Artificial intelligence is taking our world by storm. Through deep learning and neural networks it is already capable of many things we, human beings can do. Securing companies from cyber attacks is one example. Writing simple sport recaps and reports is another example. Through natural language processing, companies like IBM, Google and others now try to master what we humans are so good at: learning languages.
So easy, yet so difficult
Think about the last conversation you had. Either face-to-face, through a phone call or a messaging app. What was it about? Who did you speak to? Did you think a lot about what you wanted to say? How and what you say probably took a couple of seconds, but didn’t slow down the conversation. Using the right grammar didn’t make you hesitate and the context of the conversation was clear from the start.
See what I did there? I made you think about a skill that is natural for us human beings: understanding and speaking languages. Language is a natural thing that we don’t have to think about and that we can learn pretty easily. It is also a pretty big part of our culture. In fact ‘culture’ won’t exist the way we know it without any form of language. And yet, it is one of the most ambiguous things in our society. Academics still haven’t figured out the origin of languages and speaking languages mostly depends on context. This becomes very clear later in our lives. When we learn a second or third language.
Want to learn Spanish? Go to a small village in the Andes and try to survive for a month without speaking a world of English. This is called ‘immersion’ and it is how we learnt our mother tongue as well. Our human brain can process and acquire the many rules by listening to other people speaking that language. But still, we can not explain why we use a certain word in a certain context. When we get older, learning these rules gets harder. Learning an extra language besides your mother tongue isn’t that natural anymore and it demands many hours of studying to be able to order your meal in a restaurant during your holiday abroad. Choosing your words takes longer, it demands a lot of effort to understand people and you make mistakes, all the time.
When are you going to start talking about AI?
Learning a new language ourselves gives us a clear look in the struggle artificial intelligence has when trying to master languages and conversing with human beings. To understand this process, we, humans should, first of all, path ourselves on the back. We are the masters of language learning. Robots are even mimicking our brain trying to master languages. It is the last piece of land of human intelligence that artificial intelligence has yet to conquer. Winning the last battle won’t be easy. It is no longer merely about data processing, it is about context and common sense.
The field of study that focuses on the interactions between human language and computers is called Natural Language Processing, or NLP for short. It sits at the intersection of computer science, artificial intelligence, and computational linguistics. Google is the leading force behind this field of study and made a lot of progress already (more on that in a bit). Other leading organisations are the Stanford NLP Group, the University of Toronto and the University of Oxford.
Let’s take a look at what this ‘Natural language processing’ is all about. Basically it can be divided into two components: natural language understanding (NLU) and natural language generation (NLG). For both components the input and output can be written text or speech. NLU processes the input, NLG is charged with producing phrases and sentences in the form of natural language. This is where the ambiguity of language comes into play. And that on many levels: on a word level (is ‘fly’ used as a noun or as a verb), on a syntax level (there are 100s of ways to interpret the same sentence) and on a ‘reference’ level (what is related to whom, in a conversation).
Interested in the more technical stuff? Find more about it here
Google was one of the first to take a big step forward in the processing of languages. It has developed world’s most accurate parser (analysis of words in the sentence for grammar and arranging words in a manner that shows the relationship among the words) called SyntaxNet. This open source neural network framework is situated in the NLU part of natural language processing. Out of this framework, Google has developed a parser that can help you analyse the English language with 94% accuracy. Its name: Parsey McParseface. The most beautiful part is that Google has made this open source, kickstarting the many possibilities of language processing for developers (Think about text-based personal assistants in messaging app or Google’s Assistant in the Pixel phone).
Another notable development in the field of natural language processing is probabilistic programming languages. Although described by Noah D. Goodman and Andreas Stuhlmüller as a computer language that helps at representing and using uncertain knowledge. The implementations of this form of programming languages mostly lie in the processing of language. A great example of PPL is the Webppl language, which is capable of sensing hyperboles or puns in a conversation. Indeed, one day robots will be better at sensing sarcasm better than Sheldon from The Big Bang Theory does.
Back in 1966, MIT professor Joseph Weizenbaum developed a chatbot called ELIZA that acted like a cartoon psychotherapist. It repeated key parts of statements, encouraging its “client” to continue the conversation. It was part of a project to make natural language conversation possible with a computer. AI avant la lettre, you could say, as ELIZA was not capable of learning any new scripts. Those must be added in real-time to be able to continue the conversation. Weizenbaum wrote down is findings in the book ‘Computer Power and Human Reason: From Judgment to Calculation‘. In this book, He stressed the fact that a computer is not capable to grasp people’s emotions (yet).
Okay Google, what’s in it for me?
“This is all very well, but what it’s in for us?” you might ask yourself. Natural language processing is a pretty complex technology about a pretty commonplace thing. That’s why, you won’t sense this type of technology until something goes wrong. Siri does not understand your question, or your Amazon Echo plays the wrong song. That’s a difficult position to be in as an emerging field of study. But yet without us knowing it, it slowly is taking over small parts of our lives.
Inspired by the all-in-one-app approach of WeChat, Facebook is developing its ‘M’ personal assistant for Messenger. A bot will be capable of understanding what you are looking for and then (in the initial stage of the software) will send this demand to a group of Facebook employees that order your shoes, book a hotel or cancel your restaurant reservation.
Last week The Next Web reported that AI will be concluding 60% of the customer service workload. How? Through a combination of data analytics and improving the work of customer service reps through AI (companies like Assist.ai and Digital Genius). But yet again, it looks like AI will be on the predictive part of the equation. Thus, this will result in better product development and more happy clients. Robots won’t steal our jobs they will make them easier.
Imagine that: you talk to a robot the same way you talk to a friend
But basically, AI has implementations in every conversation based service where there are certain patterns and easy yes/no scripts. The only boundaries that is yet to be on the full force level are the specific aspects that make our way of conversation ‘natural’. The future and most of all the productive implementation of natural language processing software relies on those aspects. Basically what will happen in the next few years can be reduced to the question: if AI will be a gimmick or a productive tool that improves our lives on a day to day basis.
Something is heating up. Research in natural language processing is growing and receiving more funding. But it will take a while until we’ll have a conservation with a robot the same way we talk to our friends and family. Do we really want do?
Over to you now.
Would it be a good thing that robots can understand and speak the same way we do?
Would it improve our lives?
And what are the consequences of this type of development?