You’re smart. You’re informed. You’re probably wondering why the people in the smart rooms whisper among themselves about machine learning. It is the usual recipe for buzz: potential and puzzlement about how to capitalise on it.

It’s fair to say machine learning is the hot tip everyone’s heard but few know what to do with. At Hunted Hive, we do know. Read on for some details to help you sieve the rumours you’re hearing.

Machine learning in theory

Machine learning could be thought of as the onramp to Artificial Intelligence. In essence, the practice involves using heavy-duty mathematics that feed back into themselves. By doing this, machine learning systems teach themselves to find and extract meaningful information from large and/or complicated datasets.

Yes, statistical analysis has been around for centuries, but machine learning is different in that it isn’t just a technique to find results, it’s a technique that finds its own ways to find results.

This unique capability be used for:

  • Data filtering

  • Data mining

  • Data analysis

  • Optimisation

  • Prediction

  • Reporting / Analytics.

The breadth and depth of capability is so impressive (and unknown), that it distracts lay-people from the machine learning’s limitations. It can’t, for example, self-enable.

The wildest rumours also ignore the fact that machine learning is just a highly advanced, single-purpose tool.

Building it and deploying it still requires a great deal of work from a human development team.

Nor is machine learning useful in situations where there is little quantitative data or for which the data doesn’t meet extreme validity criteria.

The quality of the data is crucial because a machine learning system can only work with what it is given and has no way to ‘choose’ whether the data it works with is accurate - that’s a separate functionality again and still largely in the human realm.

Things may change in the future, but for the moment you could say that machine learning is smart in that it analyses data, but not intelligent in the sense that it “understands” data.

Machine learning in practice

With the broad theoretical basis out of the way, the natural question is what this esoteric capability is good for. In practical terms, machine learning can find applications wherever there are large datasets for which advanced insights are highly valued. Imagine machine learning tools with which:

  • Investors can identify inefficiencies in the stock, real estate or energy markets

  • Brokers can assess and minimise insurance and investment

  • Marketers can identify hidden trends in campaign response

  • Consultants can find solutions through aggregating complex reporting information

  • CTOs can predict and track security vulnerabilities and threats

  • Efficiency advisors can identify business process inefficiencies

  • Businesses generally can automate processes and reduce admin loads.

Because the field shows so much promise, universities and think-tanks around the world make advances in machine learning algorithms and systems on a near-constant basis. A new raft of potential applications arises every week.


Commercialising machine learning

With knowledge of machine learning fundamentals spreading quickly, the limiting factor in advanced statistical analysis is no longer the scarcity and expense of human experts.

It means statistical experiment and analysis can be launched from datasets almost as soon as they are acquired and pre-processed.

For example, finding optimal investment strategies in commercial property could be conducted as soon as the right data on pricing, features, location and economic factors are acquired, cleansed and input.

The Hunted Hive break-down of a machine learning development project looks like this:


Because a machine learning system has no intuitive or cognitive biases (unlike any human learning system), successful deployment will either:

  • affirm insights that are already held (common sense)

  • yield evidence to encourage new conjectures

  • find meaningful trends in datasets too large for traditional means to work with

  • yield insights that are mathematically provable yet so counter-intuitive that a human expert would be unable to hypothesise for them.

Whatever the application, using machine learning still requires considerable research and development to make a dataset usable by the system.

Further, each dataset exposes new challenges in identifying useful patterns or procedures. The algorithms that work with one dataset may not be applicable for another.

Bringing machine learning to the real world

Perhaps you’re now getting a sense why so many people are talking about machine learning, but so few are well informed. It’s exciting yet hard to grasp, let alone stay current with.

By having some idea of its limitations and uses while focusing on its practical applications, business people can, however, have a nose for when rumours deserve to stay just that.

It also means when you’ll be able to sense whether the machine learning buzz surrounding a topic might have something to it.

To know more about the business of machine learning, ask us.


Hunted Hive has been researching and working with machine learning for several years and has real-world experience with the technology and techniques.