Human beings are known for a myriad of different things, but most importantly, they are known for getting better on a consistent basis. This tendency to improve, no matter the situation, has brought the world some huge milestones, with technology appearing as a major member of the stated group. The reason why we hold technology in such a high regard is, by and large, predicated upon its skill-set, which guided us towards a reality that nobody could have ever imagined otherwise. Nevertheless, if we look beyond the surface for a second, it will become clear how the whole runner was also very much inspired from the way we applied those skills across a real world environment. The latter component, in fact, did a lot to give the creation a spectrum-wide presence, and as a result, initiate a full-blown tech revolution. Of course, this revolution then went on to scale up the human experience through some outright unique avenues, but even after achieving a feat so notable, technology will somehow continue to bring forth the right goods. The same has turned more and more evident in recent times, and assuming one new discovery ends up with the desired impact, it will only put that trend on a higher pedestal moving forward.

The researching teams at Massachusetts Institute of Technology and IBM have successfully created a dedicated tool to help users better understand the saliency of their AI models. To fulfill the promised value proposition, these researchers developed saliency cards, which provide standardized documentation of how a method operates, including its strengths and weaknesses and explanations to help users interpret it correctly. Such an extensive lowdown, on its part, should guide the user big time in selecting an appropriate saliency method for both the type of machine-learning model they are using and the task that model is performing. But how does the stated tool manages to inform the user on a more granular level? Well, it does so by setting the table to conduct a concurrent comparison of different methods and pick a technique most fitting for the task at hand. Interestingly enough, this information is extended under the umbrella of 10 different attributes In order to ensure that subsequent results are as efficient as possible, the said attributes capture the way saliency is calculated, the relationship between the saliency method and the model, and how a user perceives its outputs.

“Saliency cards are designed to give a quick, glanceable summary of a saliency method and also break it down into the most critical, human-centric attributes. They are really designed for everyone, from machine-learning researchers to lay users who are trying to understand which method to use and choose one for the first time,” said Angie Boggust, a graduate student in electrical engineering and computer science at MIT and member of the Visualization Group of the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL).

Following the creation of their tool, the two involved teams conducted a user study with eight domain experts, from computer scientists to even a radiologist. Going by the published results, all participants reported the concise descriptions helped them prioritize attributes and compare methods. Not just that, even the experts from sectors that aren’t too heavy on AI and machine learning exposure claimed they felt comfortable in using the all-new saliency cards.

For future, the plan is to uncover some of the more under-evaluated attributes, while simultaneously design task-specific saliency methods. Beyond that, the researchers are also hoping to better understand how people perceive saliency method outputs, something that could viably instigate high-quality visualizations.

“We are really hopeful that these will be living documents that grow as new saliency methods and evaluations are developed. In the end, this is really just the start of a larger conversation around what the attributes of a saliency method are and how those play into different tasks,” said Boggust.