New AI tools, capabilities, and services are released almost daily. They promise to revolutionize the way government operates. When evaluating these AI tools and capabilities, note that there’s more to AI than simply building models. This is particularly true when considering more customizable options of interactive AI platforms or building from scratch.
You’ll need to evaluate development environments, infrastructure, data management, data manipulation and visualization, and computing power technologies. Some of these are offered as services through software, platform or infrastructure as a service (SaaS, PaaS and IaaS). Some are available as hardware or software installations, and some are available open source.
Though not an exhaustive list, the tools and platforms outlined below highlight what you may need to create an AI solution.
Cloud & Infrastructure
Many AI tools and solutions are tied to a cloud platform. Elastic storage, computing infrastructure, and many pre-package ML libraries help accelerate ML model development and training. Agencies with limited in-house computing resources need a cloud platform; so do ML models that require intense computing resources for training, such as for Deep Learning and GPU acceleration. A cloud platform can be more economical when the computing requirements are short-term and sporadic, depending on data security requirements.
Use orchestration tools to help manage complex tasks and workflows across the infrastructure. A variety of open source tools are available.
DevSecOps is the integrated practice of bringing together software development, IT operations, and the security team. It’s a critical part of successful AI delivery.
At a high level, DevSecOps includes an environment to manage development tools such as:
- programming languages like Python, R, Java and C++
- code repositories
- build and unit testing tools
- version control management for code and models
- tools to manage code quality
- version control management to perform security scans, and monitor and perform testing and ongoing performance
Data collection, ingestion, management, and manipulation are AI development’s more critical and challenging elements. Tools are available to handle various tasks associated with data operations, including: tools for data acquisition, data cataloging, data collection and management frameworks, data ingestion frameworks, data labeling tools, data processing tools and libraries, data sharing services, and data storage.
Not every AI solution requires every tool. The relevant tools depend on the size, complexity, structure, and location of the data being used to train AI models.
Artificial Intelligence (AI) and Machine Learning (ML)
Many new tools and products support AI development. These include data science toolkits (combined offerings for applied mathematical statistics); visualization tools to explore data, understand model performance, and present results; and machine learning frameworks that provide pre-build architectures and models to reduce the development effort.
AutoML tools—tools that automate many of the model training and deployment processes—are an emerging area of AI tool development. They further reduce the access barrier to AI technology. A number of these solutions offer low-coding ML tools.
Many of these tools are open source, but many are offered as commercial products. Selecting the right tools for the job will require careful evaluation and involve all members of the Integrated Product Team (IPT). These tools are often provided by and operated by the central AI resource.