IBM ups the game for artificial intelligence
A new platform promises to help identify problems with facial recognition and other issues
One of artificial intelligence’s known weaknesses is bias. Now a new platform by IBM aims to give businesses the tool to detect that bias — and fix it.
IBM on Monday announced AI OpenScale, a new artificial-intelligence platform that among other things is supposed to clear up how AI makes decisions.
“How do we help organizations have more trust and transparency in AI?” said Ritika Gunnar, vice president of IBM’s Watson AI technology, in an interview Friday.
She gave this example: Insurance companies are using AI to help them accept or reject claims. How the technology is making those decisions is important.
“Bias can occur because the model itself is biased, or data is biased,” Gunnar said. “OpenScale detects bias as it is actually happening.” Then the system can give recommendations, or provide a synthesized model to de-bias automatically, she said.
Other examples of biased AI abound. Reuters
reported last week that Amazon shut down a secret automated recruiting tool because its AI was showing bias against women.
“They trained it with all the data they had,” said Paul Teich, principal analyst with Doublehorn, on Monday. He pointed out that that data was probably affected by the low numbers of minorities at Amazon and other tech companies. “They could’ve used a tool like (IBM’s OpenScale) before they deployed it.”
Artificial intelligence is also used in facial recognition, which a study found is most accurate for white men, and least accurate for darker-skinned people, especially women. That’s important because law enforcement and other officials are increasingly using
facial recognition.
“It’s also crucial to have diverse groups of people involved in AI development in order to limit potential bias,” said Charles King, Pund-IT analyst, on Monday.
In addition, IBM’s OpenScale aims to address the lack of available AI expertise among businesses.
A recent study of 5,000 C-level executives by IBM showed that 82 percent of businesses want to use artificial intelligence, but 63 percent don’t have in-house AI talent, the company said in a press release Monday. Also, 60 percent of those businesses worry about trust and compliance issues.
“This principle of having visibility and transparency into how AI works, we have been doing this for quite some time at IBM, in many different industries,” Gunnar said.
King agrees that Big Blue’s experience, size and scale should help.
“One of the points of the OpenScale design is that by providing a single, easyto-use platform, AI planning and implementation processes can be entrusted to large, more inclusive groups,” King said.