Like writing, the spear, and the wheel, language is one of the many things that enabled humanity to gain an early advantage in our formative years.
Language and Machine Learning are Inseparable
Like writing, the spear, and the wheel, language is one of the many things that enabled humanity to gain an early advantage in our formative years. It’s the linchpin piece of many things we take for granted today, such as diplomacy, negotiation, dating, and arguing, though its ever-expanding complexities mean that language remains an area of fierce study, both by experts and optimistic beginners.
Efforts to create a computer that thinks and acts just like a human – an artificial intelligence – inevitably require an understanding of just how our species learns a language. MIT scientist Josh Tenenbaum notes that the entire AI concept has language “at the heart of it”, simply because it’s the foundation of human intelligence. We have been using spoken and written languages for around 5,000 years, after all.
The problem is that language can be difficult to learn, even for biological humans. Only 43% of all the people on earth can speak a second language, with trilingualism (speaking at least three languages) almost a global myth, at 13%. Of course, some areas do have a much higher or lower grasp of the language. In Europe, for example, 98% of Luxembourgers speak a second language. In Ireland, that figure is 60%.
While the education system is the beginning of most learners’ flirtation with other languages, adults tend to learn in a self-guided way, using the internet or software. The website Preply offers a 1-on-1 tuition experience in a range of languages that can be tailored to a person’s specific goals, such as business or travel. This kind of service also provides visual learning aids, such as the one about collective nouns, below.
Observation and Experience
Scientists have historically tried to teach language to machines using the computer equivalent of post-it notes to explain things like grammar. However, this approach is, quite literally, unnatural. While humans do come bundled with our own version of language software, we learn through observation and experience. Older language-orientated AI is simply stuffed with rules and words until something happens.
So, what’s the solution? Tech website Engadget claims that a ‘blank slate’ AI could provide a salve to scientists’ learning woes. By watching subtitled videos, the AI is expected to connect the words it sees to actions and objects. Over time, the idea is that our budget android will not only understand the meaning of words but will be able to sort correct from incorrect as far as usage is concerned.
An obvious question to ask is – don’t we already have language-capable AI in the form of Siri and Alexa? There’s an important distinction between being an AI and using AI. While clever, personal assistants only know how to interpret words related to an existing set of tasks. They’re not able to learn new things or approximate human traits like problem-solving, reasoning, or even more sinister things like deception.
After all, the only thing scarier than a truly self-aware computer is one that pretends it isn’t.
It’s the linchpin piece of many things we take for granted today, such as diplomacy, negotiation, dating, and arguing, though its ever-expanding complexities mean that language remains an area of fierce study, both by experts and optimistic beginners.