Science fiction continues to undermine our general understanding of artificial intelligence. Even well-educated people frequently project human characteristics onto systems, software and algorithms devoid of human feelings, ethics and morals. Machine learning (ML), for example, is often framed as machines learning and making decisions like humans do. That’s a dangerous misconception. People use a variety of methods and strategies to make decisions ranging from formal empirical analysis to rules of thumb (heuristics). ML systems rely on mathematical models and regression analysis to find relationships between a dependent variable such as the price of oil and independent variables like changing demands, production capacity, political stability and the health of the global economy.
Regression analysis methods range in complexity from simple linear regression to Polynomial, Bayesian and Random Forest regression, and Convolutional Neural Networks. ML systems use regression analysis algorithms that have been coded in software to analyze sets of relevant data. They can be designed to analyze and present what they’re ‘learning’ to decision-makers or make autonomous decisions without human involvement.
The power of ML systems lies in their capacity to find correlations in very large data sets involving hundreds, even thousands of variables. Analysts sometimes select the set of independent variables and create hypotheses regarding their influence on the target dependent variable. ML algorithms then test the hypothesis against selected data sets and use the results to predict future behaviors.
For example, If a stock trading company wants to find out how different economic performance variables correlate with changes in the stock market, then ML systems can be used to test thousands of combinations (hypothesis) in a matter of seconds. More importantly, they can keep processing new data to fine tune correlations and improve the accuracy of future predictions. This is how automated stock trading systems predict future market behaviors, then structure and automatically execute trades in a tiny fraction of the time required by their human counterparts.
Limitations
ML systems have many limitations, particularly when applied to complex social and economic systems. The most conspicuous and intractable limitation is that, however detailed, data is only an approximation of reality. Data can be noisy, i.e. contain false signals that overlap valid ones, as when a hiss (white noise) affects a telephone conversation. It can also be improperly selected and grouped for analysis leading to classification problems. Using the best analytical strategy and methods is critically important, particularly in cases involving time-dependent data streams. In other words, extracting value from ML systems involves much more than picking, analyzing and acting on data. It often requires a small team of experts to ensure that the right data sets, methods, and algorithms are selected to deliver relevant results.
Conclusion
Machine learning systems emerged from academic, research and development environments over the past ten years. They are indispensable in applications that require fast analysis of large, complex datasets and time sensitive decision-making. Unfortunately, there is a relatively small population of experts with deep understandings of ML systems and appreciation for the tradeoffs involved in automating decision-making. Forming and leading teams with the necessary expertise is a critical first step to creating, deploying and leveraging the power of these systems.
Artificial intelligence and machine learning are innovative technologies with the capacity to disrupt how companies operate and compete in the modern economy. Their impacts are already being felt across industry segments, from manufacturing to healthcare. The primary challenge for business leaders at this point is finding the technical, management and executive talent they need to exploit the power of these amazing technologies.