IBM’s Sriram Raghavan on what's being done to remove biases from AI engines

Multinational technology giant IBM has released a free software service for enterprises that will help data scientists eradicate biases from their artificial intelligence (AI) engines and get insights about decision-making processes, a top executive told TechCircle.

“The new software service that runs on the IBM Cloud helps our customers understand exactly why the AI engine is coming out with a particular recommendation or result,” said Sriram Raghavan, vice president of IBM Research for India and chief technology officer of IBM India and South Asia. “This was not generally available before.”

He said that customer feedback had prompted the firm to release this service, which IBM’s India research lab helped put together.

"A lot of our customers including those from India are aggressively trying to deploy AI for revenue generation. But they do understand the need for unbiased and insightful AI systems which will be critical for their operations," Raghavan said.

Using the example of the process to approve bank loans, Raghavan explained that when an AI engine rejects or approves a loan decision, the banker will see via a visual dashboard exactly why the AI did so and in turn get insights that may result in operational changes.

AI engines typically work with large datasets to draw conclusions. IBM says many AI systems continue to be trained using bad data that can contain implicit racial, gender, or ideological biases. 

“A bias in an AI engine could result in it either recommending a particular result or rejecting one based on the data or parameters it has been set on,” Raghavan said. 

IBM’s new service runs a real-time check on the data lake and the AI engine at the same time to understand why and when the AI engine is turning biased.

“In case the bank is OK with the bias, they can go ahead with the engine’s decision,” he said, adding that the company was also working on tools that would help enterprises to reset their AI engines.

Currently, the service can recommend the addition of a particular data set to mitigate bias in the system.

Raghavan also said that the new software service will work with models built from a wide variety of machine learning frameworks and AI-build environments such as Watson, Tensorflow, SparkML, AWS SageMaker, and AzureML. 

“This means organisations can take advantage of these new controls for most of the popular AI frameworks used by enterprises,” Raghavan said. 

The service can also be programmed to monitor the unique decision factors of any business workflow. This means organisations can customise it for specific use cases.

Further, IBM is making an open-source toolkit available for bias mitigation available so that scientists, academia and engineers can develop more solutions.

Raghavan said that IBM is also starting a new consulting service to help companies design business processes and human-AI interfaces to further minimise the impact of bias in decision making.  

According to an IBM research, while 82% of enterprises are considering AI deployments, 60% fear liability issues and 63% lack the in-house talent to confidently manage the technology.