Chatbots: Too Smart to Not be Dumb?

The amount of slang in Spanish is overwhelming, and people invent a new term or acronym every day. This is very noticeable in instant messaging, where people try to convey ideas with the least possible characters and effort, not just to save time but also to be “in.”

I am not opposed to the development of a language, and this is inevitable, but nowadays, at least where I live, this “evolution” has been stretched to the point that you must sound sloppy or even grammatically unintelligent if you want to fit into a conversation or be popular in a chat group.

One of the problems is that once a term or expression is perused, the person or persons who use it often can no longer differentiate the proper tongue of Cervantes from colloquial talk. Moreover –and this more worrisome and much more common – people forget how to spell properly. The slang is not even consistent: the variations and modifications (grammar mistakes, actually) are random and the newly coined phrases do not always make logical sense.

Let us take the use of the accent mark on a vowel (á, é, í, ó, ú) as an example. In Spanish, one of the uses of the accent mark is to differentiate words which otherwise would have the same spelling. For instance, “si” (without an accent mark) means “if” but “sí” (with an accent mark) means “yes” and is also a reflexive pronoun. When it comes to chatting in Spanish, it is not just convenient but also fashionable to omit all the accent marks (and most of the punctuation marks). The phrase “si quiero” can therefore mean two things, either “if I want” or “yes, I want it,” so the actual intention of the chatter can only be inferred from the context. Likewise, “no hay como jugar tenis” might mean “there’s nothing like playing tennis” or “it’s not possible to play tennis.” Or, “?como mismo?” can be “shall I eat?” as well as “how, finally?” And so on.

(There is even another factor involved, a very human one: we are afraid of looking ignorant or stupid and being a perfectionist with language is more likely to disclose what we do not know or understand. We thus hide behind the vagueness of acronyms, emoticons, ambiguity, and clichés.)

In short, you can see why I could live without slang – but my point here is the following: how can a grammatically educated chatbot, with a command of ample and precise vocabulary, pass for an actual person?

I have been confronted with many a chatbot that was unintentionally unintelligible because of the human who programmed it transferred his flaws, but when the chatbot is not too advanced the problem is transparent, so it is not a problem. However, what happens when the AI is smart enough to pass for a human? Will it learn to "dumb itself down" occasionally to be believable?

It will, of course. It must be happening already.

For AI to achieve human-level intelligence, it must excel in fundamental tasks, such as coping with natural language. Creators of machine-learning AI boast that once the system undergoes training, based on (enormous) datasets, its algorithm can successfully apply this knowledge to new data until it gets things right. Eventually, machines will mimic the way humans communicate.

So isn’t it ironic? The most sophisticated software getting smarter, learning faster, emulating more and more the genius of us humans, and thus throwing up errors, imprecision, and mysteriousness here and there to maintain the illusion of a human dialogue?

This is kind of dumb – please forgive the redundancy. It is beyond ironic. But it is true: the AI behind chatbots will be too smart to not be dumb.

I suppose a truly clever AI that needs to pass as a person will strike a balance between "dumb," swanky, and pragmatic that will allow it to achieve its goals. I also suppose this will be ultimately okay: robots and chatbots are here to serve.

Alas, there will be a downside: requiem for Spanish.