Natural Language Processing has seen quite a revival with Internet giants putting a lot of smart heads at solving NLP’s toughest problems. Incredible progress in processor speed allows to stack neural networks so deep that the successes in computer vision are repeated in NLP. Even though words are arbitrary symbols and as such fundamentally different from imagery, sentiment analysis, document classification, semantic search, and even Machine Translation, all work pretty well now.
Project managers have to be careful with vendors’ claims and managers’ dreams to just throw in lots of data and the machine will figure it out by itself. Still, Machine Learning tuned with human knowledge and domain terminology achieves production quality for many tasks and in most languages.
More dramatic, however, is the opposite effect: the pivotal role NLP plays for Artificial Intelligence. Despite the impressive advances in AI like beating humans in Go, one fundamental capability remains elusive: language. Siri and Alexa can follow simple commands and answer basic questions (in a handful languages) but can’t hold a conversation. They have no real understanding of the words they use.
Language will determine whether machines become a part of our everyday life or whether they remain mysterious black boxes. As MIT’s Josh Tenenbaum brings it to the point “There’s no way you can have an AI system that’s humanlike that doesn’t have language at the heart of it”. Sure, we can have immensely powerful software but without language our relationship with AI will be far less collaborative. And probably far less friendly.
It is thus hard to envision how we will interact with AI without language, without being able to ask machines, “Why?” and “What are you thinking about?”. This will require a lot of knowledge about language but also a huge amount of domain and common-sense knowledge about the world. Only then we can solve the hardest of all challenges. So hard, that it has been stated almost 70 years ago as the ultimate test for the existence of Artificial Intelligence: a meaningful conversation with a machine.
Given the above and given that most all developments are English only, or with some luck, include French and German, how do we prevent a situation where most EU citizens cannot benefit from AI?
1 comment on "No AI without language at its heart"
Semantics is the answer, based on linked data (e.x. http://schema.org). In the semantic web machines are not only able to process long strings of characters and index tons of data. They are also able to store, manage and retrieve information based on meaning and logical relationships. So, semantics adds another layer to the web and is able to show related facts instead of just matching words.
Such logical-systems will be able to translate the meaning, rather than just words. Also programming will come to a new level. It will resemble communication with a machine with describing the desired result. Looking even further, such technical background may become the basis for a strong AI, and may be applied in Nano-Biological Computing (https://www.youtube.com/watch?v=xcHcNyC6O84) that may appear within the next decade. We will witness the birth of new smart assistants significantly surpassing current poor systems like Siri.