Cognitive computing is smashing our conception of ‘ground truth’

NEW YORK — When deep learning startup AlchemyAPI exposed its natural language processing system to the Internet, it determined that dogs are people because of the way folks talked about their pets.

That might ring true to some dog owners, but it’s not accurate in a broader context. That hilarious determination reflects the challenges — and opportunities — inherent to machine learning.

[aditude-amp id="flyingcarpet" targeting='{"env":"staging","page_type":"article","post_id":1178996,"post_type":"story","post_chan":"none","tags":null,"ai":false,"category":"none","all_categories":"big-data,","session":"C"}']

“The whole idea of a ground truth, which used to be a big thing in computing, is really starting to go away,” said AlchemyAPI CEO Elliot Turner in conversation with GigaOM writer Stacey Higginbotham at the Structure Data conference on Thursday. “Vampires are real in a certain context, but not in others.”

IBM’s artificially intelligent computer Watson also grapples with contextual scenarios, said IBM’s Stephen Gold on-stage at Structure Data. We think of computer logic in absolute terms, he said; in our minds, it should figure out that two plus two is four. But to Watson, that could also represent a car configuration — two front seats and two back seats — or a familial unit with two parents and two children.

“There is no absolute outcome,” said Gold, who serves as the vice president of sales and marketing for the Watson Group. “These systems grow in confidence as information is presented — and they learn.”

Many of these artificial intelligence frameworks date back to the early ’90s, but organizations shelved a lot of neural network techniques because computers were just too slow. But they’re progressing rapidly: Watson is 90 percent smaller and 24 times faster than its initial form factor in 2006.

That leads Gold to believe that we’ll see freemium models for cognitive computing emerge in the near future.

“Although the issue of cost is very real,” he said, “there will be people who step up and say, ‘I will pay for it because of the value it brings.'”

But these technologies still need humans to direct them, said Turner; they will augment human capabilities, not replace them. Just because Watson can now understand X-rays doesn’t mean that doctors will become obsolete.

“I believe in the strength of the human spirit,” he said. “While the systems that are coming online are amazing… you can still have a person read a document better than a machine can today. Same thing for vision. We just have to focus on the things that make us special, and move away from the historical view of rote memorization.”

[aditude-amp id="medium1" targeting='{"env":"staging","page_type":"article","post_id":1178996,"post_type":"story","post_chan":"none","tags":null,"ai":false,"category":"none","all_categories":"big-data,","session":"C"}']

“I think the challenge and the skills is how we educate,” added Gold. “Understanding natural language and machine learning, understanding analytics and big data — we need to modify our educational system to get people who can truly build these new systems.”