The rise of large language models (LLMs), exemplified by Chat GPT, has sparked a fervent debate over the capabilities and constraints of artificial intelligence. Proponents of LLMs argue that they represent a significant step forward in our ability to comprehend language and the world around us. However, opponents argue that LLMs are shackled by their inability to replicate human intelligence and moral reasoning.
At the center of this discussion is the question of whether LLMs can truly grasp the intricacies of language and context in the same way that humans do. Critics contend that although LLMs may be able to perform certain tasks at a superhuman level, like data analysis or language translation, they lack the underlying knowledge and experience that humans possess. This deficiency results in their inability to reason and think critically like humans.
The New York Times article “Noam Chomsky: The False Promise of ChatGPT” delves into the limitations of large language models (LLMs) such as GPT-3, particularly their incapacity to replicate human intelligence and moral reasoning. The article highlights that while LLMs may be adept at specific tasks, they still lack the crucial ability to reason and think critically like humans, due to their insufficient understanding of the world.
The emergence of large language models (LLMs) such as Chat GPT has triggered a spirited debate about the capabilities and limitations of artificial intelligence. Supporters of LLMs argue that they are a remarkable step forward in our ability to understand language and the world around us. Detractors, on the other hand, claim that LLMs are constrained by their inability to replicate human intelligence and moral reasoning.
At the core of this argument lies the question of whether LLMs are truly capable of grasping the nuances of language and context in the same way that humans do. Critics contend that although LLMs may be capable of generating coherent sentences and responses, they lack the profound understanding and intuition that humans possess when it comes to interpreting and analyzing language. Consequently, LLMs may struggle to comprehend and respond to the intricacies of human communication, such as sarcasm, irony, or metaphor.
Additionally, opponents argue that LLMs are unable to engage in moral reasoning in the same way that humans can. Although LLMs can identify patterns and make predictions based on data, they are incapable of participating in the type of ethical reasoning that humans are. This is because human moral reasoning is deeply rooted in our understanding of the world and our experiences, which are difficult, if not impossible, to replicate in a machine. In other words, to truly understand the shared human experience, one must be embodied in the world, just as human beings are. The kind of data found in text is limited to a very specific type of understanding that does not come close to capturing the full range of human experience.
Nevertheless, proponents of LLMs contend that these models represent a significant advancement in our ability to process language and comprehend the world. LLMs can process vast amounts of data and identify patterns and relationships that humans may not be able to discern. This enables LLMs to perform tasks at a superhuman level, such as language translation or data analysis. LLMs can also summarize large amounts of text and provide context that may be difficult for humans to achieve. Perhaps most importantly, LLMs can assist in solving problems. Whether you’re struggling to estimate the cost of a project, determine its duration, identify its steps, or even determine how to complete it, LLMs can assist you, because this type of information is contained within text. You don’t need to have uniquely human experiences to complete tasks such as taxes or legal work.
GPT’s ability to process large amounts of legal documents, statutes, and cases is a potential game-changer for the legal field. It can digest complex legal language and present it in an easily understandable format. The natural language processing capabilities of GPT can also be leveraged to generate legal briefs, contracts, and other legal documents. In essence, GPT can become a virtual legal assistant that can handle tedious tasks and free up time for attorneys to focus on more complex legal work.
Proponents of LLMs argue that they can be trained to understand the nuances of language and context. As LLMs are trained on ever-increasing amounts of data, they can learn to recognize and respond to subtleties of human communication, such as sarcasm or irony. While LLMs may not be able to tell a joke yet, they are getting better at understanding human communication.
Despite these advantages, there is still a significant issue of bias in LLMs, which has been a major concern in the field of artificial intelligence. While LLMs may generate technically correct responses, they may also perpetuate and amplify biases present in the data they are trained on. As a result, LLMs may generate responses that are discriminatory or perpetuate existing biases. However, with proper rules and training, LLMs can be programmed to identify and eliminate biases in the data, potentially making them less biased and more objective than humans in certain tasks.
At the end of the day, LLMs are as good as the data they are fed, and the rules they are trained on. Bad rules or bad data = bad output.
Another key issue in the debate over Chat GPT and LLMs is the idea of improbable explanations. While LLMs may be able to process large amounts of data and make predictions based on patterns, they are not designed to challenge the prevailing paradigm or to come up with new and innovative ideas. This is where human intelligence still has an edge over LLMs. Humans have the ability to make connections between seemingly unrelated concepts and to conceive of improbable explanations that can lead to breakthroughs. In fact, many creative breakthroughs are unexpected, improbable, and often unexplainable or replicable. There is definitely a mystery to how humans have come up with creative solutions over time, and there’s no known way for LLMs to be able to do that.
