Natural Language Processing and AI— Machines with Impressive Communication Skills
Intelligence can be measured in various domains, as we’d seen already. In terms of language, though, how would you measure intelligence? Artificial intelligence is after all attempts at making machines “intelligent” by building machines that mimic the way humans think, decide, and execute. Where does language fit in, why must language be taken seriously, and how far has AI come?
Language as a Skill: As far as we know, humans are the only animals or living beings within our knowledge range that have come up with “languages” for a long time now. Most acquired skills are developed from a need and effort put in to satisfy the need. Since humans lived as communities and inherently had the ability to communicate in several different ways, the need for expressions, expressions to express one’s thoughts developed. This is how, along with other skills, our ancestors learnt to sketch, paint, sculpt, and slowly, words were formed. Unique sounds were attached to small words or symbols so one could say and hear the same in any discussion. Structures were assigned for words, either in the form of symbols or what later evolved to be called “letters”, that’s how alphabets were born.
Language — an Inherent, Common Human Skill: You could say that skills associated with language were skills that people from all the different races shared, across the world, by simply looking at the history and literature in the many thousands of different languages that have existed. This should prove to us that unlike, say for example, a very high computing speed or an eidetic memory, the usage of language is not a rare skill at all, and unlike typing using a keyboard, it is not a newly acquired skill either.
The Sophistication of Natural Human Language: There are debates going on about how well we humans have understood other animals’ abilities and ways. Some would argue that non-human animals would have their own unique ways of communicating with one another, which we have still not understood. While that could be true, there is no evidence that proves that animals’ ways of expressions are more efficient than ours are. For example, we can go from continent to continent and still use either a common language or other means to express complicated thoughts and ideas. We have literature from ancient times that prove how skilled we are in both thoughts and expressions.
“Language shapes the way we think, and determines what we can think about.
— Benjamin Lee Whorf”
Language as a Part of Human Cognitive Abilities:
Human language (in almost any human language) uses words such as nouns (which comes from assigning names) and verbs to describe actions (in different period of time, relatively speaking), slang, proverbs, metaphors, and many other figures of speech. While these words are used to describe thoughts as accurately as possible, combined with this, there are elements in the tone of voice, facial expressions, and other forms of body language that need to be taken into account to derive the meaning of a certain piece of communication. And most people would be able to do it all with ease as long as they can understand either the written/spoken language or the gestures or both.
How often do you think about how quickly your brain is able to process so many things at once? In the middle of a discussion, if someone crack a joke, in a matter of seconds, you understand that it’s a joke, and your instant reaction would be to laugh or react positively to it. And if certain sentences or parts of sentences trigger or offend you because there is something negative that your brain associates with the said words, again, within a matter of seconds, your brain can and would recognize that whatever you just heard was offensive for you, so you’d react negatively to it. None of this is simple, easy stuff!
With languages, you are at home anywhere.
— Edward De Waal
Slang and Dialects:
The same language need not and will often not be be spoken the same way across different populations that use the language for communication. this is something we can easily understand — that almost every language has its differences created by different groups of people, closely associated with their cultures and history. But this is not all that complicated because somehow, if you can manage to collect them all, you’d get how it is spoken by different sets of people and if you put it all together, you have a language that is complete.
Grammar and Context:
Grammar is what gives a language structure. You could call it a book of rules that tells you what’s acceptable and what is not so that communication can be clear and misinterpretation might be avoided as much as possible. While understanding grammar might itself not be all that hard, again, the more used to a language and its grammar you become, the more easily you would be able to think of grammatically correct utterances before you speak and notice grammatical mistakes when you come across them.
Context is a lot more complicated to understand than grammar would ever be. A sentence could be completely perfect grammatically speaking, but still the meaning could vary based on the context. Understanding this, again, requires complex skills. You need to be aware of the circumstances, the subject being talked about, general knowledge, and other aspects involved to interpret the context accurately. Once again, most of us can do this easily and naturally when it comes to commonly discussed topics or subjects we are familiar with.
