Demystifying AI and ML
We all know JARVIS — Tony Stark’s virtual buddy — from the Marvel Iron Man series. Have you ever wondered how Tony and JARVIS are able to interact with each other? Well, Jarvis is an advanced AI system that utilizes Machine Learning to communicate with Tony. In this multi-part series, we’re going to dive into the mechanics of Machine Learning to demystify JARVIS or any ML software.
Arthur Samuel, a pioneer in Artificial Intelligence, describes Machine Learning (ML) as a field of study that gives computers the ability to learn without being explicitly programmed.” In other words, Machine Learning is a technology that enables computers to learn cognitive skills like speech comprehension and image recognition.
The key component of ML is software that represents a mathematical function that uses input parameters to give an answer or make a prediction. Development of the software involves a training and a deployment phase. For instance, take an example of software that can take a digitized image of a plant and recognize the plant by its name. During the training phase, the software is fed various digitized images of plants with corresponding names, which are used to formulate and refine a mathematical function for plant identification. During the deployment phase, the software will use the refined mathematical function to identify a plant from a digitized image. This is very similar to how humans learn new cognitive skills based on prior experiences and can expand their knowledge based on new ones. Don’t ponder on this too much- we’ll discuss this in a subsequent post in the series. Now, let’s explore another example of ML: Amazon Alexa.
Everyone loves Alexa’s clever jokes, vast knowledge, and her ability to perform NLP (wait what’s NLP?!). NLP is an acronym for Natural Language Processing, which is a computer’s ability to understand human speech/written text as well as its intent. Alexa has the capacity to do many things because of her vast library of skills. Users can ask her for recommendations, recipes, information, music, jokes, and more.
Let’s break down the working of Alexa. First, an array of microphones embedded in the device are used to detect users’ voices. Once the “wake word” is detected, the signal is sent to the Cloud where speech recognition software can convert signals from audio to text. NLP software will then interpret the text and determine what the user requested. Let’s take a closer look at the fundamentals of the inner workings of the Cloud.
During the training phase, the program must be given a diverse set of voice input with different accents and sentences to ensure that in the application phase, the software can understand and interpret a wide array of users globally. The main element of Alexa is it can constantly learn and improve itself. If Alexa makes a mistake in the translation and processing of a user’s request, the user will typically correct it by interjecting and rewording their request. This response will automatically register with Alexa so she can apply this in the future.
ML programs are also being deployed in the field of utilities. California is notorious for its widespread and destructive wildfires. As the Earth becomes warmer, wildfires are becoming increasingly more common outside of the known “fire season”. The San Diego Gas & Electric company (SDG&E) formed a solution that would aid in locating and prioritizing equipment problems- especially those in areas with a high-rise of wildfires. The solution? Drones with ML capability to detect and prioritize equipment problems.
On a typical schedule, maintenance crews must do a lot of manual labor which includes, inspecting miles of power lines or poles and logging those notes into their system. They also rank the fixes on urgency and prioritize the most critical and vulnerable cases. This schedule does not include emergency repairs that result from unexpected situations.
There simply isn’t enough manpower to continuously check the systems. Engineers proposed a solution that used drones with cameras to verify whether the equipment was stable and in good working condition, especially in unsafe districts that are at risk of wildfires. The drones would use machine learning models (image recognition software) to identify the type of utility equipment as well as rate its condition. With the score, workers could determine which equipment was in most dire need of their attention without being present in those locations themselves.
However, like all ML programs — there’s a learning process. The program had to be shown labeled and tagged data (2.3 million photos!) and SDG&E’s own employees used their own prior knowledge in the labeling and classification of the photos. This allows the model to be more accurate with expert knowledge. Of course, if the model is ever unsure- it will send the photo to a human supervisor to ensure that the images are being categorized correctly.
Most assume that AI is a new concept but, it is a notion that has long been thought of since the 1950s. Alan Turing, a young mathematician and computer scientist began to contemplate the thought of computers having similar decision-making skills to humans. Turing developed a method known as the Turing test which can be used to determine whether computers had responses like humans. Fast forward seventy years later, and AI and ML are seen almost everywhere. This is due to the increase in computing power, as well as the shift to GPUs.
Graphics Processing Units (GPUs) allow for parallel computing to occur at a faster rate. Essentially, parallel computing is multiple computations or algorithms that have the capability to run simultaneously. Because Machine Learning algorithms usually need to analyze extensive data and perform computations — this creates the need for larger processing power or GPUs to speed the rate at which this evaluation occurs.
AI can be applied to a multitude of industries and modernize the way these industries function. Andrew Ng, founder of the Google Brain lab and a pioneer in Artificial Intelligence research, describes this rapid spread as AI being the new electricity. Now, this should have given you a decent overview of Machine Learning. In this series, we’ll be exploring different ML subsets including supervised, unsupervised, deep, and reinforced learning. I’ll see you next time when we’ll be discussing supervised learning.
References
Esposito Professor of Business & Economics at Harvard University and Grenoble École de Management, Mark, et al. “What Is Machine Learning?” The Conversation, 9 May 2022, https://theconversation.com/what-is-machine-learning 76759#:~:text=In%201959%2C%20Arthur%20Samuel%2C%20a,or%20to%20make%20accurate%20predictions.
“The Machine Learning Behind Alexa’s AI Systems .” Performance by Sohan Mishra , YouTube, YouTube, 24 Sept. 2018, https://www.youtube.com/watch?v=Dkg1ULBASNA. Accessed 23 Aug. 2022.
Staff, Amazon. “How Machine Learning and Drones Are Helping Prevent Wildfires.” US About Amazon, US About Amazon, 17 May 2022, https://www.aboutamazon.com/news/aws/how-machine-learning-and-drones-are-helping-prevent-wildfires.
Wikipedia. “Alan Turing.” Wikipedia, Wikimedia Foundation, 16 Aug. 2022, https://en.wikipedia.org/wiki/Alan_Turing.