When AI Meets IT
Discover more from AI in Business
When AI Meets IT
Getting IT professionals in synch with AI development teams is a green field the needs to take shape for AI systems to deliver on strategy for the enterprise
By John P. Desmond, Editor, AI in Business
Information technology culture arguably began with the introduction of IBM mainframes in the 1960s, with their arcane operating systems and programming languages requiring a new data center culture to figure out how to get them to work and keep them running.
In medium to large enterprises, data center culture has remained, surviving through successive generations of technology such as the rise of personal computers and client/server computing that tied powerful desktop machines to the mainframes. Big waves of computing infrastructure changes usually require an overhaul in the way software is developed, changes in the languages used, and changes in the job titles in the organization needed to get the work done.
AI may be the most disruptive technology evolution yet, especially now in the new era of generative AI and large language models such as ChatGPT-4, which is capable of automating some degree of the coding required to launch new applications.
How the new AI regime fits into the enterprise data center culture, is an ongoing transition. Here is a look at how it is being addressed.
“I have had many discussions with IT professionals who handle AI and ML projects as the main part of their jobs,” stated Jan Burian, a senior director of IDC Manufacturing Insights EMEA, based in Prague, in an account he authored in a recent issue of IndustryWeek.
An early conclusion was that management of AI projects is best handled with an agile approach, rather than a waterfall approach, where waterfall breaks down the steps into separate phases, and agile employs iterations of development cycles in a continuous feedback loop.
“During the initial stages of the projects, AI/ML experts realized how isolated ML knowledge is from other IT skills available in the traditional IT department,” Burian stated, making a key point. At the same time AI is poised to take over software development, the AI experts may have no one in the IT department who speaks their language.
The standard questions that come up in the IT department apply: buy or build? Who will operate the model? Who will train the staff? “Who will fine-tine the hyperparameters of neural networks?” Burian queried. These are the right questions, and they lead down a path “closing with the million-dollar question: Who should be on the development and deployment team?”
The team members need titles, and they need to decide how the neural networks should be monitored, what infrastructure is needed to support the AI system, and whether it is in line with the company’s development strategies. Also, “What safety measures will we need to take?” Burian queried.
Example of Effort Required for AI Systems in Auto Manufacturing
An example from the automotive industry shows the complexity of putting an AI system into production. To train a model to identify 17 types of defects, over 140,000 samples had to be used for training. The software system – including infrastructure, backend, integrated layers and databases – went through 25 iterations. “And the neural network model has 52 iterations!” Burian stated.
Three main groups of project stakeholders were identified by the experts: customers, who will be operating the environment where the AI system is being deployed; translators who can explain the situation, status or problem to data scientists; and developers, who are ideal if they combine “the skill sets of data scientist, data engineer, data architect, UX designer and delivery manager,” Burian stated.
In conclusion, “AI-powered technology has definitely proven its value for a handful of use cases across industries. However, the path toward a fully industrial and scalable solution is still bumpy and winding,” Burian stated.
Another observer sees that a “data-driven culture” is needed to successfully implement AI systems. Writing recently in GovLoop, Greg Godbout, chief growth officer for Fearless, a digital services firm based in Baltimore, stated, “We recommend using two maturity models to assess data-driven culture evolution: data maturity and analytics maturity.”
Here are the stages of data maturity: basic, organization needs to build structure into its data collection to do serious analysis; reactive, data is starting to be collected into a single data warehouse to enable a holistic view; proactive, the data strategy is mature enough to enable the building of statistical models; strategic, where data is being used to make predictive models; and innovation, where machine learning, AI and robotic process automation are being used to automate decisions and data is being enriched with third-party information.
Here are the stages of analytical maturity: descriptive, in which spreadsheets are used for data exploration; explanatory, where historical data is used to understand the present; exploratory, where statistical analysis and data visualization are used to explain the circumstances of an event; predictive, where predictive models are used to forecast future scenarios; and prescriptive, where AI and ML are used to make recommendations.
“Before we can evolve into a data-driven culture, we have to understand where we are in our usage and analysis of data.” Godbout stated.
IT Can Try to Put Safeguards in Place
One observer sees the role of IT to put some safeguards in place for AI system development and deployment.
“CIOs need to also set limits to protect against impulsive AI investment,” stated Jason Conyard, CIO of VMware, in a recent account in Information Week. “At this point in the hype cycle, it’s important that we understand the limitations and implications of AI so that we can strategically make the right investments.”
After identifying how the AI project would deliver on the strategic business mission, the team needs to ensure the foundation is in place for a successful effort. “Operations and IT teams are needed to provide feedback on the feasibility and scalability of AI-based tool development and deployment and to assure that the necessary IT infrastructure required to support AI-based app deployment is in place,” stated Mike Erlihson, principal data scientist at Salt Security, which offers an API security platform.
He also suggested that AI project teams “prioritize and test a few AI use cases before scaling up the investment.”
Read the source articles and information in IndustryWeek, in GovLoop and in Information Week.
(Write to the editor here.)