Chatbots and natural language processing

One of mankind’s old dreams is to artificially clone him or herself in a machine. The mechanical Turk was the first attempt of this anthropomorphism of the machine, while being a scam, but that’s another story. Artifical Intelligence progresses inexorably as it digests data and offers many anthropomorphic features, including the understanding of a text, its translation and the response it can provide. Understanding a text is not easy. It is enough to see how long it takes a normally constituted human being to understand and analyze a text, even simple. These are years of exposure to a language, then learning its rules and finally the ability to analyze context and nuances. I give some examples in my book (in French) Artificial Intelligence with AWS.

However, one of the beauties of AI is its ability to learn quickly and well, especially since it has an important body of data. When it comes to language and text, man produces petabytes every day, so that’s not what’s missing. This is how the big industry players, the GAFAM and the BAT, have developed some pretty stunning natural language processing algorithms.

The truth is in the volume

For a supervised learning algorithm to work, it needs data, a lot of data. However, a lot remains vague… tens of thousands of data points at the mimimum, ideally several million or even billions. This is one of the reasons why the tenors of the web are also those of AI. Since all these years we have been providing them with our data for free, usually without our knowledge (it’s also another story), they have had time to discover lots of interesting information that they have hastened to “monetize”.

The fact is that without a massive volume of data, there is no AI capable way of supervised learning. A text follows rules, grammar and spelling, which can be codified and digitally put into an AI algorithm. After a few more or less long, complex and complete learning passages, our AI becomes able to understand what is being said and provide an answer or interaction.

What was the question?

Ask a search engine a question, they will interpret your request and provide you with a set of answers, the most relevant of which will appear at the top. The notion of relevance is unfortunately today often perverted by economic contingencies (the famous sponsored links or targeted advertisements), but the general idea is to provide a relevant answer, period.

Add a specific glossary to your favorite subject and our AI will complement its vocabulary with the words and idioms that will make it more relevant. Inject him with an ability to detect emotions and you have an ersatz of humanity. All that remains is to add translation capabilities and you get the basis of a conversational robot. Lex at Amazon, Azure Bot Service at Microsoft or Watson Assistant at IBM are services that allow you to quickly and simply create conversational robots. We can talk about Conversation as a Service (CaaS), with which is sufficient to inject a little contextual intelligence to create an interface that bluffs more than one.

The art of being a plumber

If the creation of the bot is a quick and simple act, what will make it powerful is its interfacing with other systems. We then move from conversation to action. A question or search can result in an order that will trigger a payment and delivery. The ability to program actions without worrying about the underlying infrastructure, thank you serverless, allows to multiply the power of speech. AWS Lambda or Azure Functions are one of those ways to interact with developers.

The developer (full stack or not) then becomes a plumber, who makes pipes join by adapting custom sleeves. So, with a few clicks and lines of code, we join user-friendly interface in natural language and CRM, ERP or other database. If it sounds like magic, check out and test the examples of conversational robots posted online by the aforementioned companies.

Ultimate beauty, enhanced learning mechanisms improve systems over time. Thus, we have a bot that continues to learn, works 24 hours a day, 7 days a week, is not unionized and never goes on strike. Dream or nightmare, this is a reality that is difficult to ignore and should be used. As with any technological advance, if you don’t, your competitor will and for you, it will probably be too late.

Photo by NeONBRAND on Unsplash

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Proudly powered by WordPress | Theme: Baskerville 2 by Anders Noren.

Up ↑

%d bloggers like this: