How do we get AI out of the early adopter phase? It's about people

Commentary: Artificial intelligence is finally seeing heavy production use, but not as broadly as it will once the technology becomes more accessible.

istock-889289314azure.jpg

Image: metamorworks, Getty Images/iStockphoto

The good news? According to a new O'Reilly survey, the more experience companies have with artificial intelligence (AI) in production, the less pushback they're getting from naysayers. The bad news? Many continue to struggle to figure out where to use AI, whether they're experienced with AI or just kicking the tires. Worse news? It's still hard to find competent talent to help unlock this conundrum, but this may be more an issue with inaccessible AI technology than with employees.

Who are these people?

But first, it's worth looking at the current composition of the AI landscape. (For additional insight into the O'Reilly survey data, please check out "85% of organizations are using AI in deployed applications.") Among the 25 different verticals represented in the 1,388 survey respondents, the biggest category by far was Software, with 17% of respondents. Second largest? Finance (roughly 12%). No other industry broke 10%, with a range of verticals that seem like they should be doing more (Media/Entertainment, Logistics/Transportation) rounding out the bottom of the list. 

In other words, while there's plenty of adoption across industries, AI is still dominated by verticals that tend to be early adopters (Tech/Software and Finance). That's not a bad thing--it's just indicative of where we are in terms of mainstream adoption.

SEE: Special report: Managing AI and ML in the enterprise (ZDNet) | Download the free PDF version (TechRepublic)

This also plays out in terms of where within these organizations AI is making a dent. It's still mostly an R&D thing (close to half of respondents), with another third coming from IT. 

That said, it's telling that the ratios of evaluation to "mature" adoption have flip-flopped in the past year, according to O'Reilly. In 2019, 54% of respondents were evaluating AI, and a much smaller percentage (27%) had reached mature adoption, meaning they were using AI in analysis and production. But this year, over half of those surveyed have jumped to mature adoption, with a third in evaluation. A mere 15% say they're not doing anything at all with AI.

So things are definitely moving. But when things get bogged down, what's to blame?

Speed bumps on the road to AI

While finding talent used to be the biggest challenge to effective AI adoption, that's now the third-most cited concern:

4-bottlenecks.png

Image: O'Reilly

The data becomes more interesting when divided up into those companies in the "mature" stage of adoption and those in the "evaluation" stage:

5-bottlenecks-with-maturity.png

Image: O'Reilly

For those companies that are still kicking the tires on AI, it's natural that cultural antibodies would fight against it. It's new and (as yet) unproven. As it gets implemented in production, however, people start to see the value and the antibodies dissipate, as can be seen in the vastly diminished cultural headwinds for those companies that have reached the mature phase.

What seems strange, however, is that both sets struggle to "identify appropriate business use cases." This doesn't seem to improve much once a company gets past evaluation into mature production. Why? 

It's hard to glean too much from the survey data, but I wonder if it has anything to do with where AI is being used. As noted briefly above, R&D (48% of respondents cite this) and IT (33%) are the two top consumers of AI, while the areas of the company best positioned to know where AI might benefit them (e.g., Marketing (21%), Manufacturing (13%), Sales (12%), Logistics (11%), etc. see far less adoption. IT and R&D are likely running AI projects for some of these groups, but adoption may remain blocked by AI's inaccessibility to less technically proficient areas of the company. (This may correlate with a seeming mismatch between enterprise interest in digital transformation and its lack of investments in re-skilling or up-skilling of people.) 

SEE: Managing AI and ML in the enterprise 2020: Tech leaders increase project development and implementation (TechRepublic Premium)  

Years ago Gartner analyst Svetlana Sicular called this out: "Organizations already have people who know their own data better than mystical data scientists....Learning Hadoop is easier than learning the company's business." Since she wrote this, the tools for data science may have changed (perhaps less Apache Hadoop and more TensorFlow), but the need to democratize access to employees who understand the business has not. 

In short, while AI has made great strides within the enterprise, more work is needed to expand its accessibility to a broader swath of the enterprise. 

Disclosure: I work for AWS, but nothing herein relates to my work there.

Also see