IBM has introduced new artificial intelligence (AI) software to help businesses detect bias in decision making and explain how AI makes decisions.
In the Wednesday press release, IBM called it “a major step in breaking open the black box of AI and also announced that IBM Research will make an AI bias and detection ‘toolkit’ available in an open source community for others to take advantage of, calling it a way “to encourage global collaboration around addressing bias in AI.”
Michael Martin, senior executive, internet of things lead at IBM called this a big shift, saying this is “a democratization of these tools that allows [IBM] to put this type of technology into the hands of everyday users and everyday IT type operators.”
“This is not available anywhere else at the moment” – Michael Martin IBM senior executive
He told IT World Canada that unlike other AI bias detecting tools that check for bias solely in training data or during development, IBM’s new feature will use that type of data but also work in real-time.
Running in real-time means that the fully automated software is able to analyze data as it is happening at a business level, and businesses will be able to decipher certain practices and understand the reasoning behind certain outcomes produced by AI systems.
Real-life use cases
The bias detection works by analyzing the decision making processes for a company’s AI system, helps flag potentially unfair outcomes and can even be programmed to meet specific businesses needs, stated the release.
IBM has been working on this software for a while, said Martin and before it became generally available, it worked with some of its customers to test and get insights about the tool. Martin says IBM has been working with companies like KPMG, Cargills Bank in Sri Lanka, Woodside Energy in Australia and U.S.-based Autodesk Inc., he told IT World Canada that Autodesk, which makes software for a number of industries including engineering, manufacturing and entertainment has seen impressive results using Watson and the bias detection software.
Autodesk’s AI system responds to sixty thousand conversations a month, said Martin and using IBM’s services has increased customer response time by 99 per cent and also helps detect bias in those responses by analyzing whether the AI system is misunderstanding or incorrectly answering a question.
How it works
Once the information is collected by the bias detection software it can be viewed on a visual dashboard, which can show companies what factors an AI system weighs when making a decision, its confidence in the recommendation and even the factors behind that confidence. User can also view other statistics including the AI system’s accuracy, performance and fairness.
IBM is also offering a consulting service along with the software to help companies during set up and afterward and it’s not just for customers using IBM’s machine learning and AI platforms, said Martin, the consulting service will help implement the bias detection with whatever platforms a company is using.
Since its available in an open source community, the bias detection software can be adapted to work in other machine learning and AI frameworks that would typically be considered competitors. IBM announced that this “AI Fairness 360 toolkit” will include algorithms, code and tutorials for the wider tech community that could be made compatible with TensorFlow, Microsoft Azure machine learning, Spark machine learning, and Amazon’s SageMaker.
Martin told IT World Canada that this type of openness is important to IBM, “[we] believe in open source community and sharing. In the past companies might have made a tool that just worked on their products, but that limits your market opportunity.” He said being open makes IBM’s tool more readily available but it’s all about the customer as well. “Today if a company wants to survive they have to be very customer focused and customer-centric.”
Importance of AI
“There is a significant shift underway in how business leaders look at AI’s potential to drive business value and revenue growth,” according to an IBM report on the value of AI in 2018. It stated that 82 per cent of enterprises are now considering or moving ahead with AI adoption but 60 per cent fear liability issues and another 63 per cent simply lack the skills to harness AI’s potential.
“With this software, we’re taking the AI tools out of the black box. We’re taking them out of the dark closet that only the PhD data scientists can get out, and we’re exposing them for everybody,” said Martin, “It’s very critical that we democratize these tools and put them in the hands of the users so it can be meaningful, and can be used in real time, which is just critical.”
The bias detection software is currently available and works across IBM’s machine learning and AI tools, including Watson of course, and is accessible through the IBM cloud.