You can't skip straight to the AI party

In 2014, Amazon set out to build an AI that could read resumes and pick the best candidates. They trained it on ten years of resumes the company had received. Most of those came from men.

So the system learned what it was taught. Reuters later reported that it marked down resumes containing the word "women's," as in "women's chess club captain." Amazon tried to correct it, lost confidence, and scrapped the project.

The model wasn't broken. It faithfully learned the pattern in its data. Nobody had looked hard enough at the data.

The truth about the magic

Data science is the bedrock of AI. Every large language model, every predictive model, every "magic" generative tool traces back to analytics done right. Ignore that, and you're building castles in the sand.

Learn, teach, apply

Let's make it simple. Analytics is how we learn. Data science is how we teach. AI is how we apply.

Before you can automate or "intelligentize" anything (yes, I'm coining that), you need clean data, meaningful patterns, and tested models. That's the gritty, behind-the-scenes work of data science. Without it, generative AI is just a fancy summarizing parrot.

Trust needs a paper trail

NIST's AI Risk Management Framework lists accountability and transparency among the marks of trustworthy AI. In plain terms, that means knowing where your data came from, being able to reproduce a result, and having sound analysis underneath.

In short: data science.

So no, the old skills aren't dead. They are very much alive. Possibly undead, and coming for YOUR brains.

Four questions before you buy the AI

• What data will it learn from, and who has examined it?
• Is that data complete, current, and representative? Or does it carry yesterday's habits forward, the way Amazon's did?
• Can we trace an answer back to its source?
• Who owns data quality after launch? If the answer is "nobody," you have your answer.

The door and the key

Let's stop pretending you can skip the hard stuff and jump straight to the AI party. Data science is the door, and analytics is the key.

You don't get artificial intelligence without natural intelligence first.

A shorter version of this article first appeared on LinkedIn. Join the discussion there.