The spread of artificial intelligence (AI) is not slowing down: 85% of organizations said they are evaluating or using AI in production, a report from the technology and business training company O’Reilly found. More than half of companies identified themselves as mature adopters of AI, or as using AI for analysis or in production.

O’Reilly’s AI Adoption in the Enterprise 2020 report, released on Wednesday, determined that AI growth and popularity is continuing apace. To prepare for this onset of AI use, organizations must make sure they have a solid foundation for the technology to flourish, it found.

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

The 2019 edition of O’Reilly’s report indicated that AI was still in the experimental phase. Last year, only 27% of organizations considered themselves in the mature adoption phase of AI. While 2019 hosted more AI testing, 2020 is presenting tangible action, the report found.

AI applications

Most AI is being used in research and development, which was cited by nearly half of respondents. Coming in second place was IT, followed by customer service, and marketing/advertising/PR, the report found.

Nearly all industries are present on the full list, indicating that AI is pervading all scopes of the enterprise, according to the report.

Challenges in AI adoption

Adopting any new technology is difficult, especially such a high-level one as AI. The biggest obstacle in 2020 is the same as in 2019: Lack of institutional support (22%), the report found.

Other obstacles included identifying appropriate business use cases (20%), skills gap (17%), and data quality issues (16%).

When comparing the challenges of organizations in the mature phase and the evaluation phase, the number of respondents facing a lack of institutional support doubled for those cited as in the evaluation stage. Perhaps late Ai adopters are more resistant to the integration of AI, according to the report.

The shortage of AI and machine learning skills is a consistent and persistent problem, according to the report. The shortage of machine learning modelers and data scientists (58%) topped the list as having the biggest skills gap.

At No. 2 came the challenge of understanding and maintaining a set of business use cases, which was cited by nearly half of respondents. Data engineering (40%) came in third, according to the report.

The most striking part of both findings is the consistency year-over-year, as the same skills areas were problems in 2019, the report found.

As stated in the report, “The uncomfortable truth is that the most critical skill shortages cannot easily be addressed,” which means these obstacles may take a long time to solve.

Managing AI risks

The risks for managing AI were evaluated in organizations in the mature and evaluation phases. In both categories, unexpected outcomes/predictions was the most common risk factor, cited by nearly two-thirds of the mature and about 53% of the still-evaluating, the report found.

For mature adopters, the need to control for interpretability and transparency of machine learning models was the second-most cited risk factor (~55%), while those still evaluating AI placed fairness, bias, and ethics (~40%) in second place.

While the risks of AI can be intimidating, the benefits can be worth the risk.

Most popular AI tools

For all adopters, supervised learning was the most popular machine learning technique. Organizations in the evaluation phase cited deep learning as slightly more used than supervised learning, but both were the clear frontrunners. Deep learning was the second most used technology for mature adopters.

Both evaluators and mature adopters cited model-based methods as the third-most used AI technology. However, some stark differences did exist: Nearly 23% of mature AI practices used transfer learning, which is nearly double that of less mature organizations.

While some differences were present, all organizations appeared to lean toward similar technologies.

When looking at the most popular AI-related tools, TensorFlow was the clear winner, reinforcing its dominance from last year, the report found. Its staying power reveals that deep learning and neural networks are becoming much more widely used, according to the report.

Four out of the five most popular tools were Python-based, indicating Python’s significant presence in AI and machine learning practices, the report found.

Data governance is falling behind

A little more than one-fifth of businesses said they have implemented formal data governance processes or tools to support or complement their AI projects, the report found. This is fairly consistent with last year’s data.

The good news, however, is that more than 26% of respondents said their organizations plan to launch formal data governance processes by 2021, with 35% expecting this to happen in the next few years.

AI adopters still view data governance as an additive instead of an essential ingredient, the report found, adding, “Ideally, data provenance, data lineage, consistent data definitions, rich metadata management, and other essentials of good data governance would be baked into, not grafted on top of an AI project.”

Organizations that want sustainable, successful AI integration must integrate data governance from the start to create a solid foundation for innovation, the report found.

For more, check out How is your company managing its AI and ML initiatives? on ZDNet.

Image: Igor Borisenko, Getty Images/iStockphoto