What is Machine Learning and How Google Assistant Uses It

Machine learning is pretty self explanatory. It is a process that we use to teach machines to recognize things. Here this ‘things’ can be a lots of things.

Machine Learning is a lot like human learning. We tell the machine that ‘something’ is ‘Something’. It analyzes the ‘something’ and saves the analysis with the result we provided, ‘Something’. The next time we tell the machine that ‘something else’ is also ‘Something’ the machine analyzes the ‘something else’ and also finds out the difference and similarity between the two things applying statistical reasoning and at last modifies the first analysis so that both the things can be matched with it to a certain degree. Now, we tell the machine that ‘another thing’ is ‘Another thing’ and it does the same thing it did at the first time, analyze and save analysis with the result provided. This is called training the machine so that it can recognize what is what.

Let’s talk about Google Assistant now. How does it understand what we say? How does it returns us the necessary information we were asking about? Why does it makes mistakes when we newly start to use it but get better at and make less mistakes as we continue to use it. It hears the sound of our speech, records it and tries to find out from its database if anyone else made this kind of sound before and what it means. It applies statistical reasoning to its search to find the best match and returns the meaning. Sometimes you will notice that when you start talking to the assistant it doesn’t recognize your words properly but as you keep talking to it corrects itself automatically. How does this happen? What happens is Google Assistant is already trained about sentence patterns. It can predict and resonate which words should be where in a sentence. It applies this ability and correctly recognizes your speech.

For example let’s say you asked “What can I have for lunch today?” and the Google Assistant matched your speech sound to “Vat can I have foe lunch today”. Why did the assistant failed with “what” and “for”? It could be because of a lot of things like pronunciation, noise, slurred speech, accent etc. But the assistant is intelligent, it knows what it heard doesn’t make any sense and it finds out that “vat” sound is close to “what” and “foe” is close to “for” and with this it can come up with the right sentence. It takes feedback from you whether it was right or wrong and trains itself.

The assistant also uses machine learning paired with AI to get the answer to the question it was asked. It uses Natural Language Processing (NLP) to find keywords from sentences and return results based on the keywords. This NLP is also made with help of Machine Learning to recognize which words are keyword in a sentence.

I hope I have been able to deliver you the concept of Machine Learning. I would love to receive feedback and answer any questions you have. Thanks.