(Photo by Mohammad Esmaili on Unsplash)

When it comes to AI, we need more dishwashers.

I don't mean AI-powered dishwashers. Dishwashers already work just fine as-is. It's hard to imagine how throwing AI into that mix would improve anything.

No. What I mean is: we need to design AI-based applications, agents, and robots with a dishwasher mindset. We need to rewrite tasks to make the most of how AI works.

Look at your typical household dishwasher. It's great! You push a button, walk away for an hour, and you come back to clean plates and silverware.

You and the dishwasher can achieve the same goal – clean dishes – but you take different paths to get there. The dishwasher doesn't have robot arms that pick up a plate, run it under soapy water, rub a sponge across it, rinse it off, then pick up the next one. Instead, it sprays an entire chamber full of dishes with soapy water, rinses everything with clean water, drains that water, then heats the whole thing up to dry. Spinning water jets, not sponges, take care of the grunt work.

The same story plays out for washing machines, car washes, and other devices to which we offload our manual tasks.

That brings us back to AI. When you see AI goofs making headlines, ask yourself whether that problem stems from:

1. AI's inherent flaws. Because every model will be wrong now and then.
2. Misuse of AI. Companies sometimes push AI into roles for which it's not suitable.
3. Poor task fit. AI could have met the end-goal just fine, but someone tried to get it to tackle the task like a person.

Let's call that third group "dishwasher tasks." Here, AI could succeed if we were to let it be AI instead of asking it to imitate a human. Because sometimes, to take advantage of a technology, it's better to rewrite the procedure.

This first occurred to me several years ago, when reading James Gentle's Matrix Algebra textbook. I'll have to paraphrase the idea, because for the life of me I can't find the exact quote, but it boils down to: People and computers can both do matrix math. But if you implement a person's approach in code, it will be very slow.

(And if you think that slow matrix math isn't a big deal, note that the algorithms behind data science, machine learning, and AI are all based on matrix transforms. If those systems ran at a fraction of their current speed, this field would not be nearly as useful.)

Keep your dishwasher in mind as you assign tasks to AI. Consider starting with the end-goal, then working backwards to determine how this could be done in AI fashion.