How NLP and Artificial Intelligence can help in defeating COVID-19

There's a lot of confusion with the term AI. Artificial Intelligence is a fancy word used by marketers to associate products with cutting edge technology, which in most cases is just an algorithm.

Algorithm or Artificial Intelligence?

So what's the difference between an algorithm and AI?

A computer system that works with structured data is an algorithm.

For example, when you are shopping with Amazon, the algorithm gives you product suggestions based on the history of the previous user's

preferences recorded in a structured database.

On the contrary, AI is a system that has the capability of analyzing data as is.

For example, autonomous cars are controlled by Artificial Intelligence.

AI analyzes the current road situation (unstructured data) and gives commands to make relevant reactions (break, accelerate, turn to the left/right).

Natural Language Processing is a communicational AI. A system that can interpret human speech - raw data, that can not be analyzed by a simple computer algorithm.

In general, computers can't analyze raw data. Most living creatures (not to say humans) are better at seeing and examining real-life pictures than the most powerful supercomputers in the world.

On the other side, machines are out of competition at computing, and we all need them to perform our everyday work.

AI combines the best parts of both worlds.

But how you teach computers to analyze unstructured data?

There's where Machine Learning comes in. ML and human AI-trainer help the computer to interpret raw data into commands.

For example, we are training COVID-19 chatbot to understand commands like "book me a call with the doctor" or "check my medical state". There can be hundreds of ways to say that phrase. But to make AI understand the command, you don't need to have all possible phrases in the database. Instead, human AI-trainer adds training phrases to this command (intent) in Dialogflow, and when you get at least 20 of them, ML starts working. It finds any mathematical patterns for those training phrases, like the number of symbols, nouns, spaces, and thousands of other parameters, so when the 21st phrase with a similar mathematical model comes in, it understands it as a particular command. The more training phrases per command you have, the more accurate is the result.

In COVID-19 chatbot, we have more than 1500 training phrases, so it has 84% success rate.

Why we developed COVID-19 Chatbot

According to the World Health Organisation (WHO), in most cases, COVID-19 infected person doesn't need any specific medical treatment. Only people with fever, cough, and difficulty breathing should seek medical attention. Older persons and persons with pre-existing medical conditions (such as high blood pressure, heart disease, lung disease, cancer, or diabetes) appear to develop severe illness more often than others and need special attention.

Imagine an infectious diseases hospital or a doctor who helps people during the pandemic. They have to handle an enormous number of routine communications to qualify incoming requests, and schedule calls ONLY with patients that need urgent help. With our chatbot, they can automate qualification and appointment scheduling processes.

Please note that the bot does not make a diagnosis. It only handles routine communications and passes data to doctors by email and schedules calls in Google calendar.

Of course, we will give this chatbot to hospitals at no cost as our input in fight with global pandemic.

12 views0 comments