On AI adoption in Europe. Q&A with Thomas Forss and Magnus Westerlund
Q1. What is the mission of StageZero Technologies?
StageZero is an AI data company on a mission to solve access to AI training data for speech recognition and Natural Language Processing (NLP). Our company is uniquely suited for this as we have built a technology that integrates with apps and games to get access to hundreds of millions of users for the purpose of collecting and annotating training data.
Q2. You have recently published a report on the AI adoption in Europe. (*)
We have indeed, our report is about looking at how high-performing companies generate value using AI, what activities they focus on, what they do differently from other companies, how they approach implementation, and how they see the future. In addition to that, we have asked companies to rate different problems they have when implementing AI to see what they struggle with.
Q3. What methodology did you use to compile this report?
We aimed the survey at leaders and experts in data science working at companies implementing or using AI in some way in their products or services. We then sent it out based on LinkedIn titles and at least three years of self-reported experience in the field. It was sent out to people with titles such as Head of AI, Lead Data Scientist, Chief Information Officer, Senior Data Scientist, and Data Scientist in Europe. We collect responses for three months before compiling the results into the report. The main insights were extracted by asking people about how they perceive their company’s AI adoption: Ahead of the competition, same as the competition, or behind the competition, and then comparing the results of the answers between the groups.
Q4. Who were the respondents to this survey? How many did you consult and in which industries?
The survey was sent out to about 2000 people in different geographies in Europe ranging from small to large companies. We received 60 responses of which we disqualified some based-on region and a few more due to not implementing or using AI. We ended up with 37 qualified respondents from various industries (), covering 73 use cases in total.
Q5. What are the main insights you have learned from this report?
When it comes to the high performers, we see the following trends:
- They are more willing to collaborate, and more frequently work with third parties when it comes to acquiring or annotating data.
- They are further along when it comes to MLOps, and they fine-tune and retrain deployed models more often.
- High performers more often have a centralized decision-making structure when it comes to AI decisions.
MLOps (Machine Learning Operations) provides a systematic approach to the deployment, management, and optimization of machine learning models in production. It helps organizations overcome common challenges associated with the deployment of AI models, such as reproducibility, scalability, and collaboration. A key strength that MLOps brings is that it enables organizations to continuously monitor, update, and improve their models, ensuring that they remain relevant and accurate over time. As high performers in AI are characterized by a willingness to collaborate and work with third parties for data acquisition and annotation, adoption of MLOps allows them to manage these collaborations.
Q6. Is the lack of data the main challenge you have identified?
Yes, the lack of training data and quality training data was the challenge that most people responded to saying it was a challenge. Only 5% said it was not an issue while 3% were unsure. 49% have some issues, and 43% said they definitely have issues.
Q7. As my friend and colleague Gio Wiederhold wrote “An underlying assumption is that the data is a fair representation of the real world. That is often not true.” (**)
Did you find something similar in your findings?
The respondents represent the expert’s and leaders’ opinions and not necessarily always the company’s official stance. Furthermore, the sample size of the survey had an uneven spread of respondents from industries. This meant that we could only look at general insights from the entire field, not by industry. Next year we hope to be able to provide insights on a more granular level.
Q8. What are the main challenges you discovered in your survey in AI adoption?
Our respondents say the following:
- Access to training data is the main concern, and it limits the number of implementations companies are working on.
- Privacy laws are seen as a challenge by an overwhelming majority. This could turn into a major issue in a few years if different regions end up having different incompatible legislations.
- Synthetic data is already quite commonly used.
The European context requires companies to explore alternative methods for obtaining data. The increasing prevalence of synthetic data may indicate a promising development for AI development in the European Union.
Qx. Anything else you wish to add?
Organizations looking to leverage AI should prioritize the establishment of effective data practices for labeling and collection/creation. This includes active involvement from management to build collaborative partnerships. As companies move towards Industry 5.0 principles, with a focus on sustainability and human-centric production, it further highlights the importance of having strong data practices in AI. Here, synthetic data can play a crucial role in Industry 5.0 strategies by providing organizations with access to diverse and high-quality data more rapidly, while also addressing privacy and ethical concerns. This is a trend to keep an eye on in the future.
Thomas Forss is the co-founder and CEO of StageZero Technologies. He’s been in the field of NLP and AI for over 12 years, working in both academia and different AI startups. He holds a doctorate from Åbo Akademi University in Finland.
Magnus Westerlund (DSc) is a Principal Lecturer in Information Technology and Director of the Laboratory for Trustworthy AI at Arcada University of Applied Sciences in Helsinki, Finland. He is also an associate Professor at Kristiania University College, Oslo, Norway.
(**) Moving on
Draft, Gio Wiederhold, 30 October 2022