Machine learning not ready for the cloud, yet

Organizations with sophisticated machine learning systems usually have built them themselves, survey shows.

It stands to reason that many organizations interested in artificial intelligence and machine learning, which requires some sophisticated skills, will turn to cloud-based services to make it happen. However, that's not has happened yet.

clouds-over-chicago-cropped2-nov-2015-photo-by-joe-mckendrick.jpg
Photo: Joe McKendrick

Companies that are making significant headway with machine learning are ones that have invested heavily in developing or acquiring appropriate skills, such as data scientists and data engineers. So far, machine learning systems tend to be ones developed in-house, versus tapped into from the cloud or other outside sources.

That's the word from Ben Lorica and Paco Nathan, analysts at O'Reilly, which released a survey of 1,000 data specialists from across the globe. "Organizations that have more experience deploying machine learning models to production are more likely to use these newer job titles -- data scientist, data engineer,machine learning engineer, deep learning engineer," they observe. "About half of the respondents stated that machine learning models were built by their data science teams. However, that number rises considerably as organizations gain more experience."

Only 12% of those who belonged to organizations that are just beginning to explore machine learning stated that they relied on external consultants, whereas three out of four (73%) of those who belonged to the most sophisticated companies relied on their internal data science teams. Only three percent of respondents currently rely on AutoML services offered by cloud providers.

Will cloud-based machine learning open up possibilities to organizations with restricted budgets or skills? David Linthicum says there are now robust machine learning services available from cloud providers, including AWS, Google and Microsoft. "These systems are cheap to operate," he says. "You only have to pay a few dollars an hour, on average, to drive your very own machine-learning application such as the ones outlined above." Add to that cheap data storage and software developer kits and APIs. The only drawbacks are these services are bound to the cloud providers, and hybrid data environments may be complicated to implement.

So, we may see more machine learning services adopted from the cloud in the months and years to come.

Job titles specific to machine learning are already widely used at organizations with extensive machine learning experience, the O'Reilly survey shows: data scientist (81%), machine learning engineer (39%), deep learning engineer (20%).

Machine learning, of course, has it's share of bias, but most organizations are not keeping tabs on potential instances -- 40% report checking for fairness and bias. Even among the most sophisticated machine-learning companies, only about half monitor for bias -- 54%. Privacy in AI results is also only safeguarded across 40% of organizations. More than half (53%) of respondents who work for companies with extensive experience in machine learning check for privacy. The EU's GDPR mandates "privacy-by-design" mandates may push more enterprises to monitor and assure that privacy is protected within AI systems.