How much can a computer understand?

​Machine learning has revolutionized computers understanding of language in just a few years. Yet they still do not truly understand what it is that they know.

Some people are afraid that there will be a time when computers get so smart that they form liaisons and take over the world. Shalom Lappin, Professor of Computational Linguistics at the University of Gothenburg, is not one of them.

Shalom Lappin, University of Gothenburg“The revolution in artificial intelligence that has arrived with the deep learning technology is still in its infancy, and even though it’s developing fast, I don’t think the idea of malicious super-agents is a real prospect that we have to worry about in the near future.”

Computers only have the ability to reason about the task for which they are trained. They can find patterns and associations in millions of data, thus become better than humans in playing chess, translating and writing texts and driving a car. But as soon as we leave a specific and defined area and jump to another they are lost.

“My personal feeling is that we will never quite get to the point where machines have something that resembles general reasoning power. But I could well be wrong”, says Shalom Lappin.

Computers’ weakness is that they don’t really understand anything, adds Richard Johansson, Associate Professor at the division of Data Science at Chalmers. But with the help of machine learning, they can be very good at recognizing informative patterns. Therefore, they are very good at understanding languages in both speech and writing.

Richard Johansson, Chalmers“The development is gradual, sometimes in leaps. Google translate is much better now than ten years ago, and today it provides useful translations, but I hardly believe that professional translators use it to any large extent”, says Richard Johansson, and mentions a problem area for machine translation:

“The word "it" is translated with "den" or "det" in Swedish, and sometimes you may need to go back a few sentences to understand what “it” refers to to be correct.

The improvement of Google's machine translations in recent years is largely due to the switch from statistical translation to a deep learning model. This allows the system to understand the context and thus generate improved translations.

How good can computers get at recognizing patterns and relations? The sky is the limit, says Shalom Lappin, and mentions the development of facial recognition as an example where the computer is already superior to humans.

Translations are the flagship area of machine learning technology today, but how well do computers perform on language processing? Imagine, for example, reading a text to a computer and getting it processed and rewritten. Richard Johansson thinks it’s not impossible.

“First, your speech signal will be converted into text, and that technology is relatively good already. Then the text will be rendered into a grammatically well-structured language. I think that is fully operational within a few years.”

Is there a risk that the personal language will disappear?
"Yes, should people get lazy and formulate carelessly, the language can certainly get more generic and less personal. On the other hand, I think it will take a long time before computers can make major changes in texts, such as moving paragraphs or sentences to make them well-structured", says Richard Johansson.

  • Artificial intelligence refers to computers that imitate human cognitive functions such as learning and problem solving.
  • Machine learning is, simply put, algorithms that are trained to draw conclusions based on large amounts of data.
  • Deep learning is machine learning that uses so-called neural networks (see below) as a model for learning.
  • Artificial neural networks are self-learning algorithms that imitate the model of biological neural networks (such as the brain). Artificial neural networks can often handle problems that are difficult to solve with conventional task-specific programming. A neural network must be trained with examples before it can fulfill its intended function.

Welcome to our Initiative seminar on Digitalisation:
Security & Privacy | Machine Intelligence

On 15 March 2018, Chalmers organise a second Initiative seminar on Digitalisation. This time we present a more in-depth programme – with half a day on Security and Privacy and the other half-day on Machine Intelligence.  

See the programme and register for the seminar >

Published: Wed 17 Jan 2018. Modified: Mon 26 Feb 2018