Side Of AI
Many people labor over the topic of AI; what it is, what it isn’t, how amazing or how dangerous it is to our future. For many of us, it’s all sorts of confusing. But I can tell you with certainty you already know a ton about AI without ever having studied it, and it has a lot to do with your ability to find Waldo, or debate was it Dylan McDermott vs Dermot Mulroney (funny SNL sketch).
In this “Side Of AI” post, I’ll give you just a taste of this massive topic by familiarizing you with things you already know (you’re welcome), while explaining how that stuff is fundamentally what AI is all about.
So, do you think Zoe Daschenel looks like Kay Perry? A lot of people do. The reason is, humans are very good at facial recognition. Actually, we are amazing at pattern recognition, overall. Your common 500 piece puzzle is one big pattern recognition test, and we love doing it because we are pattern recognizing machines. If you get anything out of this post, it’s that AI is essentially based on what your brain is doing when it’s doing these kinds of tasks.
But let’s shift away from pop culture and talk about everyone’s favorite topic, statistics. Don’t get too bored, you know statistics too. Anytime you’ve calculated the average or mean of some group of numbers, you are doing statistics, and essentially doing something very similar to AI. But a key difference is the scale or perhaps I should say the speed of the statistics performed.
To explain I’ll create a story, about you, our reader entering the Twilight Zone: Witness you, our reader, a hapless character thrown into our story about an on the spot statistician. A numbers person who dutifully observes demographics on unknown souls who move to new townships. An informational warrior who’s sword is a calculator and creed is to maintain focus on data about schools, crime, affordability, and shifting political landscapes. However, you are shifted into a strange existence, untethered by time and relationships. For a 1000 lifetimes, our loyal statistician observes civilizations come and go, with people settling into new townships over and over again. And then, our statistician is back to our time, but with the ageless knowledge of a 1000 lifetimes of observation, highly trained on predicting where people will choose to live.
This bizarre scenario is the idea behind training AI to derive new information out of complex amounts of data. AI frameworks use things called neural networks. These are models for analyzing massive amounts of data over and over again using common statistical formulas like linear regression. But it’s missing intuition, fuzzy logic, interconnected memory, and experiences. To make up for actual intelligence, we analyze and run calculations against the data in far more permutations than our brains ever would… at the scale of lifetimes. It’s a brute force method to find patterns, clusters, and correlations in the data.
AI is only mimicking ourselves and doing a very poor job at it. The intelligence is simulated and a bit of an illusion like a common card trick. Every AI application starts with what humans know. What we call a cat vs a dog or a car vs a truck or clean vs messy hair. It’s all human biased data to start with, then we force the data to go through many more iterations and calculations than we ever would. An AI is only as good as the data it’s analyzing. We are the source of all AI.
As implied AI is not very intelligent. It takes a child very little time and information to recognize family and friends. Certainly, children do not need to perform recursive statistical analysis on every pixel in a million photos of their parents.
Now the spooky stuff. The AI algorithm created after it’s been trained on data, can’t be understood by us. Massively powerful machines pour over data for many hours or days and create a sequence of instructions and calculations we just can’t unravel. And there is the potential for new correlations and patterns we didn’t anticipate, expect or even want, hiding in the instructions. We may never know this until we inference or run the AI in a certain scenario. This is the thing Elon Musk famously warns about. What if an AI learns, correlates, or defines a pattern to deceive and harm humans?
However, fear not, most experienced in AI see this is as a misplaced projection of human intelligence and emotion thrust upon AI. AI-powered software is not self-aware and has little understanding of our human world. AI’s don’t care, they don’t have ambition or desires, they are just a series of instruction assembled from patterns in data.
How those patters are used is up to us. AI is amazingly powerful and now highly accessible. It will do SO much for us. It’s an exciting time for technology.
An example of this is an AI application created by a friend of mine, Peter Ma. This application is called Clean Water AI. For background, Peter is the world’s most prolific hackathon winner. He is a very clever and creative coder but he’s not a data scientist. He’s taken commonly available technology and with sharp insight and vision, he builds fantastic prototypes.
His Clean Water AI operates similar to a game of “what celebrity looks like Katy Perry”. When we look at her verses another celeb, we intuitively recognizing facial features that follow a pattern from a history of seeing so many faces. Clean Water AI is doing the same thing. This AI was trained to look at microscopic pictures of clean water vs water with bio-contaminants. Peter put a bunch of these images through an AI Framework which popped out an algorithm that could determine if any new image of water is clean or contaminated. And this algorithm can then be run on a low powered laptop with USB powered microscope/camera and nueral compute stick from Intel. Just a few hundred dollars of off the shelf hardware and you have a means for anyone around the world to know if their water source is clean. It’s also portable enough so you can go upstream of any water source and find a likely source of contamination. First world tech solving third world problems.
I see tons of this stuff from developers all over the world. Risab Biswas here created a plant disease detector using a similar method; give the AI framework images of diseased plans vs healthy plants and out pops an algorithm for detecting diseased plans. He’s not a renowned hackathon winner. He’s an indie developer from Kolkata India. And this solution works using a standard smartphone allowing farmers to have untrained labor or even drones checking for plant disease.
Each day news stories are published explaining how AI is solving problems as well as, if not better than, humans. From breast cancer detection to free parking space alerts, AI has the power to shink time, distance and access to information that usually requires the time and attention of a skilled human.
There are many flavors of AI from Machine Learning (ie Netflix recommendations), Reinforcement Learning (virtual bots that evolve and learn to beat humans in StarCraft), Deep Learning (autonomous vehicles knowing how to stay in a lane) and style transfer techniques like GANs responsible for DeepFakes and face-swapping software. There is much going on with AI and further topics to explore.
So next time you are looking for Waldo, or wondering who that celebrity reminds you of, know you are inferencing a trained pattern. You are acting like AI…less the A.
