At a glance: Some of the tasks people help AI systems do

This undated combo of images provided by CrowdFlower shows before and after renderings from CrowdFlower’s Human-in-the-Loop technology, which provides sophisticated tools that enable a person to label and structure every part of a normal photo and convert it into structured "training data” that an AI system can understand and interpret.

Mon, 05 Mar 2018 10:44:01 -0500

People power drives artificial intelligence in a surprising number of ways and for a wide variety of purposes. Here are just a few instances in which people play vital roles in helping machines learn things about the real world:


Self-driving cars

Users: Automakers such as Toyota, Nissan and Ford; ride-hailing firms like Uber and Waymo.

Tasks: Look at dashboard-camera photos and draw boxes around objects like cars; trace photo elements such as walls, light poles, bridges and lane markers; label every pixel with digital paint marking it as a street, car, stop sign, pedestrian or cyclist.


Facial recognition

Users: App makers, researchers

Tasks: Put dots where requested, including on the corners of eyes and mouths in a photo.


Audio transcription

Users: Digital assistants, call centers

Tasks: Transcribe what is said, from a single word to phrases.


Sentiment analysis

Users: Call centers, social media services

Tasks: Read text to determine if the words are abusive or threatening or if the writer is angry or frustrated, so that a manager can be alerted.


Image labeling

Users: Online marketplaces like eBay

Tasks: Categorize items such as shirts and dresses, determining their style, color and fit.

Copyright 2018 The Associated Press. All rights reserved. This material may not be published, broadcast, rewritten or redistributed.