The Difference Between Traditional Programming and Machine Learning
Video produced by Steve Nathans-Kelly
Everyday Chaos author David Weinberger discussed how machine learning and traditional programming approach and analyze data differently during his presentation at KMWorld Connect 2020.
In traditional programming a developer looks at what the factors are that affect the outcome that the program is attempting to predict, he explained. In the case of a business, it's what you would put on a spreadsheet and then the programmer implements in code the logic, the relationships among all of these different things, which can be as simple as if expenses go up, then revenue profits will come down or way more complex scenarios between releationships.
Whereas machine learning drops the logic, drops the relationships, and what we know about the relationship, he said.
"Machine learning starts off its natural state as a black box because it's not generally has not been designed in order for us to understand it," Weinberger said. "They are designed in order to produce accurate results. And it turns out that it does so in a way that we can understand great, if not, then we have other challenges. The, the natural state of a machine learning system is to be a black box. Doesn't mean we can't open it up in some least some ways in some circumstances, this is a problem because the machine learning machine learning systems are based upon data, data comes from a society of business, a culture, inevitably it reflects the biases in that culture or a business."
However, biases are machine learning's original sin, he explained.
"It's sort of its nature is to operate on data. That data is unless we're extraordinarily careful. It's very likely to reproduce the biases and machine learning is likely to reproduce and could, can easily amplify those biases," Weinberger said.