Crowdsourced labor and artificial intelligence company CrowdFlower Inc. announced today that it has raised $20 million in venture capital to expand its efforts further into the enterprise.

The round, led by Industry Ventures LLC, also included new investor Salesforce Ventures alongside existing investors Canvas Ventures, Microsoft Ventures and Trinity Ventures.

People power and machine learning go hand in hand at San Francisco-based CrowdFlower, which uses data and training from large groups of people – a practice known as crowdsourcing – to train machine learning algorithms to do tedious data science work.

Robin Bordoli, chief executive at CrowdFlower, believes that AI applications within the enterprise is on the verge of a “Cambrian explosion.” This is a reference to a point in the biological history of life on Earth when a huge variety of different body designs begin to appear in the fossil record. In short, “living thing” applications began to try out a lot of different ideas.

With CrowdFlower’s approach to crowdsourcing AI training, Bordoli said, enterprise data science could find its killer app. “The bottleneck for the large-scale adoption of machine learning still remains the availability of high-quality training data and human-in-the-loop workflows to handle the failure states,” he said.

The crowdsourced labor works by applying simple tasks to individual workers, such as transcribing text seen in an image, determining the sentiment of a sentence, statement or forum post, annotating images and other work that humans do well. These are processes that can be broken down into thousands of small tasks, and each individual task is executed by a small group of people.

For example, six people could all read a short statement by a customer and pick a “sentiment,” such as “happy,” “satisfied,” “unhappy” or “angry,” from a list of options. The answers are then aggregated.

Thousands of statements receive this treatment so that there’s lots of sentiment data attached to a variety of responses, which can be from reviews, social media, customer complaints or other sources. The statements and attributed sentiments then get fed into the AI big-data engine, which then learns how to rapidly identify the sentiment of new yet-unseen customer statements.

AI algorithms continue to become more sophisticated and context-aware day by day, but many still have trouble with local colloquialisms, new slang, sarcasm and Internet memes that human readers could quickly identify. As a result, human workers augmenting the AI system give it a quicker and more accurate understanding than if it were left on its own.

“A machine learning model without training data and human-in-the-loop workflows is like a rocket ship with a large engine but no fuel and no navigation system,” Bordoli said. “It won’t reach escape velocity nor will it achieve the trajectory to land on its intended target.”

CrowdFlower’s human crowdsourced AI training platform can be applied to a multitude of use cases, including self-driving cars, intelligent personal assistants, medical image labeling, content categorization, customer support ticket classification, social data insight, data enrichment for customer relationship management, product categorization and search relevance.

Over the years, CrowdFlower has raised a total of $58 million, including today’s funding, in various rounds starting with $1.2 million seed funding in 2009 and more recently $10 million in Series D funding in June 2016.

The company intends to use the new money to extend its platform’s capabilities by expanding into other AI and machine learning technologies. This will start with growing its operations by hiring new data scientists, machine learning experts and engineers, as well as sales and marketing team members.

Image: Pixabay