IBM Corp. is trying to mitigate the problem of bias in artificial intelligence-based decision-making with a new platform released early today that can inform people how AI models come to their conclusions.

The company said AI OpenScale is needed because a significant portion of businesses simply don’t trust AI enough when it comes to making the most important decisions. The main issue is that they’re unsure if their AI models are biased. They also want a greater understanding of how AI comes to its conclusions, and how it traces its logic, IBM said.

Bias is a byproduct of the black-box nature of AI applications, which don’t always reveal the logic behind their decisions. Bias can be introduced in training data as well as machine learning logic. Teasing out such aberrations can be an arduous data science exercise.

In an interview with SiliconANGLE Thursday, Jeff Welser, a vice president and lab director at IBM Research Almaden Research Center in San Jose, said bias can’t always be easily recognized. “If you have really large data sets, you might not even realize that the data are slightly biased on gender or whatever you’re analyzing,” Welser said.

The AI OpenScale platform, which is being made generally available today, is designed to remedy the bias problem through algorithms that are able to detect and correct it across the full spectrum of AI applications as they’re being run.

It works by continually monitoring AI applications and applying an “automated de-biasing technology” to flag biases that tend to build up in machine learning models. The software logs every prediction, model version and all training data in order to help organizations understand how AI applications reach decisions through plain-English explanations.

“So now we have systems to analyze data sets and see if they’re overrepresented in particular sets of characteristics,” Welser said. “It might be that you’ve overtrained on those characteristics.”

AI OpenScale also helps to make it easier to deploy AI models by enabling applications built with any open-source machine learning or deep learning model to run on any common environment. These include IBM’s own Watson and PowerAI platforms, the Seldon open-source framework, Amazon Web Services Inc.’s SageMaker and Microsoft Corp.’s AzureML, among others.

In addition, the platform comes with the Neural Network Synthesis Engine, which IBM said will allow businesses to rapidly and automatically build neural networks to run AI from scratch. NeuNetS, initially available in AI OpenScale as a beta feature, is designed for companies that don’t possess the engineering talent to build their own neural networks.

Image: geralt/Pixabay