Mark and Huckle are quite the pair. For those of you who have not met them, Mark is one of Futurety’s data scientists and Huckle is our resident machine learning agent, or as I like to call him “super awesome fancy computer”.
Mark works hard every day training Huckle on behalf of our clients so those clients can use machine learning and artificial intelligence in their organizations.
For example, Mark could show Huckle thousands of images and x-rays of healthy shoulders and teach Huckle that these are “healthy shoulders”. Then he can show Huckle thousands of bad shoulders and teach Huckle that these are “bad shoulders”. Pretty soon, Huckle can tell healthy shoulders from bad shoulders, even if he’s never seen that particular x-ray and Mark is so proud and he’s not crying he just has something in his eye.
There are several other ways of training Huckle. Huckle can analyze millions of emails and get the equivalent of a catnip treat from Mark when an email recipient opens an email. Huckle can get a big ol’ belly scratch when a recipient makes a purchase after clicking a link in an email.
Finally, Mark can feed Huckle a bunch of information and let Huckle figure out an answer. For example, if a real estate developer is trying to decide whether to build luxury homes, family homes or retirement housing in various zip codes, Huckle can cluster information like age, income, and marital status to help determine what homes to build and where.
One limitation with all of this is that Mark is just one guy. And even with 100 Marks, there is a limit to how much Huckle can be trained.
What if the very consumers of your product also become its trainers? What if you could have real people, in real-world scenarios constantly training your product?
Welcome to the age of Consumer Trained Algorithms. Since 2015, Google has talked about Rank Brain, an AI platform that makes search results more relevant. Does Google do anything as tacky as survey us on how happy we are with our search results? No way. However, when we click a result, and spend 2 minutes on the site and make a purchase, we’ve just reinforced a good experience. When we click a link, spend three seconds on the site and change our search terms, we have likewise trained the algorithm. Multiply that by a billion searches a day and we can see how Google can start to think like a person.
It’s been said that when one Tesla car learns something, all Teslas learn it. We contend that Tesla isn’t a car company. It’s not a battery company, it’s a custom neural network… essentially a giant brain of the world’s infrastructure.
It’s also been said that Tesla loses around $15,000 per car sold. Flip this thought and think more along the lines of “Tesla is paying us $15,000 each to map the roads and highways you drive.” Ultimately, we are the Mark to Tesla’s Huckle. We train the algorithm.
Facebook, Apple, you name it, they can all be thought to be doing the same thing.
I happen to fall on the side of the automation debate that believes automation is a good thing. Automation will empower us to solve more of the world’s problems and we only create a better user experience for ourselves and for others when we train technology to adapt to human needs. We don’t pay a monetary fee to use Google or Facebook. We pay less for a Tesla than we should because in exchange we help develop the technology.