Programming languages are reserved for an elite part of the population known as 'developers'. To enable 'normal people' to communicate with a computer programme, there are two options available:
- Teach them the language of computers, which on paper is far from being the best solution. Saying "hello you" in binary code means writing "01001000 01100101 01101100 01101100 01101111 00100000 01111001 01101111 01110101".
- Teach computers to understand natural language. This field is known as Natural Language Processing (NLP).
'NLP' did you say?
NLP is a field that brings together techniques enabling a computer programme to understand and analyse natural language. It is formed of two main components:
- NLU (Natural Language Understanding);
- NLG (Natural Language Generation). Certain artificial intelligences are able to write screenplays, for example. The AI named Benjamin, designed by researcher Ross Goodwin and film-maker Oscar Sharp, created the screenplay for the short film Sunspring.
Is saying 'hello' a problem for NLP?
Because Natural Language Processing is an algorithm that does what is asked of it, it is confronted with the problem of the ambiguity of natural language. An algorithm is a set of instructions to be executed without ambiguity. However, natural language is anything but unambiguous. A word can have several meanings depending on context. For example, the word 'opera' is a noun and also related to the verb 'operate'. Elsewhere, we can use several words to say the same thing. To greet somebody, we can say 'hello', 'hi', 'hey', 'yo', 'howdy' or 'wassup' for example.
When texting or chatting we also often use abbreviations, which makes understanding words even more complex for a computer programme. Another problem encountered by NLP is coreference. Take the following example: "I like Hulk Hogan because he reminds me of my childhood" he said. For us humans, understanding this sentence is easy: the first 'he' refers to 'Hulk Hogan', while the second 'he' refers to 'I'. Performing this analysis is not so easy for a computer programme. In addition, there are variations in the alphabet and grammatical syntax from language to language, misuse of language and neologisms (such as the verb 'to Google'). In short, speaking a language is an intellectual feat in itself. If you are able to read and understand this, you are a genius. Well, you beat artificial intelligence in this area at least.
A robot that speaks like a human
NLP is particularly used for chatbots. In May 2018, Google presented Google Duplex. This conversational agent is able to make an appointment over the telephone without the person on the other end of the line realising that they are not speaking with a human. According to Google, it can handle four out of five calls. If it is unable to understand the person, the call is rerouted to a call centre where a human takes over. The presentation made a big impression because it demonstrated not only a good understanding of natural language (NLU), but above all a strong ability to generate natural language (NLG). However, while they are useful and impressive in certain cases, conversational agents are far from reaching their maximum potential.
The Siri of the future: a Jarvis Jr.
"Hey Siri, can you find a summer job for my daughter in Singapore? Writer a cover letter for her, and three different CVs for different types of companies, to match what they're looking for. And make sure it doesn't clash with her horse-riding competition. And, one last thing, find her an apartment with a good location for the job you find for her!". Idriss Aberkane in Regards Connectés, episode 32.
It seems like science fiction, but we're getting there fast. Today, AI is able to make appointments with ease and write screenplays for films. So why couldn't it take care of this type of request? AI's role is to accomplish tasks that would take us a long time in a matter of minutes. While we have seen many advances in the area of artificial intelligence, it is still far from the level of human intelligence. While it is often demonised, AI will (like mobile phones and the Internet) become harnessed and popularised. It will not be rare to see conversational agents acting like real personal assistants.