Major Cost Factors Associated with AI
Video produced by Steve Nathans-Kelly
Access Innovations CEO Jay Ven Eman, Ph. D. identified four major cost factors involved in training and implementing AI systems during his presentation at KMWorld Connect 2020.
These major cost factors include: gathering lots and lots of data, normalizing and cleaning the data, training the system, and total cost of ownership, he explained.
"Some of you may have heard of OpenAI LP up in the Bay area of San Francisco. They have their GPT-3, which has a 175 billion parameters, which were derived from 300 billion words," he said. "Now, if you had to weigh 10% of those parameters, you're looking at weighing 17 billion, how are you going to do that? Another research outfit was looking at adding objective object learning into AI. And so they had 900,000 hand curated rules. If you change the topic or you change the focus, you got to rewrite all those sort of re rework them. It's not insignificant, but you can still do it and you can still get great results."
When it comes to training data, organizations need to remember the "garbage in, garabage out" rule.
"You'll always get results, but keep in mind, not all of those results are going to be meaningful. Not all of those results are going to provide you with insight," he said.