Why Can’t We Use AI to Improve AI Implementations?

Andy Jordan is President of Roffensian Consulting S.A., a Roatan, Honduras-based management consulting firm with a comprehensive project management practice. Andy always appreciates feedback and discussion on the issues raised in his articles and can be reached at [email protected]. Andy's new book Risk Management for Project Driven Organizations is now available.

  Artificial Intelligence  

If something fails around three quarters of the time (or even more often than that), it’s not going to take long before people start to question why so much money and time is being spent on it. Well, that’s where we are with artificial intelligence projects, with estimates frequently suggesting that 70-80% of them fail. (Some trusted sources suggest that it’s even higher.)

I know AI is still new, and I know the technology is constantly evolving, but that kind of failure rate isn’t sustainable. If things don’t improve, organizations are going to at least defer investments until the technology and current economic environment are more stable. And some are going to divert their limited discretionary funding elsewhere.

There are many reasons why AI implementations may not be successful. But there is clear evidence that one of the biggest issues is the data that is available to train AI tools. In most organizations, historic data is incomplete, inconsistent, inaccurate…or all of those things. And if that isn’t recognized or even identified, then the information provided by even the best AI solutions is going to be flawed.

Sadly, in most cases, it won’t be recognized. AI will train itself on whatever data is given to it. It won’t assume that the data is problematic, so it won’t be looking for data to exclude from …

Please log in or sign up below to read the rest of the article.