I am a data science engineer and a part of my job involves training machines to perform various human-like tasks. This belongs to a branch of Artifical Intelligence (AI) called machine learning.
Machine Learning deals with presenting a variety of information about a task to a machine and making it learn all the patterns from it. We define the success of the learning by measuring the machine's ability to correctly predict the outcome on data it has never seen before. This is called the model accuracy.
I have trained machines to identify gas leaks, detect anomalies like fire and smoke and much more through real-time CCTV footage. They aren't 100% accurate ( the system that I've trained has achieved around 96% accuracy), meaning, they do give out wrong predictions sometimes but they still do a very good job of identifying the anomalies.
"Does that mean it can now think like humans in identifying anomalies?"
This is the main question that I want to address.
These AI systems are masters at generalizing a given set of information and extracting patterns that humans can't see or comprehend. This is possible due to a complex set of algorithms called neural networks and the availability of nearly unlimited computing power to run them. The output that we get from training are specialized models capable of achieving a near-human level of accuracy.
Though it might sound intimidating to some, you must understand that this isn't actually any intelligence that the machine develops. It's just a representation of what has been presented to it during the training process. There is surely some room for robustness but it is nowhere close to what humans can achieve in terms of the diversified learning.
For example, if I train my system on images of fire and smoke and there is a gas leak that occurs, the system will not detect it to be an anomaly. Even without any prior knowledge on all the anomalies, we humans, have the capability to perceive them as a kind of threat and take the necessary measures. However, these machines CANT.
As of today, Artificial Intelligence comprises thousands of machines each trying to do a very specific task. Be it Google's Alpha Go or Tesla's self-driving cars all are machines capable of being better than humans, but only for that very specific task.
We are far away from having a single machine that can think or emulate thinking patterns. AI as a term has caught on only because of businesses that use it to lure customers. Right now, AI is used as a marketing term just like how Big Data, and Cloud Computing were used in the past.
Replication of Human Thought Patterns is possible, but we are light years away from it. We can discuss more on this in my next article. So, stay tuned!
If you are interested to know the technical details about machine learning, then I will recommend you to read more about it on What is Machine Learning? A simplified Understanding
