The Balance: Smart Software and Humanoid

February 13, 2019

Micro-blogging super sphere, Tumblr, recently made headlines when it declared it would crack down on adult content. This is obviously not a small task, so the company harnessed as much deep learning AI as it could. The problems started instantly: pictures of boots and jeans were getting flagged the same as pornography. We got the full picture from a recent The Next Web story, “The Challenges of Moderating Content with Deep Learning.”

According to the story:

“Obviously, the folks at Tumblr realize that there are distinct limits to the capabilities of deep learning, which is why they’re keeping humans in the loop. Now, the question is, why does a technology that is as good as—or even better than—humans at recognizing images and objects need the help to make a decision that any human could do without much effort?”

This is a similar problem that the defense and intelligence communities are struggling through. It’s funny when boots get mislabeled as pornography, but what if your deep learning software mislabels satellite images regarding hostile threats? UCLA researchers recently declared they found the limitations of deep learning and we are far closer than most folks guessed. The solution is good news for us humans, because living-breathing oversight is really still the only solution. Meaning deep learning AI is merely one tool out of many we will continue to use.

Patrick Roland, February 13, 2019


Comments are closed.

  • Archives

  • Recent Posts

  • Meta