Click to learn more about author Deepak Dube.
While machine learning is an involved science with complex models, what distinguishes transparent machine learning is that it explains itself – how it works, its predictions, its insights – so that the user understands and trusts the outcome. In this article, I explain what transparent machine learning is and the considerations for implementing it.
What Is Transparent Machine Learning?
Enterprises have reams of data, which machine learning uses for two important purposes:
- Supervisory training of a statistical model to make predictions, which automates processes and creates consistency for the enterprise
- Extracting intelligence from the data
The data has critical intelligence buried in it about the dynamics of your business. This intelligence – some well-known, some surprising – has considerable value about your organization and the rules it follows.
The former purpose allows you to do what-if experiments to fine-tune and optimize your business; the latter exposes bottlenecks that, if undetected, could be costly.
With that in mind, let’s examine three ways transparent machine learning manifests:
- Explainable AI explains the reasons behind a machine learning model’s predictions. A reason is expressed as a rule of various predictors to make it actionable. In health care, for example, the model could predict that your systolic blood pressure today would be high, even though you took medication, because you were stressed at work, exercised for less than 20 minutes, and skipped your evening walk.
- Augmented intelligence extracts the intelligence – the insights about a business – from the entire data set, which helps improve decision-making. In education, for example, augmented intelligence might analyze a learning platform’s data dump and find that students on mobile platforms tend to have a poor experience compared to those on desktops in lengthy courses that have open-book exams.
- Traceability for data processing explains at each step, in context of your data, how it is processing it and why it chose that particular way, so that you can feel comfortable with the results or override them. When a platform is doing feature engineering, for instance, it might list the predictors that were important enough to qualify for modeling.
The key theme for transparency is trusted human-machine collaboration in the enterprise.
How to Get Started with Transparent Machine Learning
Businesses turn to digital transformation to automate, increase productivity and, importantly, ensure consistent output. Machine learning can play a key role in accomplishing those goals. Instead of using any machine learning platform, a key consideration should be how transparent it is. This way, experts in your enterprise can trust the predictive models and continue to build incrementally.
- Select the right machine learning platform. When selecting a platform, consider whether it can do explainable AI and augmented intelligence. This means that the machine learning platform must associate a confidence score with each prediction or insight; you select only those predictions or insights that have confidence scores above a threshold. Verify these results against the data, filtering it to carefully study whether the results were correct.
- Define the scope of the project. Propose a well-defined and narrow scope to trial transparent machine learning. Preferably, select a project that would benefit considerably from transparency. Ideally, the project should address an important function. To alleviate project risk, start with small incremental phases with well-defined deliverables that build on the previous phase.
- Get support from professional partners. Work with partners who have experience with the platform; they can share knowledge on how to interpret, generalize, and refine results. As you experiment with the platform, configure the various parameters and thresholds so that the platform performs based on your tolerance for error.
Other Considerations with Transparent Machine Learning
First and foremost, when selecting a vendor, ask them if their machine learning platform is transparent and easy to use. Here are a few additional considerations:
1. Training for users: Machine learning platforms typically have two interfaces: graphical user interface (GUI) and application programming interface (API). A user would choose the best option depending on their skill set and may desire training to get started. A business user looking for insights from data can use GUI to quickly extract it; a data scientist interested in advanced functions like explanations for predictions might use APIs to access it.
GUI helps you get started quickly without a considerable investment of time. It allows you to configure parameters. The results of explainable AI and augmented intelligence are displayed in simple and easy-to-understand tables and charts.
Suppose you want to stitch your own data pipeline. You have your favorite model for prediction – or a tool you like for visualization – and want to use the platform only to explain your model’s predictions transparently. You would have to use APIs to access transparency functions.
2. The ability to start with a smaller project: To use transparent machine learning, the project must have the infrastructure: big data platform, cloud presence, Data Governance, adequate training data and a model for prediction that is reasonably accurate. Otherwise, you can get started with what you have, then incrementally refine – but make sure to set expectations accordingly.
Preferably, select a project with a narrow scope. This can help alleviate project risk and help you learn the concepts by observing them. At the same time, select an important project where transparency eliminates a key bottleneck. Engage the vendor to help prune the list of potential projects.
3. Keep the management abreast with status updates: Chances are that this could be your first machine learning project – certainly your first machine learning project that demands transparency. Management should solidly support you because transparency implements vital business functions – such as regulatory compliance, model enhanceability, training data bias, prescriptive analytics, and data insights about business operations – but you need to communicate to make sure that you are achieving the desired goals.
Finally, transparent machine learning can be compute-intensive; make sure the machine learning vendor can help you size the compute resources. The platform vendor should remain engaged and help customize the default setting of parameters and troubleshoot issues. In the process, you may grow into a much-sought-after expert.