About This Project

I started working on this project when I was learning about Organizational Behavior (OB). Organizational Behavior is the academic study of how people interact within groups. Concepts from OB are primarily applied by companies and businesses to try and optimize their operations. Historically, some companies have taken a lot of unethical actions that were legal towards their employees to help themselves operate effectively. An example of this is companies forcing early retirement and resignations from employees to cut down on costs like severance pay or termination pay. Thankfully, in some countries actions like that are now deemed illegal and there are laws that exist that protect employees. Optimization is right at the core of machine learning and machine learning is becoming a substantial part of analyzing business operations. This got me wondering how can companies use their employee data to optimize their operations. This project tackles the implications of integrating AI into organizations; how can it be used and are there ethical concerns to doing so.

What I did

I am not an expert on machine learning, business analytics, or data privacy laws. I just did some trivial experimentation that barely scratches the surface of a vastly open ended question.

I decided to place myself in the position of an organization that wants to run more effectively. I found a dataset that represents employee turnover. This dataset contains personal information about each employee and identifies whether the employee has resigned. I fed the processed data to a machine learning model and then tried to come up with ways the information from the model can be interpreted to optimize business decisions.

I trained a logistic classifier using scikit-learn that predicts the turnover status of employees. The model takes in an employee's data as an input and outputs a turnover prediction. The classifier has internal weightings on each category of data.

The image above shows the weightings placed by the model on each category of personal data. The red bars represent factors inducing turnover and the blue bars are the opposite.

This model could definitely be used for a good cause. If you are an organization that wants to address its turnover issues then you can use the model weightings to find possible factors that induce turnover. The model suggests that the most prominent factor behind employees leaving is a promotion issue. As an organization, you might want to put some focus on planning how to integrate more promotions to retain your employees.

The model can also have unethical applications to company policy. As mentioned before, force termination is something that companies used to unethically do to cut down on business costs, which is now illegal in a lot of places. Companies can use a model like this to delibrately induce turnover in a subtle manner that is difficult to expose. Let's assume that you're an employee who is close to retirement and the company wants to avoid severance pay by making you resign. The company can use the results from this dashboard, and try to modify the features surrounding your work conditions until the machine learning model predicts that you are highly likely to leave by yourself then start implementing those modifications in real life.

Another big issue to consider is bias. Most machine learning models are trained on data from the real world. This means that bias in the data usually translates to bias in the model. For example, if a company has a history of discriminating against a certain group of people, then the model is gonna have some form of bias towards that group. A company can use that information to attempt to identify the issues that cause this discrimination and fix it. On the other hand, if a company decides to employ this model to help with making hiring decisions without inspecting potential bias, then the model will most certainly discriminate.

Takeaways:

The project is a very small example the demonstrates what AI integration in business decisions might look like. There are useful applications but there are also lots of unethical ways this technology can be used. It will be up to governing bodies around the world to create new laws and safegaurds to regulate the use of AI in our lives. It is a very delicate problem though. How do you regulate the use such that we're able to extract all the benefits but avoid the potential harm?