So, the danger of relying too heavily on LLMs is that we may lose this ability to conceive of improbable explanations. We may become so reliant on these models to provide us with answers that we forget the importance of challenging the prevailing paradigm and thinking outside of the box. This could lead to a stagnation of knowledge and a lack of progress in fields where improbable explanations are necessary.
Imagine, for example, that people read less books because of these models, but in these books, there are certain strands of thought, that could be improbable sources of enlightenment or inspiration. There is a kind of magic to that process, and if we rely on LLMs for information or knowledge, we may cut off that magic, and with it, all the potential for truly creative solutions.
You might wonder whether creativity is simply the offspring of drudgery or really hard work, and you might be right to some extent. But a closer inspection of history shows that many of the most significant breakthroughs in science, technology, and the arts were more haphazard than planned, and more improbable than predictable.
Think of the discovery of penicillin, the creation of the printing press, or the composition of Beethoven’s Ninth Symphony. Each of these accomplishments relied on a combination of serendipitous circumstances, intuitive leaps, and hard work. While LLMs can produce impressive outputs based on existing patterns and data, they lack the ability to make random associations, explore tangential ideas, or generate new insights based on personal experiences or emotions. Therefore, while LLMs may improve productivity and efficiency, they cannot replace the creative ingenuity that has driven human progress for centuries.
To illustrate this point, let us take the example of climate change. The prevailing paradigm in the scientific community is that human activities are contributing to global warming, which will have disastrous consequences if left unchecked. However, there are still many who reject this explanation, either because they do not believe that human activities are responsible for climate change or because they do not believe that the consequences will be as severe as predicted.
If we rely solely on LLMs to provide us with answers to the issue of climate change, we may miss out on important insights and alternative explanations that challenge the prevailing paradigm. For example, there may be other factors at play, such as natural climate cycles, that are not adequately captured by the models.
Furthermore, even within the realm of human activities as a cause of climate change, there may be improbable explanations that are worth considering. For instance, some researchers have suggested that geoengineering, or the manipulation of the Earth’s environment on a large scale, could be a potential solution to climate change. While this idea may seem far-fetched, it highlights the importance of considering all possible solutions, no matter how unlikely they may seem.
Common sense and intuition
Humans can use common sense and intuition to make decisions in situations where there is no clear answer, while LLMs rely solely on data and algorithms. Thinking is not merely a conscious activity, but involves unconscious processes.
The human psyche is a vast and complex landscape, filled with many layers of consciousness and unconsciousness. Thinking, as an integral component of the psyche, is no different. While we may be aware of some of our thought processes, much of our thinking occurs beneath the surface, hidden in the depths of the unconscious mind.
To understand why thinking involves unconscious processes, we must first recognize the true nature of the unconscious. It is not simply a repository of repressed desires and traumas, but rather a vital source of creative energy and intuition. The unconscious mind is the wellspring from which our thoughts and ideas originate, bubbling up to the surface of our awareness when the conditions are right. Without the unconscious, our thinking would be limited and sterile, lacking the depth and richness that comes from tapping into the hidden depths of our psyche.
Thus, thinking involves unconscious processes because the unconscious is the source of our most creative and innovative thoughts. To tap into this wellspring, we must be open to the mysterious and unpredictable workings of the unconscious, allowing it to guide us towards new and unexpected insights. By embracing the unconscious and recognizing its importance in the thinking process, we can unleash our full potential and access a realm of thought that is beyond our conscious awareness.
There is ample evidence from cognitive psychology and neuroscience to support the idea that thinking involves unconscious processing.
One example of such evidence comes from the field of priming, which shows that exposure to a stimulus can affect subsequent behavior or thinking, even when the person is not consciously aware of the stimulus. For example, participants who were briefly exposed to the word “yellow” were faster to identify a picture of a banana than those who were exposed to an unrelated word, even though they were not consciously aware of the word “yellow”. This suggests that the unconscious mind is processing information and affecting behavior without our conscious awareness. (See “The Unconscious Mind” by John F. Kihlstrom and Terrence M. O’Brien, published in American Psychologist in 1995.) https://www.jstor.org/stable/1699849
Another line of evidence comes from brain imaging studies, which have shown that many cognitive processes involve activity in brain regions outside of conscious awareness. For example, studies have shown