- AI and machine learning developers most often rely on language and speech-based Application Programmer Interfaces (APIs).
- 50% say gathering or generating data is the most challenging aspect of continually training and fine-tuning AI models.
- AI and machine learning developers say that the complexity of managing operations is the top challenge they face when developing AI applications.
Evans Data Corporation’s latest study of AI and machine learning development provides insights into the challenges developers face when building enterprise-level, high-quality AI apps. The study, Artificial Intelligence and Machine Learning 2019, Volume 2 is based on interviews with 500 AI and machine learning developers globally. Focusing on the attitudes, adoption patterns and intentions of AI and machine learning developers worldwide, the 187-page study is one of the most comprehensive of its kind. What makes the study noteworthy is the depth of research into AI and machine learning developer’s challenges today.
Key insights from the study include the following:
- 55.9% of AI and machine learning developers rely on language APIs, followed by speech (51.1%). Developers have long built on multiple series of APIs for AI and machine learning development. What’s interesting about this survey’s results is the popularity of conversation and data discovery APIs, indicating voice-activated assistants are now part of mainstream AI and machine learning software development.
- The lack of quality tools is slowing down AI and Machine Learning app development today. The most significant barriers AI and machine learning developers face in improving AI app development also include the cost of materials and lack of necessary skills or training. Just 10% are having to deal with the challenges of working with and integrating into legacy systems, a finding that indicates AI and machine learning app development is happening in relatively new business units and development centers.
- 38% of AI developers state that the complexity of managing operations is the top challenge when developing AI applications. The second most significant challenge is developing applications that are portable across deployment environments. Choosing the right AI framework is the third greatest challenge they face in creating quality AI apps. AI and machine learning frameworks are comprised of libraries of mathematical expressions and functions for various machine learning and deep learning operations. They often include a broad base of APIs and other development tools that are designed to assist developers in integrating into previous code and capitalizing on enterprise systems for the data needed to train models and produce the app.
- The majority of developers (54.9%) are relying on private cloud infrastructure for hosting their AI, machine learning, and deep learning development. 46% are relying on public cloud infrastructure, with just over 51% relying on their organizations’ on-premise infrastructure. Many cloud service providers have developed their cloud-based environments that incorporate an array of standard AI tools, including machine learning or deep learning frameworks, data-science-specific IDEs, and machine learning notebooks.