As I look to what clients are interested in and opportunities on the horizon for AI, immediately I see evolutionary vs. revolutionary uses to help us improve systems and organizations, save money and maybe even improve the quality of life.
One person I was talking to said for businesses, AI is more about “blocking and tackling.” And that might be an apt description if you like sports metaphors.
Many of the ideas we’re exploring start with “wouldn’t it be great if we could...”
Two such projects have emerged out of the Skapa AI laboratory that I want to share with you.
Recently, we had an opportunity to dig into an EdTech challenge. We wanted to know if the training and test questions some of our EdTech clients were developing were too easy—or too hard! And AI seemed the perfect solution to test the process.
It takes a lot of planning and work to make training software or apps easy to use and to inspire learning. But it’s difficult knowing up front if the questions created met the desired outcome.
The goal was to be able to tell in advance if a question was problematic and might need to be rewritten or reworked.
We identified a client with quite a track record of training and test questions - more than 25,000 questions. To analyze that much data ordinarily would be time and budget prohibitive. And yet with that much data, we knew we could develop a feature that would offer a high level of accuracy.
Once we organized the data set, which took some time with that much data, we identified a machine learning algorithm that worked for this use case, and then trained a model to evaluate questions based on the historic data.
The outcome is “Learner Select,” a feature for tests that allows us to predict the likelihood of multiple-choice answers to be picked, helping to guide the creation of meaningful questions.
Another innovative project sprung from the leadership of our client, Virginia 811. VA811 is a not-for-profit organization created by Virginia’s utilities to protect their underground facilities by managing requests to identify underground utility services before excavation takes place.
The challenge is that inexperienced residents and contractors who plan to dig often miss clear evidence of unmarked utilities. In fact, 36% of all damage to gas lines in Virginia is caused by missing “clear evidence.” Eliminating even half of those instances would be a huge win.
Given that, the leaders at VA811 asked if an AI-driven mobile app could be developed to detect and identify many “clear evidence” situations for homeowners and less experienced excavators. Scanning the dig site with a camera, taking pictures and allowing AI to point out possible clear evidence is intended to add a layer of safety and aid in compliance with Virginia laws, helping to ensure excavators don’t miss any evidence of an unmarked utility.
Skapa has produced a proof-of-concept app that is the start of VA811 realizing its vision. Powered by Microsoft’s Azure AI Custom Vision, our developers trained the model to recognize common red flag areas. If the location is deemed “clear evidence,” the user knows they need to re-evaluate the excavation location.
The app is in not fully tested yet, and VA811 has yet to decide if it will move forward with full development. If it opts to move ahead, Custom Vision might not have the fire power for the full-blown app. We envision a more development intensive effort, using PyTorch that is more customizable and requires more advanced developer skillsets. I’ll report more if VA811 moves forward.
If you’ve been thinking about how AI might help your organization or you have an app idea, I am happy to kick it around with you. There are many exciting developments, and we sure are energized by the possibilities.
