What is a No-Code AI Lab?
No-Code AI LiveLab is where we solve real business problems with real data at Business School of AI community.
In this post I want to tell you about what is No-Code AI. If you are curious about the lab, you can read our weekly lessons from our Capstone AI Lab teams here. If there is enough interest, we can do a series to explore this topic further.
What is No-Code AI?
No-Code AI is the latest trend in AutoML platform based tools (Automated Machine Learning) that allows you to build an AI Model with zero coding.
No-Code AI is not about the tools to automate Machine Learning. It started out that way as Data Scientists needed AutoML to get out of the mundane work of fitting the same data to multiple classification algorithms and then hyper-parameter tuning a set of parameters to arrive at an optimal model for a simple binary classification algorithm.
Lets take a step back. Let us get to the same starting point so we can move forward to understand No-Code AI. And then we will get to No-Code AI Lab.
The most common problems solved by Artificial Intelligence in industry falls under 4 category.
- They organize data and find patterns.
- They identify objects.
- They make predictions and recommendations.
- They find an anomaly or call out the odd pattern.
To do this, all data is trained using historic data that is believed to have the good and bad results of past behaviors. For example, if you want to understand what causes customer churn, you need data from customers who churned (and canceled or dropped out of a product or service)and from those that did not churn (and continued on as customers) in the same dataset.
I want to talk Classification algorithms which are used widely for situations where we know that the answer is a discrete value, from potential answers that we know already. We do not know if the customer will churn or not, but we know that the answer is either a Yes or No. When we try to find which is the best marketing channel that a particular customer will react best on an offer, we may not know the answer, but we know it is either newsletter, event, ad, or a sales call.
You can skip this paragraph if you do not want to know how the brains of classification algorithms work. The algorithm fits a line through the data to form a shape that it can interpret using linear algebra. It uses it to organize the training data to see what data points fit on the line and which ones are not on the line. The complexity comes because there are several different algorithms to create this best fit line. Each has its advantage and challenges. It is tedious for the data scientist to fit the data to multiple algorithms and that is how AutoML (Automated Machine Learning) was born.
The training data needs to be organized into classes or buckets for the algorithm to make sense of it. The most common example we see of this (never in real life but in many ML courses) is to identity fruits as apples or oranges as two classes. AI is a narrow AI which means that its world is limited to fruits being either apples or oranges because that is all that the training data taught it.
That is why when I showed my face as a test image in a binary classification model, it categorized me as an dog instead of a mop and when I wore by glasses, it became 100% certain that I was a dog. (see screenshot from TeachWithGoogle AutoML aka No-Code AI Computer Vision model).
Classification algorithms can classify between two classes, then it is binary classification. If it classifies among more than two classes, it is a multi-class classification. (Technically there are 4 types of classifications. You can learn it from Jason BrownLee at Machine Learning Mastery — IF you know Python).
If you are NOT into Python, No-Code AI might be for you as a business tool. Read on.
The number of classes is a trivial detail for an algorithm that can deal with huge volume of data and see patterns and predicts with a high degree of confidence which pattern or class will help with revenues, sales, reducing costs, predicting prices, keep customers happy, saves lives and so on.
Can a Business Person Build an AI Model?
Yes. A radiologist named Erwin John T. Carpio, MD, from the Philippines used the teachablemachine withgoogle and built an Xray image diagnostics ML model to augment his work. (details here)
My student, Peace Technologist, Aishatu Gwadabe, took our AI class exercise and exported the model to flutter to create an image recognition App on her phone.
If we expand that to look at voice or raw numbers data, the possibilities are endless with No-Code AI.
This No-Code AI webinar is a sneak peak to an No-Code AI: Data Science for Business User livelab course where anyone can learn to solve business problems using AI. We will pick 3 business problems and solve them hands-on in small cohorts of 10 people within 100 days.
Click here to get the No-Code AI Webinar recording (free). Feel feel to download the curriculum for the No-Code AI course here.
We are just getting started with the Business School of AI blog. Tell us what you want to learn about. Tell us if you have experience with No-Code AI platform? I use several No-Code AI platforms to stay brand agnostic in my classes online. Each has it strength and weakness and caters to different kind of business problems, different kind of data sets and different levels of complexity in bringing AI to your company.
Until next time — Sudha