The lack of an understanding of statistics has always been a problem among people who need to base their decisions on numbers. This problem already arises in the absence of very basic statistical knowledge. How many managers report an average figure in a presentation without knowing that they should be looking at the median instead?
In the age of data abundance and technological marvels, it’s tempting to believe that statistics and machine learning algorithms hold the power to provide definitive answers. This belief can sometimes lead us down a treacherous path, blurring the lines between true understanding and a false sense of confidence. New AI tools accounting software, e-commerce suits, or in cloud services provide us predictive statistics for future sales or tell us which products are most important to our customers. Without fundamental knowledge of statistics and sector knowledge it is easy to become misguided by these statistics.
Machine learning does not render human judgment obsolete. Instead, it necessitates experts capable of connecting machines with people. To truly derive value from data, we require individuals who comprehend machine learning, the business context, and possess effective communication skills. What we truly need are socially skilled nerds.
Here are five common pitfalls when working with data that serve as good reasons for employing one:
- Being fooled by eloquent numbers:
In a world where numbers are presented with authority, it is easy to be misguided by the appearance of certainty. Machine learning algorithms chuck out predictions and insights that seem solid on the surface. Without a foundation in statistics, we might mistake these outputs for truth, failing to recognize the layers of uncertainty that often underlie them. This is particularly a problem when statistics seem large or meet our confirmation bias. - Not overseeing assumptions or logic behind statistics:
Every statistical and machine learning approach comes with a set of assumptions. These assumptions, if not understood or validated, can act as silent traps. A lack of background in statistics can cause us to overlook these critical assumptions, inadvertently sowing the seeds of misinterpretation. Statistics and machine learning are both sophisticated fields that rely on complex mathematical underpinnings. Without a proper grasp of these foundations, we risk oversimplifying or project human understanding on AI. This is especially the case for natural language models, where people often mistake the predictive capacity for words with human like understanding of subjects. - The seductive alure of spurious correlations:
Numbers can be deceptive, and without a solid statistical foundation, we risk falling into the trap of drawing conclusions from spurious correlations. Machine learning algorithms, when not guided by statistical reasoning, might pick up on coincidental relationships, leading us to believe in patterns that don’t truly exist. Who knew that US spending on science inflates the suicide statistics)? - Not understanding the real-world context of numbers:
Where the previous points where mostly concerned with statistical knowledge. Statistics are meaningless when we cannot relate them to the real world. This requires knowledge of the fields in which we are applying statistics and machine learning models. For example, without knowledge about banking, accounting and financial regulations, statistics about bank transactions are meaningless. - The expert trap
Even possessing an abundance of knowledge is insufficient if one cannot effectively communicate and prioritize information for their audience. If you wish to elevate an organisation with insights from data, you need data experts that can communicate the insights and take-aways to the people taking the decisions. This not only requires knowledge of statistics, machine learning and the subject you are analysing. It also requires a set of teaching and presentation skills. If experts only create reports that can be understood by other experts the potential impact of such insights remain rather limited.