The state of Machine Learning at the end of 2022
What is ML Insider?
The ML Insider is an annual analysis by cnvrg.io, an Intel company, of the Machine Learning industry highlighting key trends, points of interest, and challenges that AI developers experience every day. This report offers insights into how over 430 AI professionals are building, training, deploying and adapting their machine learning stack to better suit today’s modern, complex ML workflow.
Compare salary benchmarks between roles, technology stacks, challenges, and operationalization tactics across the industry, to help improve your AI efforts. Download the full ML Insider 2022 Report to see the complete results.
Who took this survey?
The survey covered 430 participants, representing organizations of all sizes from a few employees to over 5,000. 53% of respondents came from companies with 600 employees or less. The insights from the survey covered dozens of industries; among the most common were: Information Technology Services and Computer Software. Financial Services/Banking, Education, Healthcare, Consumer Goods, Telecommunications, Automotive, Insurance Media/Entertainment, and Defense represent the other half of respondents. 42% of all respondents are data scientists, 16.9% engineering/DevOps and 10.6% in software development roles.
Key Takeaways and Trends
- AI is recession-resilient
Despite the rough economic circumstances, investment in AI is expected to increase in 2023.
- AI maturity remains low
57% of respondents reported a low AI maturity with less than 4 models running in production.
- Lack of knowledge and expertise remains a top AI challenge
Consistent with 2021 data, lack of knowledge and expertise, and hiring AI/data science talent remain the top 2 challenges executing ML programs.
- A majority of organizations plan to address AI explainability in 2023
43.5% of respondents are planning to introduce explainable AI techniques in the next 12 months, while only 37% already have AI explainability techniques in place.
- Industries with more consumer regulatory pressure tend to have lower AI adoption
Defense, Automotive, and Computer Software have the highest AI adoption, while Education/E-learning, Hospital & Healthcare, as well as Media/Entertainment have seen the lowest AI adoption.
- Operationalization of AI is still heavily dependent on Developers, DevOps and Engineering
64% of respondents that found it difficult to successfully execute ML rely on Developers/ DevOps/Engineering to operationalize their AI models
- More organizations are adopting hybrid infrastructures
Compared to 2021, there has been a 13% increase in hybrid compute adoption.
- 89% of organizations are seeing the benefits of their AI solutions
The majority of organizations investing in AI report that they are benefiting from their AI solutions.
AI is recession-resilient
Amid high inflation, low market performance, supply chain issues, and uncertainties from both the war in Ukraine and the lingering pandemic, it is safe to say that the global economic state in 2022 has been less than ideal. The tech sector in particular has been hit hard by the economic situation which has led to massive company layoffs, budget cuts, and decreasing earnings from tech leaders. So, what does that mean for the future of AI in 2023?
Nearly 50% of respondents indicated that they believe organization investment in AI development will actually increase.
What do you believe the biggest impact will be on AI development
from the current economic situation?
AI maturity remains low
Overall AI maturity remains quite low. 57% of respondents reported a low AI maturity (0-4 AI models running in their product). The below figure measures AI maturity by the number of models an organization has running in their product. It should be considered that organizations may in fact have AI that is not in production or applied to a product solution.
At what level is AI being used in your organization’s applications?
Lack of knowledge and expertise remains a top AI challenge
Consistent with ML Insider 2021, lack of knowledge and expertise remains the top AI challenge for most organizations along with hiring AI talent. When combined with the fact that 61% of respondents reported having less than 8 members on their team, difficulty in obtaining talent points to the need for getting the most out of what talent already exists in an organization in order to scale AI.
What is your company’s main challenge in executing ML programs?
How long does it typically take to get from experimentation (model staging, training, testing)
A majority of organizations plan to address AI explainability in 2023
75% of AI developers agree that AI explainability is a top concern. 43.5% of respondents are planning to introduce explainable AI techniques in the next 12 months, while only 37% already have AI explainability techniques in place.
As an AI developer, AI explainability is a top concern for me
How is your team addressing explainable AI?
Industries with more consumer regulatory pressure tend to have lower AI adoption
Which industries have the highest/lowest AI maturity? Education/E-learning, Hospital & Healthcare, and Media/Entertainment have seen the lowest AI adoption, with the largest number of respondents indicating that they have 0-5 models running. The industries with the highest adoption include Defense, Telecommunications, and Insurance.
AI Maturity by Industry
Operationalization of AI is still heavily dependent on Developers, DevOps and Engineering
66% of respondents rely on Developers, DevOps and Engineering to operationalize their models. The technical complexity of operationalizing AI models requires the technical skill of developers, engineers, and DevOps which can add bottlenecks and increase time to production.
I rely on Developers/DevOps/Engineering to operationalize my model
More organizations are adopting hybrid infrastructures
Compared to 2021, there has been a 13% increase in hybrid cloud adoption. Organizations are shifting away from on premises-only and cloud-only infrastructures. The trend in 2022 is that hybrid cloud is becoming more popular. This is likely due to the flexibility to utilize different types of computing for different types of tasks, as well as the high cost of cloud. Hybrid infrastructures allow organizations to reduce costs and improve AI workload performance. While hybrid infrastructures have their benefits, it can still be a challenge for organizations to manage hybrid infrastructures. Organizations will need to adapt to these challenges and enable AI developers to easily shift workloads between clouds and on premises resources to fully take advantage of hybrid infrastructure benefits.
Where do you (mostly) run your ML workloads?
89% of organizations are seeing the benefits of their AI solutions
Your organization may be considering whether or not to invest in AI solutions. 89% of respondents reported that they are in fact seeing the benefits of their AI solutions.
Is your organization seeing the benefits of your AI solutions
Download the full ML Insider 2022 Report to see the complete results.
cnvrg.io is a full stack machine learning operating system with everything an AI developer needs to build and deploy AI on any infrastructure. cnvrg.io was built by data scientists, to help data scientists and developers automate training and deployment of ML pipelines at scale. cnvrg.io helps organizations accelerate value from data science.