Turning data into insight is one of the top business challenges today, and it becomes especially tricky when the data in question is unstructured. Artificial intelligence has a mixed track record there, but a young startup aims to get better results by bringing humans back into the picture.
Spare5 on Wednesday released a new platform that applies a combination of human insight and machine learning to help companies make sense of unstructured data, including images, video, social media content, and text messages. The result, it says, are "game-changing insights delivered cost-effectively and at scale."
The company's technology is now being used by companies including Expedia and Getty Images to enrich, clean and label unstructured data.
“Businesses need specialized human insights to solve complex data problems,” said Matt Bencke, founder and CEO of Spare5. "There is a profound difference when the right human intelligence powers machine learning."
Spare5 taps human expertise via a crowdsourcing platform. The company can call on a global community of more than 40,000 subject-matter specialists to accomplish custom micro-tasks through an app on their smartphone or desktop. Dubbed "Fives," these people can get paid to do things like rate help articles, rank photos, write image titles and descriptions, or find missing information.
The process begins when a company signs on with Spare5 and uploads a set of unstructured or incomplete data. Spare5 then turns the job into micro-tasks, writes custom instructions, and establishes a "gold standard" for answers.
To get the job done, Spare5 targets specific groups of Fives for specific tasks -- women aged between 30 and 40 who frequently shop online, for example.
Next, machine-learning algorithms filter the results for accuracy and ensure that quality is maintained.
The final result is that the customer's previously unstructured data is put into a structured format and delivered with associated insights, including weights or confidence intervals. That structured data, in turn, can be used to train new AI algorithms, complete data sets or improve recommendation engines, for example.
Sentient Technologies, for instance, makes an AI-powered shopping assistant and uses Spare5 to validate its AI-generated models with data about how people perceive nuances among retail products.
"The dirty secret of machine learning is that it is great at recognizing connections and correlations, but can’t always interpret causality, understand context or correctly identify similar pieces of information," said Nik Rouda, senior analyst with Enterprise Strategy Group. "This is where humans still have an edge."
Unstructured data can be particularly nuanced, Rouda added, as illustrated by several recent memes featuring near-identical photos of chihuahuas and muffins, for example. Even Google's technology has stumbled in a big way, mistakenly tagging black people as gorillas.
"Machine learning may be able to read everything and find obscure patterns or rare conditions humans would miss, but it also can lack valuable human experience," Rouda said.
By blending the two, Spare5 could tap into the best of both approaches.
"Improving accuracy and completeness of data going into machine learning will improve the model, and it will continue to learn the right associations," Rouda said.
Founded in 2014, Spare5 took in $10 million in Series A funding last August.