I will be developing new methods to design, deconstruct and deploy new technologies, and look forward to reaching out to companies and communities.

Imagine we are building a prediction model using 5 million past customers and we want to know how accurate these predictions will be for future customers.

What do you think our future in Singapore will be like? How would you predict outcomes? What are your reasons? Feel free to share with me.