Machine learning has become an increasingly popular topic, both in the public consciousness and for businesses wanting to streamline their processes.
The line between speculation and reality can be blurry though, leading to quite a few misconceptions about the technology and its uses.
No ad to show here.
At its heart, it simply refers to the process of making digital tools to automate decision-making, in a way that gets progressively better over time based on the data received.
For technologists like myself, we see machine learning as one more kind of resource that helps us to develop smarter solutions for our clients.
In this article I want to unpack some of the issues around this technology, as well as some examples of how it can be used effectively.
Machine learning in the real world
For a start, machine learning can be used much more tangibly than you might think. The most common picture painted in the media is of algorithms dealing with relatively abstract qualities — things like risk or financial projections. As a result, it is often thought of in terms of highly complex decisions that may not have everyday applications.
The truth, however, is that it can be used by all kinds of businesses, in all kinds of markets.
I say this because we are often asked about the relevance of emerging technologies in the South African space.
The reality is when it comes to tools that help to streamline systems, there will always be relevant applications — the question is just about how to implement them in a way that takes the customer and context into account.
An example for the potential of machine learning to be used in a tangible way can be seen in a situation we often see in retail in South Africa.
Most stores will log product returns in physical ledgers which are then filed away. This form of record keeping serves a very narrow purpose of keeping track of products brought back into store, but doesn’t add any significant value to the way in which the business operates.
By digitising this system, we would be able to assign data points to the inputs, so entries could be tagged with information around the type of return, supplier, date or store.
With a system like this in place you then have the ability to integrate machine learning into the process. You can collate information from multiple locations and identify any trends — like whether returns were associated with a specific supplier, delivery route, store or customer.
From one small consideration in terms of how data is collected, we’d be able to identify a range of issues, having a very real impact on both the bottom line of the business and customer satisfaction.
Internationally there are also many examples of machine learning being used to yield quantifiable impact.
In a similar retail world, for example, Alibaba uses machine learning and AI to generate the most efficient delivery routes to get products from its distribution centres to its customers.
By tracking information around the routes drivers take and the time it takes to get there, the company can use that data to optimise future trips. As a result, the e-tailer claims to have reduced vehicle use by 10% and overall travel distances by 30%.
This has efficiency, cost-saving and sustainability benefits, all starting with the smart collection and use of data.
But do people trust them?
Public attitudes towards algorithmic decision-making are, unfortunately, not always positive.
There has been much negative media coverage around their potential for bias and reputation for being opaque, “weapons of math destruction”.
In studies conducted by the Pew Research Centre there was an overall scepticism around the use of these technologies, with general concerns being around privacy, fairness and effectiveness.
Unsurprisingly, the general attitudes were more positive among younger respondents and less so in older demographics, suggesting that digital natives are already used to this kind of digital intervention as a part of everyday life.
Harvard Business Review offers an interesting counterpoint to the common concerns around bias in machine learning, however.
Algorithms may display some form of bias depending who how they are developed, but the truth is that they are on average less biased than the human decision-makers they replace.
This was the case in a number of cases studied, from choosing “non-traditional” candidates in hiring scenarios to making more equitable recommendations on bail decisions for individuals awaiting trial.
The reality is that humans are actually remarkably bad at making decisions.
When you add to this the unpredictability of whether or not decision-makers themselves hold hidden biases, it is clear that a properly developed algorithm can beat humans when it comes to both ethics and effectiveness.
So where to from here?
So with all this in mind, we can see how a technology like machine learning, which can be seen as both opaque and theoretical, is actually neither.
When applied intentionally, and as part of a broader strategic approach, it can add significant value to all kinds of businesses, whether the output is physical or intangible.
The key takeaway, it’s not a one-size fits all solution when it comes to solving business challenges. It is simply one part of a toolkit that comes together to create something incredibly relevant and powerful, no matter the context.
Feature image: panumas nikhomkhai via Pexels