With 180,000 real estate agents who are more familiar with building codes than computer code, Keller Williams Realty was looking for a way to compete with tech-focused startups like Redfin and Zillow.
Luckily for the 35-year-old realty company, it didn’t need to teach every agent a bunch of new computer skills. Instead, it found the answer in a homegrown app that automatically learns how to recognize the characteristics of a home or property from video footage shot by visiting agents. The app then turns that info into a searchable, automatically generated listing page.
The backbone of this system is a tool called Google Cloud AutoML, which is aimed at helping a wide range of industries automate machine learning models like the ones used for Keller Williams’ Keller Cloud.
“As KW agents increasingly use Keller Cloud apps, the AI platform will learn from their interactions and subsequently offer intelligent decision support and recommendations to strengthen agent service to consumers,” said Keller Williams chief product officer Neil Dholakia.
The undertaking is an example of how AI tools that automate the very process of building AI are making machine learning operations more accessible to companies that don’t necessarily have big armies of data scientists on hand. Amazon, Microsoft and Salesforce have also released tools similar to Google’s in a bid to own the cloud computing market, and a slew of smaller startups like DataRobot, H2O.ai and RapidMiner have already established their own niches in the space as well.
“It’s a little bit like power tools,” said Forrester AI analyst Kjell Carlsson. “With power tools, everybody might not become a carpenter, but everybody has the ability to do carpentry in a way that it was just very difficult to do before.”
The idea of automated machine learning development, or AutoML as it’s commonly abbreviated, can be confusing because most machine learning, by definition, consists of programs that automatically form their own programs through self-training. But AutoML usually refers specifically to technology that automates certain setup processes—namely, selecting a particular framework and starting parameters to best fit a given data set—and tasks that would otherwise fall to data scientists.
“If you want to create, say, a model for lead scoring, you can go and take [one of the many AutoML tools on the market] and you can literally just take the data that you have on the particular websites … and create a conversion model,” said Carlsson.
Such programs have existed in various forms among researchers and developers for years, but it wasn’t until about the last year and a half that Big Tech companies began to latch onto the idea as an enterprise product, according to Rachel Thomas, co-founder of AI educational publisher Fast.ai. It’s usually geared toward specialized tasks with unique data types that might not fit into existing prefab models, a common scenario for businesses with enterprise needs.
Use cases for AutoML range from building purchase and inventory-prediction systems to sorting ecommerce product photos or other content-recognition and tagging functions.
Meredith, the publisher behind magazine titles like People and Entertainment Weekly, used Google AutoML to scan and categorize thousands of articles and recipes. Hotels.com has been using similar automated features in Amazon SageMaker to develop recommendation and personalization tools with a lower overhead. And PayPal has used H2O.ai to create fraud-detection models.
Yet despite tools like these (and the seeming ubiquity of machine learning buzzwords in industry conference lineups and trade media), many businesses still remain overwhelmed by the prospect of AI adoption. A recent survey from International Data Corp. found that only a quarter of businesses have laid out an enterprise-wide AI strategy. A growing number of executives see AI-enabled competition as a risk to their own operations: 45% of respondents this year versus 37% in 2017.
The hope is that automation will reduce that hesitance. “These technologies that are coming out, whether they’re from us or from others. We just want to do what we can to remove the barriers for building with AI,” said Mitra Azizirad, Microsoft’s corporate vp of AI marketing and productization.
In some cases, however, worries of certain risks might be warranted. The “democratization” of AI can also mean it is subject to fewer of the safeguards that a more robust manual operation would normally be, according to PwC global emerging tech leader Scott Likens.
“Needless to say, these are powerful tools,” Likens said. “[But they] may lead to unintended consequences, biased models or algorithms that have little predictive capability. Those who are not trained in data science may also not recognize issues latent in the data itself, from the presence of bias all the way to outright violation of data privacy regulations.”
Or as Carlsson put it in keeping with the power tool analogy: “It also doesn’t stop you from cutting off your finger.”