1

Will it ever be possible for machines to understand language the way humans do?

It is a famous XKCD comic strip pointing out how "Language isn't a formal system. It is a glorious chaos".

It basically talks about how the same words could mean entirely different things based on small gestures, tonality, pauses or other innuendos of one's body language within the same region of origin let alone across regions of varying cultures and languages.

On the other hand, in computer science, people have been able to make a lot of breakthroughs in understanding language and converting them into machine-understandable formats.

The question that naturally arises is whether or not it would be ever possible for machines to win this race by being a better interpreter of the notion given a text? Would it be theoretically possible and practically implementable to train a machine to understand and talk back based on interaction and other discourse gestures?

  • Brains aren't magic. In fact, hardware-wise they kind'a suck. One can legitimately question whether humans will ever manage to make a smart computer, but it should be self evident that such a computer exists in the space of possible architectures. – Veedrac Mar 15 '18 at 14:04
  • Then that would mean that all the concerns that are ever raised about a scenario where machines become more intelligent than humans are impossible because we ourselves are not intelligent enough to build such machines? @Veedrac – m1cro1ce Mar 16 '18 at 7:21
  • I'm saying it's a legitimate question, not that I believe it. Personally I think it's likely that we will. – Veedrac Mar 16 '18 at 7:43
  • @m1cro1ce By pooling resources we will build a machine that is smarter, or we will build a machine that will build a machine that is smarter, or we will build a machine that will build a machine that will build... – christo183 Jan 8 at 6:40
3

Current approaches to natural language processing have long left behind the idea of defining a formal system by hand, instead letting the machine learn the patterns in language.

So, if we can accurately model learning, then we can expect machines to be able to learn to understand language. If language learning was complicated enough to forever stay out of reach for computation, then it is unclear why humans are able to learn language in the small timespan that they do.

The problem with machines learning language are NOT "small gestures" etc. We understand each other writing on the internet without these gestures, also machines are very capable of learning from input like gestures. It's that we don't really understand language and language learning well enough to teach the machines.

0

I would like to supplement some more comments:

First: machines are artifacts created by humans. That is, we humans had some goal in mind and we decided to create an artifact that we could use as a means to achieve it. This definition imposes no restriction on what any machine can or can not do. Reiterating, it states that machines are objects created by humans to achieve some goal. I speculate that many people think that machines have some limitation based on a superficial generalization that they made of their own common experiences. They realized that machines such as blender, cars and lawn mowers had some "limitations" and generalized this as a "limitation" of all machines.

Second: we do not know how human language actually works, yet this is not an argument against the idea that machines can learn human language. If we knew how human language works it would be possible to say with certainty whether it is possible or not to build a Turing Machine that understands human language. The interesting thing is that even if we think hypothetically in the case that the Turing Machine is not able to compute the human language, it's not enough for us to build a machine that is more powerful than the Turing Machine to process the language. We have good reason to believe that the Turing machine can understand human language. I will not discuss the reasons in this text, but what I find most important is not to confuse what computers do today with what they can actually do. What computers do today depends on the implementations of engineering. There are many things that can be done with computers but we still do not know how to implement them concretely.

Third: many people find that showing complications in language is a barrier for a computer to understand the human language. The right question in this situation would be: why would this be a problem? What prevents these particularities from being treated? Critics have to be able to answer these questions.

Fourth: I would like to add an historical note. Many people think that computers are just machines, but what they do not know is that the concept of computing was created to capture the mechanism that we humans use to perform mathematical accounts. Because at the time that Turing defined computing those who did computing were people. What's amazing about this definition is that it's so good that you can even build machines with it, which is why we call it the Turing Machine. The fact that it is possible to build a machine to perform computations clearly demonstrates the power of the theory. Because computers have been a success and the history of computing is not widely publicized, people think that computing has only come with machines.

  • It would help to cite references for any claims. This would give people a place to go to get more information and strengthen your answer. Welcome to this SE! – Frank Hubeny Sep 29 '18 at 3:51
  • I made some minor edits. You may roll these back or continue editing. Again, welcome. – Frank Hubeny Sep 29 '18 at 4:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.