When Lars Hård, an artificial intelligence veteran, discusses developing a recommendation engine for a consumer product like perfume, he talks about harvesting many kinds of data. Product data. Consumer search engine data. Design data having to do with colors that men or women prefer. Pricing data.
To Hård, founder and chief technology officer at Expertmaker, data in all its structured and unstructured forms represents signals that machine learning systems pick up to refine results for future interactions. The data, and those interactions, also figure into Hård’s work of applying principles of evolutionary biology to both business use cases, like shopping assistants, and medical research.
Expertmaker is a company based in Malmo, Sweden, that makes recommendation systems and virtual assistants with an emphasis on mobile devices. You can find the company’s technology in applications like the eBay Explorer shopping assistant.
On this episode of the Data Informed podcast, Hård discusses how his work seeks to build on the heritage of the famous “if you like Book A, you might enjoy Book B” screen at Amazon.com. He describes the process of working with very large datasets from structured and unstructured sources to develop what he calls an optimized map of data correlations that will lead to end-user action.
“There might be thousands of levers and knobs you need to put in the right position to reach something that is profoundly good,” Hård says. “So we try to automate as many processes as possible using as many algorithms as possible that have the capacity to approach these optimization problems.”
He also discusses how he has adopted the principles of evolutionary biology to tackle other big problems. For example, he has established another company, called Experlytics, which seeks to apply artificial intelligence models to medical research, to pinpoint promising correlations and directions for researchers to explore.
This podcast was recorded in a lobby area of the Predictive Analytics World conference in Boston on September 30.