Outperforming the human: How can machine learning help your business?
29 March 2018
Every week, it seems someone produces a new estimate for how much data we produce. According to the research company IDC, by 2025 we will have created 163 zettabytes. That’s the equivalent of 40 trillion DVDs, or the entire Netflix catalogue 489 million times over. Technology allows us to gather, organise and analyse these vast volumes of data – but a potential weakness remains, and that stems from human involvement.
Whether we know it or not, humans have biases, where we have a tendency, inclination, or prejudice towards something or someone. An example of a well-recognised bias is confirmation bias. This involves, subconsciously or consciously, interpreting information in a way that supports our pre-existing beliefs rather than giving equal weighting to information that contradicts our hypothesis. We also have strategies derived from previous experiences, where we try to be efficient and take shortcuts to solve a problem (heuristics), which might mean we miss certain factors; for example our judgments of likelihood that the birth order of son-daughter-son-daughter is more representative of random outcome than son-son-son-son. These biases and heuristics can seep into the way we make decisions and how we use data, pushing the answers from the data into a direction that’s dictated by our preconceptions.
Machine learning (ML), while not eliminating this risk, can play a powerful role in either removing or highlighting biases and heuristics. Ask Google for a definition of machine learning and you get: “a type of artificial intelligence (AI) that provides computers with the ability to learn without being explicitly programmed. Machine learning focuses on the development of algorithms that can change when exposed to new data.” The key point here is the power for computers to learn ‘without being explicitly programmed’, this means we can minimise human bias in how data is interpreted, and let the computer tell us exactly what’s significant within our data. Critically, there’s still room for bias to creep into our models through the ML model learning from bias in the data itself if we’re not careful. We go to great lengths to try and remove bias from this underlying data prior to and during model building, and this is an area of growing research as models become increasingly complex and we lose transparency. PwC recently made a submission to the AI All Party Parliamentary Group and gave evidence covering this and broader considerations across artificial intelligence.
The role of the human in data analysis is changing. It’s no longer just our job to assess individual records, analyse a sample of a population or to help make a decision… it is now our job to responsibly direct, develop and harness the power of machine learning and AI in general. In parallel, we have to make sure that we have the right processes, frameworks and assurances in place to be able to interpret and react to their results.
Although this may sound complex, it can be broken down to a simple construct. What we are dealing with is algorithms that have the power to see connections and analyse data on a scale and at a speed that humans simply aren’t capable of. A human cannot scan the data of hundreds of thousands of customers every day, or hour or minute, to assess the ever-changing risk of each one based on their behaviour; a machine learning model can. Could a human identify the hundreds of complex, and ever changing, variables that predict the demand your organisation is likely to face over the next year? Very unlikely!
It is a "no-brainer" that today all organisations should at least explore the power of Machine Learning and AI within their business, but it is essential that at the same time careful consideration is given to the processes, frameworks and assurances in place around it.