You'll Pay Now or You'll Pay Later
Happy tax week.
AI is about the data.
“Good” means accurate and trustworthy, not “desired”
If you took a 2 year old and you taught that child that a table was, in fact, called a “chair,” the 2 year old would believe it until you told it otherwise.
The data you gave the 2 year old was bad. A table is called “table”, not “chair.”
Good data is important. Good data is how you know how much money you made last year, how many Instagram followers you have or how far away your in-laws live.
Bad or questionable data is bad. The quality of the data presented is how much you trust or don’t trust the media, how you believe certain people and don’t believe other (“I just can’t tell if they’re lying”) and how we don’t know the average level of female education and entrepreneurship in the third world.
Good data is really, really valuable. In this piece, “Good” means accurate and trustworthy, not “desired.”
An AI trained with bad data is intelligent, but with bad reference points. So, if the data is bad, the predictions, categorizations, and outcomes will also be bad. The algorithm thinks it’s doing everything right (and it is) but the data set it up for failure.
Good data is expensive because it requires lots of work. Lots of work requires lots of labor. Lots of labor means large expense. This typically bothers CFOs.
The issue is that you either pay now or pay later. Either, you pay now and train the AI properly, or you pay later and have humans doing work algorithms could do. Training the AI properly enables easier and more flexible scaling, and you don’t need to retrain unless you need new data included in predictions. You’ll always pay hourly for that labor.
So, what’s the point?
AI will permeate every industry and every business. Quickly. It won’t be like “Oh my business can wait a few years” or “Technology doesn’t really affect my business” because competitors will implement it, lower prices immediately (while keeping the same margin), and undercut.
Or other startups will use AI to automate most or all of your business and ship a product of the same quality for half the cost.
Maybe they’ll ship twice as much product for half the cost per unit.
AI removes the human component of decision making and repeatable tasks.
Those tasks are only as repeatable, and the decision making only as strong, as the data you give it.
You’ll pay now or you’ll pay later, isn’t it worth it to invest the extra 20% into training the algorithm the right way the first time?