The Things AI Can't Do: A Comparison of Human and Machine Intelligence

Overview

Artificial Intelligence (AI) has come a long way since its inception and is now a vital part of our daily lives. With advancements in technology, AI is now an integral tool in several industries, leading to a significant increase in the automation of certain jobs and routine tasks. However, its widespread use also raises concerns about its potential to surpass human capacity in certain areas. With AI capable of writing essays, generating unique images, face recognition, and even predicting market trends, it is only natural to wonder what it is that humans excel at and AI falls behind. As AI continues to evolve, it becomes important to consider its limitations.

Human Intelligence vs. Machine Intelligence

AI and human workers differ in their abilities and qualities, making it incorrect to assume that AI will replace human workers. While AI-based machines are fast, accurate, and consistently rational, they lack intuition, emotions, and cultural sensitivity, which are crucial qualities that humans possess. Machine intelligence and human intelligence differ from each other. Due to their ability to learn and make judgments depending on the data they are presented with, advanced computers are regarded as intelligent. However, human intelligence is different and goes beyond the ability to make information-based decisions. AI can be understood as a computer acting and deciding in ways that seem intelligent and mimic human intelligence. Based on Alan Turing's philosophy, AI imitates human behavior, decision-making, and communication. This type of intelligence is useful in organizational settings as it can identify patterns in data and perform repetitive tasks without physical exhaustion. However, AI's abilities are limited to the available data on which they are trained, while humans can imagine, feel, judge, and adapt to changing situations. AI is best suited for repetitive and low-level tasks in a closed management system with clear rules. However, human abilities are more versatile, allowing them to shift focus from short-term to long-term concerns. Such abilities are unique to humans and do not rely on a constant flow of external data, as with AI.

AI's Common-Sense Conundrum

Humans have "common sense" because we form long-lasting mental images of the things in our world – what they look like, how they behave, and what they can and can't do. Deep neural networks do not make such mental models. They don't have

specific, meaningful representations of objects like a house or a cup of coffee. Instead, they use statistical information from raw data to make helpful predictions using probability. For many tasks, this statistical approach works well. However, it's not always accurate and can cause basic common-sense mistakes that humans would never make.

One example of machine learning's lack of common sense is observed in image recognition tasks. Despite being trained on a large dataset of images, a machine learning model may not always be able to distinguish between similar objects or understand the context of an image. For instance, a machine learning model trained to recognize animals may mislabel a picture of a zebra crossing a road as a "giraffe on a bicycle." This is because the model has not learned the common-sense knowledge that zebras cannot ride bicycles, and it is simply making a prediction based on the patterns it observed in the training data. This lack of common sense highlights the limitations of current machine learning models and the need for new approaches to incorporate prior knowledge and better understand the context of the problem. According to Elias Bareinboim, AI systems are clueless regarding causation. Understanding cause and effect is a big aspect of what we call common sense, and it's an area in which AI systems today "are clueless." Elias Bareinboim is the director of the new Causal Artificial Intelligence Lab at Columbia University. The topic of common sense has been discussed within the AI community since its inception. John McCarthy's 1958 paper on AI, considered one of the earliest in the field, was titled "Programs with Common Sense."

Empathy and Emotional Intelligence

Artificial Intelligence (AI) systems have advanced significantly in detecting emotions and responding to them in pre-programmed ways. However, they still fall short in terms of truly understanding and experiencing emotions in the same manner as humans. For instance, an AI system might be able to detect that a customer is angry, but it cannot feel the frustration and emotions the customer is undergoing. This limits its ability to provide genuine empathy and support to customers.

Moreover, AI systems might categorize our words into positive or negative categories, but they do not fully comprehend the emotions and the subtext behind them. Cultural references, sarcasm, and nuances in language can drastically alter the meaning and emotions expressed, but AI still lacks the sophistication to understand such delicate nuances. In many cases, the things left unsaid may also imply emotions. The lack of ability to understand this subtext makes it difficult for AI to provide genuine emotional support.

AI researchers agree that AI systems cannot have emotions of their own but can mimic emotions, such as empathy. However, much debate surrounds whether a simulation of emotion truly demonstrates an understanding of emotions or is still artificial. Even when an AI system can recognize and react to emotions, it is still unclear whether it qualifies as emotionally intelligent.

Continuous Learning and Adaptation

In real-world environments, new information is constantly being generated. Humans can easily integrate this information into their behaviour, making real-time adjustments as necessary. On the other hand, AI is limited by a batch-based training and deployment process that doesn't allow for continuous adaptation. The field of machine learning refers to this type of human ability as "continual or lifelong learning."

The traditional AI development process consists of two phases: training and deployment. An AI model uses a fixed dataset during training to learn how to perform a specific task. Once training is complete, the model is deployed to analyze new data based on what it learned from the training data. However, if the model needs to be updated with further information, it must undergo a time-consuming retraining process. Conventional deep learning methods do not allow for continuous learning, at least not yet. A phenomenon known as "catastrophic forgetting" is the primary obstacle, as new information can overwrite previous learnings in a neural network. Humans have the ability to learn and adapt continuously. Currently, AI systems lack this capability, but advancements are underway in continuous learning.

Conclusion

AI has come a long way and now plays a vital role in our daily lives. However, it has limitations as it lacks human qualities such as idea generation, intuition, emotions, common sense, and cultural sensitivity. Despite these limitations, AI has proven to be a powerful tool in various industries and applications such as data analysis, automation, and decision-making. Its intelligence is limited to the dataset it is trained on, and it cannot make mental models, which affects its predictions. Additionally, AI cannot truly experience emotions like humans, and its batch-based training and deployment process limit its ability to learn in real-time. Therefore, it is safe to say that though AI can do a lot. It cannot replace human creativity. It is this creativity that is the essence of all art. This creativity or the power of creating ideas that we humans have is what is most important and needs to be protected.