Main barrier to AI in the enterprise now skills, not culture

Profile picture for user ddpreez By Derek du Preez April 20, 2021 Audio mode
The latest AI adoption report from O’Reilly - a think tank set up by the influential Tim O’Reilly - sees some interesting shifts happening in the enterprise.

Image of a signpost with the words ‘new skills’ on it
(Image by Gerd Altmann from Pixabay )

It has been predicted for years, but the AI skills crisis is finally becoming a reality for enterprise buyers. Just last year it was company culture that was the major bottleneck for those looking to deploy AI tools, but it seems that the acceptance of ‘Artificial Intelligence' as a workable technology has grown significantly - and the focus has now shifted to finding the people to make that happen. 

This is according to the latest survey from O'Reilly - a think tank of sorts that was set up by the influential Tim O'Reilly, who spearheaded the popularity of open source software and led with ideas such as Web 2.0. 

The survey was carried out by subscribers to O'Reilly's AI newsletters, where the number of respondents grew significantly on last year's survey. Almost three times as many people (3,574) participated in the research this time round, suggesting that more people are using AI than ever before. 

The report notes: 

Of the 3,574 respondents who completed this year's survey, 3,099 were working with AI in some way: considering it, evaluating it, or putting products into production.

Unsurprisingly, the largest cohort are based in the United States (39%), followed by India, Canadian, Germany, the UK and Spain. In terms of industries, computers, electronics and technology came out on top, representing 17% of respondents, followed by financial services (15%), healthcare (95) and education (8%). O'Reilly is seeing relatively little use of AI in the pharmaceutical and chemical industries at present (2%), as well as the energy (3%), manufacturing (4%) and retail (4%) sectors. 

This data is interesting, as it gives us a clear sense of where AI is being used and how. In addition to the above, 26% of respondents described their use of AI as ‘mature', meaning that they had revenue-bearing AI products in production. That's impressively high, from where I'm sitting. Some 35% said that they were ‘evaluating' AI, which is approximately the same as last year, and 13% said that they weren't making use of AI or considering using it, which is a couple of percentage points down on 2020's findings. 

Barriers are changing

It's the barriers to adoption data that's of particular interest, however. It's important to understand this, not just so we have a better sense of how we problem solve for the future, but also to give us an understanding of what stage of AI use buyers are at - and more broadly how the industry needs to respond. 

The report notes how the focus has shifted from company culture, to skills and data. It states: 

Looking at the problems respondents faced in AI adoption provides another way to gauge the overall maturity of AI as a field. Last year, the major bottleneck holding back adoption was company culture (22%), followed by the difficulty of identifying appropriate use cases (20%). This year, cultural problems are in fourth place (14%) and finding appropriate use cases is in third (17%). That's a very signifi‐ cant change, particularly for corporate culture. Companies have accepted AI to a much greater degree, although finding appropriate problems to solve still remains a challenge.

The biggest problems in this year's survey are lack of skilled people and difficulty in hiring (19%) and data quality (18%). It's no surprise that the demand for AI expertise has exceeded the supply, but it's important to realize that it's now become the biggest bar to wider adoption. The biggest skills gaps were ML modelers and data scien‐ tists (52%), understanding business use cases (49%), and data engi‐ neering (42%). The need for people managing and maintaining computing infrastructure was comparatively low (24%), hinting that companies are solving their infrastructure requirements in the cloud.

It's gratifying to note that organizations starting to realize the impor‐ tance of data quality (18%). We've known about "garbage in, garbage out" for a long time; that goes double for AI. Bad data yields bad results at scale.

Techniques and usage

The survey also provides us with some interesting insights into how AI is currently being used. For instance, 82% of respondents said that they are using AI techniques such as supervised learning, and 67% are using deep learning. 

However, beyond this there was a big drop off. Human-in-the-loop, knowledge graphs, reinforcement learning, simulation, and planning and reasoning all saw usage below 40%. 

Interestingly, only a very small number of respondents wrote in natural language processing as a response. This is noteworthy given the hype in the industry regarding the use of natural language processing, with many vendors placing big bets on the use of the technology in call centres and for customer service. The results don't bode too well for its success, thus far. 

In terms of tools, scikit-lean, TensorFlow, PyTorch and Keras each scored over 45%. Following these, tools included Amazon's SageMaker (25%), Microsoft's Azure ML Studio (21%), and Google Cloud's ML Engine (18%). The difference between adoption between the latter group and the more popular first group is also particularly noteworthy, in my opinion. 

My take

This data is very useful in guiding the conversation around AI use in the enterprise and cutting through the hype that is put out by some of the larger vendors in the market. It seems that adoption is desired, but that we aren't going to get anywhere particularly productive if we don't figure out how to get the right skills in place. This is important given the role that AI could play in the future and what's at stake. 

Mike Loukides, vice president of content strategy at O'Reilly and the report's author, says: 

Enterprise AI has grown; the sheer number of survey respondents will tell you that, but deployment of AI applications into production has remained roughly constant, and with it, overall maturity in the field. 

It's no surprise that the demand for AI expertise has exceeded the supply-that's been predicted for years-but it's important to realise that it's now become the biggest barrier to wider adoption.