Living in the Seattle area, I have the opportunity to be exposed to the latest and greatest artificial intelligence (AI) experiences, like the Amazon Go Store, which knows when you pick up an item in the store for checkout and when you put one back, all culminating in an app that simplifies your checkout experience with automation.
These are the types of AI experiences that businesses hope for and can attain if they harness AI not only for the rewards, but also with an eye on managing the risks and ensuring their own readiness.
SEE: The ethical challenges of AI: A leader’s guide (free PDF) (TechRepublic)
Alex Fly, CEO of AI solution provider Quickpath, calls this the “three Rs” of artificial intelligence: Reward, risk, and readiness.
“What CIOs and other individuals at the C-level [in organizations] should note is that AI is a methodology that uses an experimental framework,” said Fly.
When you implement AI, whether it is operating on big data, traditional data, or a blend of the two, the testing process is iterative. You begin with small steps, and you test the accuracy of your data and your algorithms. You do this by determining how closely data and algorithms capture the realities of your business and deliver the insights that you want.
In some cases, the experiment produces results right away. In other cases, there is a need for continuous improvement. In still other cases, the experiment doesn’t work.
“The key is to pilot your AI first,” said Fly. “Measure your results against your benchmarks and your expectations. If your first efforts don’t achieve what you want, refine those models. Perfecting an AI application is an iterative process of continuous improvement. By incrementally improving your results, you are lowering your risk of producing inaccurate results.”
The concept of iterative testing can have varying impacts on projects. For example, if the velocity of data in a given business process you are adapting for AI is rapid, you can iteratively test and re-deploy quickly; however, if the business process and flows of data are slow, the iterative AI testing cycle will be slow, too, which can try the patience of upper management and project sponsors.
“One important key to AI success is transparency,” said Fly.
So, if the AI testing process by necessity must be slow, management should be informed of it upfront. If the AI project is successfully implemented and it impacts your customers’ expectations of privacy, such as an insurance company contracting with third parties to obtain customer mileage information in order to compute auto premiums, consumers should be informed of the practice and the AI upfront—and not in the fine print of policies.
“There is also the question of IT readiness,” explains Fly. “Do you have the right skill sets on your IT and data science teams to support and monitor the AI, and to implement it on websites, in mobile apps, and in systems? With AI’s great rewards come great responsibilities. These include managing risks and also assuring your readiness as you reap the benefits of AI.”
Fly continues, “If you’re impacting in new ways how your customer interacts with you or how people work within your own organization with an AI app, customer sensitivity and employee readiness for an AI introduction should be assessed.”
“A sound approach with any new AI is one of ‘crawl, walk, and then run,'” said Fly. This lets you know if the organization is ready for the change you want to introduce. You and your stakeholders should also verify that the business case the AI was designed for will be able to be met, and if you have the right data for the AI algorithms to operate on.