One of the most common reasons for artificial intelligence products fail? Bad data

Shaibani Ahuja, the first vice president of the IT strategy, said during a round table discussion on luckBrainstorm Technology Conference in Park City, Utah.

But the agent, it turned out, was not the problem. Ahuja said: “What we have noticed is an essential problem in our data,” Ahuja said. When her team achieved what happened, they found that Salesforce published contradictory “knowledge articles” on its website.

Ahuja said: “It was not actually. It was the agent who helped us to determine a always existing problem.” “We turned it into an audited agent, he has already examined our content on our public website for abnormal cases. Once our basic data is cleaned, we indicated it again, and it was functional.”

New AI products will not be good as basic data, according to AHUJA and other speakers who participated in the discussion. Ashok Srivastava, Senior Vice President and Head of Artificial Intelligence in IntuitHe said he was not surprised by the conversation results Study of the Massachusetts Institute of Technology This has found that 95 % of artificial intelligence pilots in large companies have failed, due to old systems in large companies.

“The truth is that the basis of artificial intelligence – which is data – does not invest in people,” said Srivastava. “Therefore, you have data of the nineties of the last century sitting in a high -end database here, you have Amnesty International here, and the CEO may ask you to do something, and will not only work.”

Sean Browsch, Senior Vice President of Artificial Intelligence and Data in AmgenHe added that it is also difficult for major companies to move from the pilot to adoption at the level of the institution.

“Pilots in large companies do not provide an investment return,” he said. “They may provide learning, and they may provide proof points, and they may provide inspiration. But the road to size – this is where you get the return on any major technical program.”

In order to see companies returning to investing from the new artificial intelligence tools, they will have to sort through both data and the problem of scaling.