Can AI Actually Learn A Language By Itself?
Welcome back to our little corner of the internet! Today we’ll be investigating how an AI can apparently learn a language by itself! Let’s begin!
According to the McGill Newsroom website, researchers from McGill University, MIT, and Cornell University combined have taken a big step forward in the tech industry by developing an AI that can “learn the rules and patterns of human languages on its own” — which technically could allow it to have a sufficient understanding of a language.
If all that is said in this article is true, this would mean we have taken a significant stride in our understanding of Artificial Intelligence — and that we are getting to a stage where little to no human interference is required to allow AI to learn, comprehend and act.
How Does It Work?
This tech works through allowing the AI Model created by researchers to identify patterns in languages and in turn have the ability to state rules and grammatical concepts of that language (such as case, gender and declension). In order to increase the accuracy of how this works, the researchers used a machine learning technique known as Bayesian Program Learning (more on that later).
The AI also learns a language through related datasets which “discover” the interweaving of phonology (the study of sound patterns) and morphology (the study of word structure).
But what could this lead to?
You probably know that the simple fact that AI can learn things by itself is one that could in fact lead to its takeover (or at least, that’s what people like Elon Musk warn).
Maybe the popular theory that “machines could become so advanced that they could hack into essential systems such as power grids and financial systems, taking control of vital aspects of human society” could come true, knowing that AI could perhaps learn (using one-shot data that we’ve provided)
What a good sci-fi story that would make! Good thing we don’t have to worry about that for now, as AI researchers and futurists suggest that there’s only a 10% chance that a take over of any sorts will happen.
However, (to shine some positive light on the topic), as AI and tech like Quantum Computers become further developed, our lives could become a lot easier!
What is Bayesian Program Learning?
As promised before, I’m going to dive deeper into what Bayesian Program Learning is in this section.
According to The Data Scientist, Bayesian Program Learning is a solution to one-shot learning. One-shot learning is when a human, for example, only needs one instance or example to gain a fully-fledged understanding of a concept. I think they have given a good example to describe this:
You know that a child only needs to see an object fall once to know (in essence) that there is a force called gravity pulling objects down; and that if they drop something, it will fall downwards. AI learns similarly when it comes to Bayesian Program Learning.
Apparently, there is also evidence to suggest that there is a mechanism in the brain which becomes activated for one-shot learning.
The reason why this stands out to me is that Machine Learning concepts are actually known for not being able to generalise (or understand a general concept) from just one example. AI would have to see a lot of examples of an object falling to be able to understand gravity to the same extent (or further) as the child.
If you’re interested in looking further into Bayesian Program Learning, then I suggest you take a deep look into this. Please do leave a follow and a few claps behind if you enjoyed this! Thank you for reading this article, see you in the next one!
P.S. Thank you ever so much to my dedicated readers for over 40 followers within my first week!
