O’Reilly, a training and learning platform, conducts an annual survey regarding AI adoption in businesses. Last December and January, it asked recipients of its newsletters to answer a questionnaire on this topic. Participation was lower than last year, perhaps due to the end of year vacations, the highlight is that the responses are similar to those of 2021, although after verification, only 10% of them are from the same people.

This report on enterprise AI adoption was written by Mike Loukides, vice president of content strategy for O’Reilly Media, Inc. He focuses on programming languages, Unix, data and artificial intelligence, ethics, the future of programming… In this report, he examines the different ways in which artificial intelligence is being implemented, the techniques and tools used by companies to better understand the results of its adoption over the past year.

The 2022 figures show that the percentage of organizations reporting AI applications in production, generating revenue in production, has remained constant over the past two years at 26%, which Mike Loukides says indicates that AI has moved on from the hype. He states:

“For years, AI has been at the center of the technology world. Now that the hype has faded, it’s time for AI to prove that it can deliver real value, whether it’s cost savings, increased productivity for businesses, or the creation of applications that can generate real value for human lives. This will undoubtedly require practitioners to develop better ways to collaborate between AI systems and humans, and more sophisticated methods to train AI models that can bypass the biases and stereotypes that plague human decision making.”

AI adoption

31% of companies report not using AI (up from 13% in 2021), 43% are evaluating adoption, and 26% have implemented AI applications.

The main increase, from 18% to 31%, of manufacturing respondents with AI is in Oceania.

Next, North America and Europe had the highest percentage of respondents: 27%, followed by Asia (24%) and South America (22%). As for Africa, it had only 13% of respondents adopting AI in manufacturing (13%) but the highest number of non-users (42%).

A lack of governance

A large number of organizations lack AI governance. Of the 26% of respondents with AI products in production, only 49% have a governance plan in place to oversee how projects are created, measured and observed (49%) versus 51% for those without.

When it comes to risk assessment, unexpected outcomes (68%) remained the top concern for mature organizations, followed closely by model interpretability and model degradation (61% each). Confidentiality (54%), fairness (51%) and security (42%) were the least cited risks by organizations.

The report’s key figures for mature practices

• TensorFlow and scikit-learn (both 63%) are the most widely used AI tools, followed by PyTorch (50%), Keras (40%), and AWS SageMaker (26%).
• AutoML tools are used to automatically generate models within 67% of organizations compared to 49% of organizations the previous year, an increase of 37%.
• There was also a 20% increase in the use of automated tools for deployment and monitoring. The most commonly used tools are MLflow (26%), Kubeflow (21%) and TensorFlow Extended (TFX, 15%).
• The top bottlenecks to AI adoption are lack of skilled people and lack of data or data quality issues (both at 20%).
• Organizations with mature practices and those currently evaluating AI agree that lack of skilled people is a significant barrier to AI adoption, although only 7% of respondents in each group cited it as the most significant problem. Experts in ML modeling and data science (45%), data engineering (43%), and managing a set of business use cases (40%) were cited the most.
• Retail and financial services had the highest percentage of mature practices (37% and 35%, respectively). Education and government (9%) have the lowest percentage of respondents but are the most considering AI adoption (46% and 50%, respectively).

Laura Baldwin, president of O’Reilly, concludes:

“While AI adoption is slowing, it is certainly not stagnating. There is significant venture capital investment in AI, with 20% of all funding going to AI companies. This likely means that AI growth is plateauing in the near term, but these investments will pay off later in the decade. In the meantime, companies should not lose sight of the purpose of AI: to improve people’s lives. The AI community must take the steps necessary to create applications that generate real human value, or we risk entering a period of reduced funding for artificial intelligence.”

Translated from O’Reilly publie son rapport 2022 sur l’adoption de l’IA en entreprise