Notes from Tuesday, October 2, 2018
Yesterday was going to be pretty epic. It was the first day of trying to explain the world of artificial intelligence, machine learning, deep learning, and neural networks to people in a meaningful way. Yeah, I went back and read the prose from yesterday. It was a massive false start. In some ways that is sort of how artificial intelligence ended up changing things. Sometimes you end up trying to align the solution you have to real world problems. Other times you just face a really big challenge and figuring out how to do it requires thinking outside the box. A few of those examples might be pretty interesting to think about. Our friends at Google really wanted to organize information and figuring out how to crawl in internet was something that ended up being a definable and repeatable process. It is something that can be done and the steps are clear. Extending that to searching photographs, videos, and creating an accurate knowledge graph involves solving very different types of problems.
One of my favorite examples over the years has been how image classification systems including the one being used by Google Photos cannot tell if Peppercorn the Dog is in fact a dog or a cat. My guess is that most of the time if you had a reasonably good photograph you can successfully categorize the pet as a dog or a cat. A lot of my photographs are labeled Peppercorn the dog… that context clue should be enough to help classify my 13 year old Australian cattle dog as well a dog. I went through the trouble of helping the algorithm tag and classify Peppercorn as a pet in Google Photos. Strangely enough that did help the algorithm learn. Translating that categorization problem to understanding classifiers can help you figure out how hard it is to sort the world around us based on a given set of heuristics. Trying to setup a framework to figure out what are all the things in a photograph is really interesting. That is a challenge that a technology like Tensorflow could help with. Software like that helps by putting together a method to classify and identify things. Sure the way I go about that is based on the things I know and can figure out.
Solving really hard problems like setting a self driving car or sorting my 10,000 digital photographs into albums take a certain type of technology. Finding the right technology solutions to solve really hard problems is where the fields of artificial intelligence, machine learning, deep learning, and neural networks are getting really interesting. That is really and understatement. My interest in the convergence of technology and modernity has been growing for years. Sometimes I spend hours just thinking about the intersection of technology and modernity. Perhaps that is why these inquiries into artificial intelligence are so timely.