Machine learning and artificial intelligence principles are rapidly changing how computing works. One of the key ways that this is happening is with weighted or probabilistic inputs that change inputs from a truly deterministic system into something more abstract.
In artificial neural networks, individual neurons or units receive probabilistic inputs. They then make a determination as to the output or result. This is what professionals are talking about when they talk about replacing the old world of programming with a new world of “training” or “teaching” computers.
Traditionally, the default was to use programming to get computing results. Programming is a fixed set of deterministic inputs – rules that the computer will loyally follow.
By contrast, allowing for probabilistic inputs is an abstraction of these rules, a kind of “slackening of the reins” to free up the computer to make more advanced decisions. In a way, the probabilistic inputs are unknowable from an outside perspective and not predetermined. This is closer to the way our actual brains work, and that’s why machine learning and artificial intelligence algorithms using this approach are being hailed as the next frontier of artificial cognitive development.
Here’s an easy way to think about weighted or probabilistic inputs. In traditional programming, you had the type of “if/then” statement that generally says: if THIS, then THAT.
Moving beyond the rule-based approach involves changing what THIS is. In a rule-based approach, THIS is some text input or rule: If you think of it as a binary – we know if it’s true or not, and so does the computer. So you can predict the computer’s response to any given input.
In the new approach, THIS is actually a collection of input that may be in any given state. So since an outside observer wouldn’t easily be able to model what THIS consists of, he or she couldn’t accurately predict what THAT result might be.
Think about this principle applied to all sorts of fields and industries, from market segmentation to financial verification to entertainment to water and sewer management, and you have the real power of machine learning, deep learning and artificial intelligence to direct human affairs in a very new way. For example, in the field of fraud management, experts point out that rules-only systems are not very good at figuring out the difference between suspicious or risky behavior and normal behavior – machine learning systems armed with sophisticated input models are more capable of making decisions about what activity might be questionable.
Another way to think of it is that the world went through an era of identifying code as a new frontier for learning and decision-making. In and of itself, deterministic code-based outcomes were powerful in terms of modeling all sorts of human activity and decisions. We applied all of these ideas to marketing, sales, public administration, etc. But now, experts are talking about “the end of coding,” as in this very insightful and instructive piece in Wired. The idea predominant here is the same idea, that in the next era, instead of coding, we’ll have a system where we train computers to think in ways that are closer to how we think, to be able to learn over time and make decisions accordingly. Much of this has been accomplished by moving from a deterministic computing approach to one that is abstracted with more sophisticated inputs.