Page 17 - ACTL Journal_Sum24

Assuming that not all program attendees came with the same level of AI understanding, Professor Grossman began with a ten-minute overview that thoroughly demystified AI algorithms, machine learning, natural language processing and generative AI. To paraphrase (as this article will do for all of the panel): So what is artificial intelligence? Well, it’s an umbrella term that was first used at a conference in Dartmouth in 1956. And basically what it meant was that computers were doing intelligent things, performing what we would normally consider cognitive tasks that require thought, reasoning, judgment, and so on and so forth, things we once thought to be the sole province of humans.
So it’s not any one particular technology or one particular function; it’s whatever a computer can’t do until it can. And then, once we get used to it we just call it software. So, if you think back, things like spam filters were sort of mystical and magical when first introduced. Now we don’t give them too much of a thought; they’re just software. That’s all AI is. It’s slightly different than automation and robotics. Automation is something that a human once did that now a machine does but it doesn’t involve that cognitive component. So a wash- ing machine is automation; a washing machine that separates whites and darks is AI.
Robotics is the hardware end. You can have robotics with AI or without. Robotics without AI might be a surgeon using robotic arms to perform a surgery. With AI, the machine de- cides where to cut.
AI is about algorithms, machine learning, and natural lan- guage processing.
In supervised learning, suppose we want to teach a comput- er to distinguish between pictures of puppies and kitties. We don’t sit down and write all the characteristics of puppies and kitties. Instead, we give the computer labeled examples: This is a puppy. This is a kitty. This is a puppy. This is a kitty. The machine becomes able to extract what the characteristics of puppies are versus what the characteristics of kitties are and it becomes able, when given an unlabeled example, to figure out whether something is more puppy-like or kitty-like.
Reinforcement learning allows us to do more complex things when we have a dynamic problem. Reinforcement learning is a separate concept that combines exploration, i.e. going into new data, with exploitation, i.e. returning to old data, and going a little deeper. The computer is told whether it gets things right or wrong and, from that, learns how to handle dynamic problems.
Deep learning is the final frontier for AI. Deep learning is a whole bunch of algorithms stacked on top of each other, sort of like a stack of pancakes and each of the algorithms is made up to look or seem like a little bit of a brain. And each of those algorithms is doing something different. At the top level a pre- diction is made.
So what does deep learning allow us to do? It allows us to do much more complicated tasks. Think about an autonomous vehicle. We have lidar, radar, sonar, GPS, photography, weath-
SUMMER 2024 JOURNAL 16
Algorithms are like a recipe to bake a cake; a computer algorithm is simply a bunch of steps to have a computer do something, nothing fancier than that.