Take this sentence for example: “Geometry, as a logical system, is a means and even the most powerful means to make children feel the strength of the human spirit that is of their own spirit.” — H. Freudenthal
Umm … This sentence seems to be slightly more complicated to understand than even geometry itself! (Just kidding!) Now, if you take the first half of the sentence, say, “Geometry, as a logical system, is a means and even the most powerful means to make children”, it seems grammatically correct but still sounds ridiculous, right? Geometry can help you make children? Mathematicians might want to argue that there’s geometry even in reproduction, (once again, kidding!) but yeah, the common man or woman reading this would be able to easily figure out that something is comically funny here! This is something I’d call a pretty good example of understanding context. See, it takes an understanding of context more than an understanding or knowledge of grammar to understand expressions.
Deciding How to Respond:
This is another key part of human intelligence — decision making skills, which again, we can naturally, easily do (whether it’s the best, we don’t know, but humans are naturally capable of this skill). This is another layer of complexity that the human brain naturally possesses and can be trained to perform better. And communication is not just about listening and understanding but also includes responding. When you identify different elements in sentence or a piece of communication, how do you identify which kind of response would which part, and when must you say something and not say something?
Communication Skills as A Part of Intelligence — How Does AI Come In? You can see how vast the concept of language itself is, and that is why it is universally agreed that communication skills can be considered one among the many measures of intelligence in people. Even someone who can’t speak and/or hear would still need communication and has to learn to communicate in some form. In fact, this is one of the most important traits that separate humans from non-human animals. Not only is communication sophisticated but also a must-have skill for human beings. And one of the goals of creating artificial intelligent systems is to help or assist humans with the tasks that can be done easily but would take time so that humans can use their cognitive and other skills in things that are more important.
How far Have We Reached?
Can a machine understand context? Can it be programmed to understand context in any situation using machine learning? Can a machine understand and identify puns? Can a machine crack a joke on its own, unlike the chatbots that are fed with a database of jokes and pull one for you when you ask them for one? Would a machine know when you’re being sarcastic?
Wait, wait, hold your horses! One thing at a time. Before we begin, we need to firstly understand what we want to do. When we understand the goal clearly, we then begin from step 1 — we build from scratch. That is how successful projects get built.
Natural Language Processing is an area of artificial intelligence that is truly challenging and realistic to a great extent, which is something motivating to work with. Scientists and researchers have managed to create stunning, effective AIs that do different tasks in the field of NLU, and this is a field that shall continue to grow. Most of these work either entirely or most effectively in English, “the universal language”, but translation is picking up in its effectiveness, too. Even recently at work, we saw a complicated sentence in Spanish being translated into English flawlessly, so that is a small victory in the massive world of machine learning.
All parts of the human brain are intertwined to go with one another. Common sense is closely associated with parts of the brain that process language, too. What happens when there is not enough information in a sentence or in a set of sentences, not enough information we need to process something? Like a person wanting a delivery but giving an address? When must you ask and when must you be able to identify on your own? Once again, this is something human beings know all too well. A regular customer in a small shop who always buys things only for himself and has lived in the same house for decades is not going place an order for somebody else, unless he himself explicitly says so. That’s the kind of understanding between two humans — you automatically know when to ask and when a piece of information is not needed. (Not saying assuming is right, but we’re talking about casual communication style here.) This ability to reason in terms of language can also be achieved in a machine, researchers believe.
How Far Have We Come in NLU?
For now, tasks that only humans were thought of being able to do, machines are doing. Machines are learning to outperform themselves. For now, it’s not entirely flawless, and we wouldn’t want it to perfect either, but can it be assigned monotonous tasks that required human-like understanding of language? Looks like we are already there! We are not in a position to use these skilled technologies and apply complex algorithms in situations that can put human lives at risk. That would be highly unethical. One such example is letting AI decide the fate of people convicted of crimes in a courtroom. This is not a smart move — it’s stupidity!
But as mentioned earlier, for relatively simpler tasks that are demanding and taxing, AI can be highly reliable, and to a great extent, accurate with all the advanced algorithms that are now being deployed. And this stream of discoveries and advancements will only continue to grow to achieve more complicated goals. Consider this just the beginning.
