Sign up for our newsletter and get the latest big data news and analysis.

Before We Argue About Ethics, Models Need Transparency

Pedro Alves, CEO of Ople,

In this special guest feature, Pedro Alves, CEO, and Rick Saletta, Senior Marketing Executive of Ople, discuss how model transparency is an essential capability. It facilitates the jobs of a data science team, while helping bring models’ end users to better comprehension of and comfort with the decisions driven by those models. Transparent AI is an absolute requirement for a proper AI implementation within an organization, as well as the key to ensuring a more meaningful ethical perspective as AI is woven into our everyday lives. Ople is a software startup which automates the model building process and accelerates deployment of production-quality AI models. The team at Ople aims to make the use of AI easy, affordable, and ubiquitous.

Rick Saletta, Senior Marketing Executive of Ople

When we talk about bias, ethics, and AI, we are getting ahead of ourselves. First and foremost, we need to talk about model transparency. We can convene all the advisory boards we want and chat with all the fanciful futurists. However, until we can consistently see how an AI model reaches decisions, we do not even have a way of framing the ethics conversation.

Model transparency is a table stake requirement for AI applications of all kinds. With transparency, we can shape AI models to reflect more humane values. Transparent AI is vital to AI research, currently grappling with a reproducibility challenge. In the business world, lesser or unregulated industries including retail, digital advertising, and manufacturing, companies demand or require transparent AI for ROI reasons.

Transparency is not a “nice to have” feature for the future. It’s here, and AI models are gaining the capacity to show how they reached certain results. These explanations are human-readable and can be remarkably straightforward. A team of computer scientists developed an agent that could explain why the AI model made a specific move in the video game, Frogger.

The stakes are high in law enforcement and other applications where AI-driven decisions must be transparent. True due process cannot exist where there is inherent bias.

Transparent AI can also have a profound impact the top line for businesses. When a company acts on insights gained from AI models, it wants as much insight as possible into the decision-making process. The more information a company has, the more targeted and effective the action can be. An online retailer might have a model that predicts if items in a customer’s cart will be purchased or dropped. Predicting that items will be dropped can trigger a response to incentivize the customer to make the purchase. This automated action will certainly be more effective than no action at all.

Now, imagine that the AI model could not only predict the likelihood of the items being dropped, but also explain why it is predicting the items will be dropped. And, what if the AI model could suggest actions that have incentivized a purchase in similar scenarios. Targeted actions are more personal and more effective at increasing sales. This level of insight can be achieved with advanced techniques of AI model explainability.

Model transparency will aid data scientists’ understanding of what information is crucial, and what additional information should be acquired to optimize the model. Transparency enables data scientists to use their time more effectively, highlighting problems such as data leaks and anomalous model behavior. Transparency also helps data scientists identify where data could have led to mischaracterizations, bias, and other unwanted outcomes in high-stakes applications.

Model transparency is an essential capability. It facilitates the jobs of a data science team, while helping bring models’ end users to better comprehension of and comfort with the decisions driven by those models. Transparent AI is an absolute requirement for a proper AI implementation within an organization, as well as the key to ensuring a more meaningful ethical perspective as AI is woven into our everyday lives.


Sign up for the free insideBIGDATA newsletter.

Leave a Comment

*

Resource Links: