Tech Talk: AI Are Learning To Read Humans

Artificial intelligence these days is mostly powered by neural networks and machine learning, allowing its capabilities to be greatly expanded in a very rapid fashion, mostly by guiding an AI in what to learn, and letting it learn. One area where AI programs generally have trouble learning because of common factors being somewhat scarce on a specific level is human interaction. Every person is different, so every person acts different. Researchers are beginning to find common ground in the outward symptoms of depression, however, an ailment that is as complex as it is easy to hide. Despite depression's complexities and easy to hide nature, researchers may be on the verge of being able to create a machine learning algorithm to teach AI about depression and other such issues.

Their findings are based mostly on speech patterns. Things like vowel spacing, inflection, tone and color in casual conversation could give away a depression sufferer, and researchers are working hard at figuring out how exactly to feed an AI the symptoms that are known to be common among depression sufferers and tell it to figure out the rest. A paper on exactly how one might go about doing that was recently published by the IEEE. In the paper, along with the typical talk of how to handle the AI and how best to take advantage of the machine learning capabilities, researchers touch on some of the factors to look at in depressed patients. These factors, when handed over to an AI equipped for machine learning, could very well turn into a very human and very complex understanding of depression. This would, of course, lead to more early catching and treatment, easier access to necessary treatment for some patients, and other such marvels. This, however, is only the tip of the iceberg.

Neural networks, the backbone of most machine learning infrastructures, use a large amount of individual computing nodes, sometimes even virtual machines, all programmed slightly differently to mimic some of the properties of the human brain. These neural networks are agile, smart and, of course, highly scalable. This means that a machine learning algorithm running on an AI with a neural network behind it could not only learn what depression looks and sounds like and what to do if it's encountered, but could learn the same thing for many other human states, such as encountering somebody who is bubbly, or somebody who is very talkative. Essentially, the bots can learn to read both verbal and body language just as well as the most extensively trained psychologist, eventually.

If this comes to pass and the AIs also become capable of passing a Turing Test, where a participant chats with a robot and must figure out if it is human or not, we could very well be looking at an entirely new way of interacting with computers, if not the genesis of an entirely new breed of AI altogether. This new, super-powerful and super-smart human-friendly AI could analyze matters from a human perspective, but with the scope and scale of a machine. Suddenly, concepts like robots with feelings and an AI president are a bit less far-fetched.

This, of course, begs a few big questions. First off, just how advanced will these AI become? The answer to that, essentially, is infinite, for now. With breakthroughs being made in manufacturing and computer part creation just about every day, there is no real way to tell when the technology could plateau, or if it will at all. Another good question is what purpose the AI will serve for their creators; some, like Google, are in it mostly for the money. There are many independent researchers who create AI as well, some for money from selling the programs, some for the joy of it, and some to test the limits of the technology. For an AI to roll out on the kind of massive scale that would be required for an AI that can read a human's emotions to be anything but a gimmick, however, those three reasons simply won't cut it. Authorization for that sort of thing before an entity like the UN will have to be for the good of mankind, most likely in the public domain. Such AI would likely also have to submit to extensive testing and even rigid requirements, and that's news that not everybody involved in the AI field is going to want to hear.

You May Like These
More Like This:
Android Headlines We Are Hiring Apply Now