Blog: The Importance of Language in Human Cognition and Artificial General Intelligence

What psychological linguistics can tell us about developing true AGI

Photo by Joel Naren on Unsplash

I talk to myself a lot. Many people do. I’d say that most people utilize internal speech to work their way through various problems. Language seems to be a crucial component to the problem solving process and how we describe the world to ourselves. How do we comprehend atomic scales and cosmic scales, and the relationship between them? Such things are in many ways incomprehensible. But we do comprehend them, through our language. If language is such an important element to our higher level thought processes, then we must take this idea into consideration when designing artificial intelligence applications that mimic, or are comparable to, human thought processes.

Internal Speech

There isn’t a lot of discussion on internal monologue and internal speech, in general. Some argue that pretty much everyone uses it, while others disagree. One possibility is that internal speech is not consciously done in certain individuals. Just like everyone dreams, perhaps everyone engages in internal speech, but it’s on too subconscious of a level to recognize for certain individuals. Some people also speak to themselves out loud. I admittedly do so. Once again, this element of self communication seems to be useful in working through higher level problems. So I think we need to learn more about this phenomenon and how it might relate to AGI.

Language and World View

Some anthropologists even feel that language shapes our very perception of reality. There’s a theory, or hypothesis, known as the Sapir-Whorf hypothesis. This hypothesis contrasts with the idea that humans essentially all think and form language the same way (universal grammar), which is an idea promoted by Noam Chomsky. While this second idea might be nice, politically speaking, as it means that we have a lot more in common, psychologically speaking, there’s reason to believe that it’s incorrect. I’ll touch on that idea more in a moment.

The Sapir-Whorf hypothesis exists in two forms: the weak form and the strong form. In the weak form, the theory suggests that language has an influence on our world view and our thinking process. The strong form states that language determines our thought process. I’m more in the strong Sapir-Whorf camp.

The structure of language does seem to impact the way we think. Keith Chen looked at languages with different structure in describing the course of events. Chen’s research, at the very least, indicates that language structure influences behavior. In his research, he looked at investing habits for speakers of different languages. He found that languages, such as English, which “oblige speakers to grammatically separate the future from the present lead them to invest less in the future.”

Language can also influence whether we place blame on an individual or not. Caitlin M. Fausey and Lera Boroditsky looked into how memory can be influenced by language. English, Spanish, and many other languages utilize agentive language. However, English speakers appear to utilize it more than Spanish language speakers, in cases of accidental situations. A possible explanation as to why is that passive voice is undesirable in English, but Spanish utilizes it frequently (Enforex). But whatever the reason, English language speakers were better able to remember the agents involved in accidents, more easily than Spanish language speakers.

While it’s possible that there is another explanation, the most reasonable explanation seems to be that utilizing passive voice and avoiding the agent involved in the accident reduces the ability to remember who was involved in the incident. However, there could be other possibilities, such as cultural differences in assigning importance of blame.

The Kuuk Thaayorre language is another interesting example of how language shapes our cognition. In the language, there are no words for left or right. Every direction is given based off of cardinal directions. It’s therefore very important to always know where the cardinal directions are. The people who speak this language also happen to be amazing to keep track of where they are.

While there is still much that we can learn, these three different studies on language and cognition make the theory that language has a significant influence on memory, behavior, and general thought processes, fairly robust.

Application to Artificial Intelligence

What does all of this theory about language have to do with artificial intelligence? If we want machines that think like humans, we need to understand how humans think. And if language is indeed so fundamental to human-like thinking, then we need to really ensure that a powerful linguistic framework is built into any AGI application. It seems reasonable that the following key elements will be needed for a true AGI to function adequately.

  • Any sufficiently advanced AGI needs to be able to learn new language dynamically. Being pre-programmed with basic language processing isn’t going to cut it.
  • An AGI needs to be able to identify multiple contexts and engage in separate conversations with multiple other people.
  • An AGI needs to be able to “talk to itself.”

Taking these points into consideration, I do have a proposal to test these ideas out. We can try to create an artificial intelligence application with a basic ability to learn language. It would also need to be able to differentiate between different speakers. Especially if we have the modified Turing Test that I mentioned in my previous AGI article, we can then compare this version to another version, which includes more robust language processing.

To do so, we could take the exact same program, but add in the additional element of an internal voice and ability to observe its own speech. It’s for this reason especially that the AGI needs to be able to understand multiple contexts and speakers.

For the first part of that task, a second instance of the same application could simply be run, where the only stream of communication is between both instances. If internal dialog is an important feature in human-style cognition, we should expect to see a more human-like discussion.

Source: Artificial Intelligence on Medium

Leave a Reply

Your email address will not be published. Required fields are marked *

Back To Top

Display your work in a bold & confident manner. Sometimes it’s easy for your creativity to stand out from the crowd.