Working Up

by Dwayne Phillips

Borrowing from The Software Crisis of the 1990s, I declare The AI Crisis.

A recent report from MIT claims that 95% of AI projects attempted by well-meaning folks fail. Gosh. That is a pretty high percentage.

I remember the software crisis of the late 1980s and all through the 1990s. Reports like this MIT report were everywhere. High percentages of software projects failed. What was wrong? What to do?

What was wrong? People were doing stupid things. “Let’s write some software to implement my great idea. I think it will work.” This brought phrases like, “Hope isn’t a plan” and such. Yes, stupid things. Hire a roomful of programmers, lock them in a room with no windows, yell at them now and then, and good things are supposed to happen, right? Good grief. Of course good things wouldn’t happen, but well-meaning folks thought it would.

One thing to come along was the Capability Maturity Model whereby people would decide what their problem was before attempting a solution. Imagine such a concept! The Capability Maturity Model helped a lot of people do better. Basic problem-solving basics.

Then the agile methods came in. Hey, we have tools that lessen the cost of experiments. Let’s run more experiments, learn more, then do more. That also helped a lot of people do better.

Still, you need some idea of the problem you are attempting to solve.

Now, here we go again, AI is everywhere. Everyone must have an AI project or be left behind in a pile of silicon dust (I think silicon dust is also called sand, but I digress). Let’s hire a roomful of AI people (what do you call those people?), lock them in a room with no windows, yell at them now and then, and good things are supposed to happen, right? Gosh.

Of course this fails. The high failure rate puts us in The AI Crisis. Someone write it down that I wrote it down.

We still need some idea of the problem we are attempting to solve.

And another thing, a lab project is not a marketable product. It is a lab project. Take the result, give it to a bunch of hardened adults, and they might produce a marketable product.

AI requires compute power—money. AI requires people—more money. AI requires direction and some hardened adults who have turned ideas into marketable products before.

Just flail around and you will become part of the 95%. Come on, we can do better.