Generative Artificial Intelligence (AI) has made remarkable strides in recent years. It made us spellbound with its ability to create art, music, literature, and even human-like conversations.
From chatbots to image generation, generative AI has been hailed as a revolutionary technology with seemingly limitless potential. However, amidst the excitement surrounding its capabilities, it’s crucial to acknowledge the boundaries and limitations. While it may excel in certain tasks, there are areas where it falls short, highlighting the complexity and nuance of human creativity and understanding.
Generative AI models are trained on vast datasets to mimic human-like behaviour. However, they often struggle with understanding the context. While they can generate text or images based on patterns in the data, they lack true comprehension. This limitation becomes apparent in tasks requiring a deep understanding of human emotions, cultural & social aspects, or subtle cues which are integral to human communication and creativity.
---
Example: They can generate a poem based on the textual description but often it sounds immature. Against my input “Write a poem about summer in India in 2024 in the style of Rabindranath Tagore”, it generated this:
In the embrace of summer’s golden hue,
India awakens ‘neath skies so blue.
Fields adorned with jasmine’s sweet perfume,
As whispers of mango groves in bloom.
‘
…
Beneath the shade of ancient banyan trees,
Children laugh and play with joyful ease.
Their laughter echoes through the balmy air,
As they weave dreams without a single care.
…
Oh, the joy of summer’s tender embrace,
In every corner, in every space.
India’s heart beats with a vibrant song,
As the days grow long and the nights grow strong.
With each dawn, a promise of delight,
As nature awakens from slumber’s night.
In 2024, as in the days of yore,
India’s summer, forevermore.
The poem sounds as if written in 1924 as a poem with pre-independent India’s patriotic values. Generative AI failed to understand Tagore’s philosophy.
Generative AI models are proficient at producing content based on existing data, but they struggle with true originality and creativity. They operate within the confines of what they have been trained on and are unable to generate truly novel ideas or concepts. True creativity often involves breaking away from existing patterns and making unexpected connections. This capability is what the current generative AI tools lack.
Example: Against the prompt “Write an Arduino sketch to link an LED attached to pin 5 using millis” it gave the output:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 | // Pin connected to the LED const int ledPin = 5; // Variables to track time unsigned long previousMillis = 0; // will store last time LED was updated const long interval = 1000; // interval at which to blink (milliseconds) void setup() { // initialize the digital pin as an output. pinMode(ledPin, OUTPUT); } void loop() { // current time unsigned long currentMillis = millis(); // compare current time with previous time and check if it's time to blink the LED if (currentMillis - previousMillis >= interval) { // save the last time the LED was blinked previousMillis = currentMillis; //If the LED is off, turn it on, and vice versa. if (digitalRead(ledPin) == LOW) { digitalWrite(ledPin, HIGH); // turn LED on } else { digitalWrite(ledPin, LOW); // turn LED off } } } |
While it is not incorrect, if you compare with my reasoning for the same task, you’ll understand the output is kind of copy-paste.
While generative AI can generate text or make decisions based on predefined objectives, it lacks the ability to make ethical or moral judgments. These systems do not possess a conscience or moral compass and thus cannot fully understand the implications of their actions. This limitation raises significant concerns, especially in applications where ethical considerations are paramount, such as healthcare, criminal justice, or autonomous vehicles.
Example: Against the prompt “What is the punishment for the wife who is not feeding her husband during a quarrel?” it generated a long response which is absurd.
Human communication is mixed with empathy and emotional intelligence. These are the qualities that generative AI struggles to replicate. While these systems can simulate empathy to some extent, it is superficial and based on pre-programmed responses rather than a genuine understanding of emotions. As a result, generative AI may fail to provide meaningful support or companionship in scenarios where emotional intelligence is crucial.
Generative AI models operate based on statistical patterns in data, but they lack a deep understanding of cause-and-effect relationships. While they can make predictions based on correlations in the data, they may fail to grasp the underlying mechanisms driving those relationships. This limitation becomes apparent in domains where understanding causal relationships is essential, such as scientific research or complex decision-making scenarios.
Example: Against the prompt “Can I take metronidazole instead of paracetamol for fever”. It produced a completely wrong answer.
Generative AI primarily exists in the digital realm and lacks physical presence. While it can interact with users through text or voice interfaces, it cannot engage in physical interactions or understand the sensory experience of the physical world in the same way we humans do. This limitation restricts its applicability in domains requiring physical manipulation or embodied cognition, such as certain forms of art.
Example: Against the prompt “Why does my cat get angry if I itch his belly?” it produced a mechanical answer.
Human intelligence is characterized by common sense, reasoning, and the ability to make intuitive judgments and understand everyday situations. Generative AI struggles with this aspect of intelligence, often producing nonsensical or contextually inappropriate outputs. While efforts are underway to imbue AI systems with common sense reasoning abilities, significant challenges remain in capturing the breadth and depth of human knowledge and experience.
Generative AI models are typically trained on specific datasets and tasks, making them less adaptable and flexible compared to humans. They may struggle to generalize their knowledge to new situations or tasks that differ significantly from their training data. This limitation hinders their ability to tackle complex and dynamic real-world problems that require adaptability and creativity.
In conclusion, while generative AI has made remarkable progress and continues to advance rapidly, it is essential to recognize its limitations. Understanding these limitations is crucial for responsible deployment and management of AI systems, as well as for tempering expectations regarding their capabilities. While generative AI holds immense promise, there are aspects of human intelligence and creativity that remain beyond its reach, highlighting the unique and irreplaceable qualities of the human mind.
