I've been slogging through this book since August. It was one of four that my coworker/friend purchased for me when I decided to teach the AI class. I don't usually stick with a book that long unless it's for a class. I tend to get antsy when I'm still reading the same book over a week. The slow read wasn't the book's fault. Nonfiction is just a slower read for me and I was tied up reading a lot of nonfiction in my classes. So I picked at it over four months.
The Algorithm isn't about AIs in general or the underlying theories of how they work. The Algorithm is specifically focused on how AIs are being used in the realm of business and human resources. The first parts of the book exposes how AIs function, how they are trained, and the limitations of the technology as it currently stands. Right now, the technology doesn't work very well at all. Any one with any kind of computer science understands the basics of AIs and complex algorithms. They are supremely cool but they are inherently buggy. An AI needs masses of data in order to learn, and the quality of the patterns it learns are reliant on the quality of that data, which introduces some issues when applied to the business world.
One of the big enticing ideas behind AI-assisted hiring is the thought that computers cannot be biased - which is true. A computer can't have an emotional reaction to an individual based on gender, race, religion, or disability. Everyone is aware that there has been a partially unconscious but wholly systemic bias towards various groups when hiring. Many tech firms are pushing AI assistance products saying that they will be more fair, and they really could be. However, what these tech companies neglect to mention is that the AI's are all being trained on data from the current work force. A work force that has been hired and promoted under an acknowledged-to-be-biased system. In examining this biased data set, the AI will, with all the things we want it to learn, also learn the biases and will perpetuate them. Of course, that assumes that the AI draws the conclusions we expect from the data. It often doesn't. So a running theme through the whole book is that as smart as AIs are, they are pretty stupid.
This is something that any one with a comp sci background already knows.
The problem is that most of the people in the business realm don't have a computer science background. Culturally, we love technology and the younger generations tend to trust it more inherently than they do other people. So they are buying into using these AI tools before the tools are really ready. In doing so they exacerbate an already problematic situation.
That's really the theme of the entire book: no matter how promising these tools may be, they just aren't ready yet.
Schellmann moves from hiring to internal and external monitoring. One of the most chilling things she notes is that there is absolutely no oversight or regulation on AI tools. It's all the wild west right now. Tech companies don't have to prove that their tools work or even submit to examination. It's a free for all and there is a lot of snake oil floating around.
As disturbing a picture as Schellmann presents about the current use of AI in business, the future of AI is hopeful. Properly regulated and tested, an AI assist can help us with many societal problems. But not yet.
No comments:
Post a Comment