Transform’s AI & Edge Week 2022, held on July 26 to 28, includes a forum on how to build AI data engines that use the correct data at the right time. One of the interesting discussions here was the reality of the iteration cycle bottleneck.
This fundamental challenge is evidenced by how it takes months for every iteration to be established for 90 percent of enterprises, as explained by one of the resource speakers at AI & Edge Week.
What is this challenge all about? What can developers do to address it? These are essential questions to explore as AI progresses and more AI applications are created over time.
The difficulty of building production AI models
The iteration cycle bottleneck is mainly about the difficulty of creating models for production. Of note, this is different from the challenge of data annotation, which is about identifying data and labeling it to ensure correct identification for the AI system. This process used to be mostly manual, but it can now be made more efficient through automation.
It is hard to convert data into production AI models. While it is easy to build prototypes, transforming them into a form ready for production is an entirely different challenge.
In addition, teams usually produce models with 50 to 100 iterations, so there is a pressing need to expedite processes without making quality compromises.
The good news
The speakers at the AI & Edge Week did not point out a specific solution to resolve the bottleneck challenge. However, at least one suggested that the problem is somewhat improving.
This is because companies have started to “think at higher levels” and are learning to use the right technologies and solutions to rapidly iterate their AI models, allowing them to proceed to production faster.
Advanced automated data annotation like those that you find if you click here is one of these products. It does not solve the iteration bottleneck problem entirely on its own, but it provides a significant contribution to speeding up the process.
Development teams need to find the right technologies and relevant AI development products to use.
Improved iteration cycles have been observed in different fields of engineering over the years. For example, in the case of self-driving cars, rocketry, and biotechnology, development processes have progressed considerably faster because iteration cycles have revved up.
Innovative companies have developed ways to iterate their models more quickly and go ahead with production.
However, it is essential to emphasize that faster iterations, deployment, and time-to-market for products should not be the sole priority.
The resource speakers agreed that it is important to balance speed with general safety and other concerns such as data privacy and the ability to meet customer needs meaningfully.
In response to the question in the title, yes, the iteration bottleneck continues to be a challenge in AI development.
However, it is slowly being addressed with the development of new technologies, methods, and related products. There is no single unified solution for it yet, but developers can be creative and use a spectrum of solutions.