TDWI | Training & Research | Business Intelligence, Analytics, Big Data, Data Warehousing

The Question Everyone Should Ask Before Deploying AI

We need AI to help us make better decisions rather than speed up the impact of our bad ones.

• By Rob Enderle
• July 1, 2024

In college, I worked part-time as a Jaguar mechanic. One of the first things I learned was to focus my selection of tools on the problem, not pick a tool first. In the case of AI, the question we should ask first is “where do we need AI’s help the most?” This is a recurring problem with technology. We have huge industry initiatives designed to sell products that are successful from a sales perspective but often lack a strong return on investment.

For Further Reading: Lessons Learned from Facebook’s Poor AI Implementation The Unfortunate Decision Process That Is Leading to AI Deployment Failures The Next Thing to Look For in AI Vendors: Interoperation

Client/server computing didn’t even work for the first decade it was deployed. Big data focused us on buying massive amounts of storage before we’d figured out how to use it. One of the biggest disasters in that push happened at a massive NSA data center that both cost billions and turned out to be pretty much worthless. Then came “digital transformation”; it sounded great but didn’t really focus on very real problems that needed fixing. After their digital transformation, many companies found they’d spent millions only to make operations worse. 

With AI, we are focused on productivity, but is productivity really our biggest problem -- or is it that we continue to make a lot of really awful decisions? Maybe we should first focus on making better decisions before we deploy AIs modeled on our process that will make these bad decisions faster -- so fast that we may be unable to recover from them.

How We Should Rethink AI Purchases

Back when Windows 95 came out, Microsoft did an incredible job of creating demand for the product. People were deploying it everyplace, including me who put it on my CEO’s PC and bricked it. It was not exactly my best career move. Worse was a guy at Intel who put it on a machine that ran a FAB and crashed the entire line. It takes weeks to restart a process like that, or it did back then. To say a mistake like that is career-limiting is an understatement.

Productivity, which is where AI seems to be focused at the moment, is important, but increasing the speed of anything before you first assure direction can, and often does, result in the company spending millions of dollars going in the wrong direction. Although that was problematic during the client/server, big data, and digital transformation campaigns, it could be deadly with AI.

Case in point was Tesla’s Autopilot effort which convinced people, many of whom died as a result, that they’d bought a true autopilot when what they had was advanced cruise control. When I was a kid, my father told a story about a guy who flew in from England, rented an RV, and asked what cruise control meant. He was told it was like autopilot, so he got on the freeway, turned on cruise control, and went back and tried to make a pot of coffee. He survived, but it ended badly.

As people run around like chickens with their heads cut off trying to deploy this largely unreliable tool, they should ask what they need fixed first. I’d argue they need to fix the quality of their decisions first, not the speed at which they make them.

Before they deploy it to make decisions, they need to ensure first it is dependable. Right now, too many buyers are thinking generative AI is the same as AGI (artificial general intelligence) -- not the case, as explained in this Forbes comparison -- and that generative AI can be trusted, which is also not true according to a Wall Street Journal article.

Final Thoughts

Like most every tech wave I’ve been involved with, this latest AI wave is way too focused on driving revenue and not focused enough on ensuring a positive result. To be successful, buyers need to have both a clear idea what they need AI to do and a deep understanding of what it can do. Neither is evident in many of the decisions I’m seeing, and have seen, made over the years.

We all need AI to be more focused on helping us make better decisions rather than speeding up the impact of our bad ones. I also think that we should tune out these industry initiatives and instead prioritize what we need fixed, and then choose the best tool. In short, we need to be smarter before we go faster. Otherwise, like that guy in the RV, things are likely to end badly.