Are you using the right tool for the job?

What does "?#artificialintelligence" do?

Are you using the right tool for the job?

Post No. 11

I am sure that like me, you found yourself facing a Phillips screw with only a standard screwdriver in your hand, yet a bit far from the toolbox. You say to yourself "I can do this." Well, I know from experience that "...results might vary."

So, the very same principle applies, and with much more important and critical consequences, when dealing with business and information tools. This is very crucial in the context of Life Sciences and Healthcare, where lives and well being are at stake. It is similarly important in other areas.

As we continue looking for new ways to leverage AI and ML in business, including their application throughout life sciences and healthcare, we must always care to use the right tool for the job.

Lets examine the article about AI Algorithm Predicts Future Crimes One Week in Advance With 90% Accuracy - original citation - Nature of Human Behavior journal - Event-level prediction of urban crime reveals a signature of enforcement bias in US cities. The premise of the article speaks to exactly the concept of using the right tools. The authors make this point several times. However, it is so easy to take this out of context and use the much more eye catching title about predicting future crimes. And herein lies the challenge. In the wrong hands, and with improper intentions, AI may be used to predict future crimes, even with a 90% accuracy, for direct law enforcement, while the intent of the original research was for it to serve as a tool in the?toolbox of urban policies and policing strategies to address crime.

This begs the broader question If AI Is Predicting Your Future, Are You Still Free? This is where I would like to introduce the Concept or Model Drift.

Every AI and ML implementation must be maintained and cared for over time. You cannot simply design and build the AI solution and then walk away assuming it will run well on its own. These systems have a tendency to suffer from what is termed concept or model drift - the underlying data tends to shift over time. Both model and training data tend to shift, impacting the overall model, and affecting the ability to maintain consistent classification. If not watched and maintained properly, your model may shift and provide skewed classification or misclassification with unintended results.

This model drift is what I believe, what eventually limits AI predictions from reaching full accuracy. However, if this tool is used incorrectly, predictions, even if not accurate, may have an impact on future outcomes, and become self fulfilling prophecies. For example, we recently see growing use of algorithms to determine a future employee's tendency to be a good hire. Eventually, the individuals that are flagged, for whatever reason, will see less exposure and less interviews, and end up not getting hired, just as the model predicted.

So, we must be very careful when choosing AI and ML tools to make sure we pick the right tools and this can only be achieved by strong collaboration with the business and subject matter experts!