Humans and computers work better together

“It’s easy to get an algorithm to 80 percent accuracy but near impossible to get an algorithm to 99 percent.” — Lukas Biewald

Source: Crowdflower

From Tesla’s autopilot to Facebook’s facial recognition, human-in-the-loop (HIL) approaches have outperformed. Startups can (should?) use this approach where appropriate.

We’re already seeing it.

Digital Genius* uses deep learning to power chat-based customer service. The company trains their models on historical chat logs. When a customer tweets a customer support question, their engine generates a response. If the algorithm is confident it’s the correct response, it will reply on its own. If not, it suggests it to a human service rep, who can approve or edit it. Either way, the model uses this data from the rep to increase its accuracy.

Magic combines human resourcefulness and artificial intelligence in its on-demand concierge offering. Magic learns faster than automated alternatives because humans are in the loop. Clara takes a similar approach. X.ai even hires AI interaction designers.

We’re at the beginning of a symbiotic relationship between humans and machines. I’m excited to see how this evolves.

*Bloomberg Beta portfolio company

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.