by Adrian Bowles
As an AI analyst, the question I hear most is, “Is AI ready for primetime?” My default answer is, “Yes. AI is ready, but are you ready?” Of course, this is a gross simplification, so let’s go a little deeper.
Overview
For every major AI sub-discipline (including machine learning, knowledge representation, vision, and natural language understanding and generation), one can find off-the-shelf services to support the development of value-producing applications. The quality of these services varies and advances are being made constantly, but one can build great applications today with AI technologies at their core. That doesn’t mean that you can build a great AI-powered application today and it doesn’t mean that every problem is an AI problem.
It’s been said that if your only tool is a hammer, every problem looks like a nail. Just because you can add machine learning or a natural language interface to an application doesn’t mean that you should or that you can do it well. The challenge is to ensure that you select the right technologies, vendors, and products based on the application you want to build, and that you have the right resources.
Process Maturity Models Provide a Starting Point
Over the years, I’ve spent a lot of time studying attributes of successful projects and teams. In the 1980s, the Software Engineering Institute at Carnegie Mellon University embarked on an ambitious project to create a maturity model for the software development process. The original SEI Capability Maturity Model (CMM) had five levels, ranging from initial to optimizing, and a set of criteria that enterprises could use to assess their own maturity. The CMM has been updated over the years and derivative maturity models have been developed to provide insights into quality beyond the development process.
My 1993 3-D Assessment Model
I briefly represented a software consulting organization in the early CMM discussions and was heavily influenced by their work. In 1993, I wrote a column for Object Magazine that outlined three dimensions that contribute to software quality: people, process, and products.
Process, of course, was based on the CMM, but it was my belief that to really understand or build a foundation for quality development, it was important to quantify attributes of the people involved and the products they used. 25 years later, I think these three dimensions stand the test of time. They are necessary, but not sufficient, to assess an organization’s AI readiness.
What’s Different About Building AI Apps?
Recently, based on client questions and engagements, I’ve started to update that model to make it appropriate for modern AI-powered applications. This assessment tool is a work in progress, and I invite readers to share their own thoughts as I refine the tool.
In addition to the people, process, and product (technology) dimensions, I’ve added a data dimension and an emphasis on the significance of application attributes. Identifying data and application requirements and mapping them to technologies and products is a critical part of the assessment process and far beyond the scope of this blog. Today, this generally requires a consulting engagement, but the goal is to codify the process to enable self-assessment in the spirit of the CMM in the future.
Success Still Depends on People: For Now
Automating application development with AI has been the subject of research and speculation for decades. The reality is that successful projects require specific roles and skills that are not yet fully automated. In our assessment process, we look at the human side of AI development in terms of requirements for specific talents and skills, the need for visionary roles, and knowledge of ethical considerations that can make or break an application.
But Wait, There’s More…
We’ve developed and we are refining questions and criteria for each of the dimensions to help clients evaluate and improve their AI-powered application development process. The work is ongoing, and we plan to share components in upcoming research, webinars, and publications based on our client engagements.
For more information, please contact me at adrian@aragonresearch.com.