The Limits of Artificial Intelligence and Language

Artificial Intelligence and the Limits of Language
Artificial Intelligence and the Limits of Language

Chaos erupted when a Google engineer recently mentioned the Artificial Intelligence chatbot as a person. TheMDA The chatbot, called a chatbot, uses a broad language model (LLM) to predict possible expressions that will follow any line of text. These methods can determine how to maintain a productive conversation. Because most conversations are reasonably predictable. LaMDA accomplished this feat with such admirable skill that the engineer Blake Lemoinebegan to question whether the device was ghosted.

Lemoine's narrative received a wide variety of responses. Some people laughed at the idea that a machine could be a human. Others have argued that while this LLM is not a person, the next one could be. Others have stated that it is not that difficult to deceive people because people can be influenced spiritually.

The diversity of responses, however, highlights a more serious issue: As these LLMs spread and become stronger, there seems to be less consensus on how we should interpret them. Over the years, these systems have surpassed countless benchmarks of "common sense" language reasoning, many of which are claimed to be insurmountable by anything but a machine capable of "thinking in the full sense that we generally reserve for individuals."

However, these systems rarely seem to have the common sense they claim to have when they pass the test, and they can often still be susceptible to sheer nonsense, unfairness, and unsafe advice. These detected facts raise the troubling question of how systems can be so intelligent while simultaneously appearing so constrained.

The real problem is not artificial intelligence. The issue is the limits of language. It is clear that these systems are doomed to a limited knowledge that will never come close to the complex thinking we observe in humans, once we let go of long-held beliefs about the relationship between cognition and language. All in all, these AI systems are among the most impressive on the planet, but they will never be like us.

In the 19th and 20th centuries, the idea that knowledge was purely linguistic—the idea that knowing anything was as simple as thinking about the appropriate sentence and figuring out how it connected to other sentences in a vast network of all the true statements we know—was a recurring theme in philosophy and science for many.

According to this logic, the ideal form of language would consist of arbitrary symbols linked together by strict rules of inference. However, natural language can also work if extra care has been taken to eliminate ambiguities and inaccuracies. In Wittgenstein's words, all true statements are "the sum total of natural knowledge".

In the 20th century, many people argued that, despite their appearance, cognitive maps and mental representations should be primarily linguistic, making psychological explorations about them controversial.

He has the belief that, for some over-educated and intellectual types, an encyclopedia contains everything known, so reading them all will enable us to fully understand everything.

It is also stated that the default paradigm inspired many of the early work in AI in various ways in accordance with logical rules.

For these researchers, an AI's knowledge consisted of a large database of manually linked real sentences.

An AI system was considered intelligent when it specified the appropriate sentence at the appropriate time, that is, if it used symbols correctly. The Turing test is based on the idea that if a machine is saying everything it's supposed to, it needs to know what it's talking about, because knowledge is exhausted by only knowing the appropriate sentences when to use it.

But this has come under the harsh criticism that has followed it ever since: Just because a machine can talk about anything doesn't mean it knows what it's talking about.

This is because language only represents a very small and very specific subset of knowledge. Language does not consume knowledge.

All languages, whether a programming language or a spoken language used in spoken, written, or symbolic logic, are based on a particular representation scheme that excels at representing individual objects and attributes at a very high level of abstraction.

But there are important differences between reading a musical note, listening to a recording of the music, and playing the instrument.

In all representative schemas, information about something is compressed, but the information included and excluded from the compression is different.

The representational scheme of language relies on more concrete information, such as representing irregular shapes, the movement of objects, the work of a complex device, or the work of a fine brush.

However, iconic information that includes things like images, records, graphs and maps, as well as the distributed information found in trained neural networks (what we often refer to as technical knowledge and muscle memory) are non-linguistic representational schemas.

“Either Picasso or twombly" What it looks like? Each diagram finds it easy to describe some information, while others find it difficult or even impossible to do so.

Recognizing how little information a linguistic representational schema carries on its own can help us better understand what makes it unique and how limited it is.

The language has an extremely narrow bandwidth for information transfer; Isolated words or sentences won't mean much if they're devoid of context.

Also, many sentences have a lot of ambiguity due to the plurality of homophones and pronouns.

As Chomsky and his cohorts have argued for decades, language is not a clear and unambiguous tool for clear communication.

This approach focuses on applying general reading comprehension techniques to understand a text, but research shows that background knowledge is actually the most important element for comprehension in children. Understanding the topic at hand is essential to understanding a sentence or section.

source: noemamag







Similar Ads

Be the first to comment

your comment