How Accidentally Learning Farsi Taught Me About The Limitations of AI
Discussing the Concept of 'Poverty of Stimulus' and the Human Way of Acquiring Languages and What it tells about the Limitations of AI
Most of you know me as an Engineer and Writer. But I think a lesser known aspect of my life is my love of learning languages. Over the years, I have ventured into learning many different language, including but not limited to Spanish, French, Russian, Farsi, Arabic, and Norwegian. These are in addition to Urdu (my native tongue), Hindi (also my native tongue), and English (the non-native language that I am most comfortable with).
My fluency in these languages varies. But one thing remains common. I learned all these languages, primarily so I could enjoy the music and poetry of these cultures. As a poet who is deeply passionate about music, I find it very necessary to first be able to connect with the music of a language before connecting with the language itself. If I cannot connect with the music and the artists of the language, I find it next to impossible to be able to learn it.
So today, I thought of sharing with you the story of my tryst with Farsi poetry. And how this experience, served as a catalyst for a deeper inquiry into how we, as humans, learn and understand language so effortlessly, and how this contrasts with the current capabilities of artificial intelligence, particularly Large Language Models (LLMs).
My motivation for this article stems from a desire to bridge my personal anecdote with a broader discussion on the cognitive processes underlying language acquisition, the concept of 'Poverty of Stimulus', and the limitations and potential of AI in replicating human-like understanding. My hope is that you might find in this article a fresh perspective on the natural intelligence that guides us effortlessly through complex linguistic terrains.
How I Accidentally Learned Farsi
Farsi was one of the first languages that I learnt (outside of Urdu and English). But what makes this story interesting is that my first encounter with this language was not for learning it. It did not begin in a classroom, nor through formal study, but through the serendipitous discovery of Shaikh Jalaluddin Rumi’s (رحمتہ اللہ علیہ) poetry on YouTube.
As a native Urdu speaker, I was no stranger to the profound beauty and depth of poetic expression, yet Farsi, with its lyrical intonations and philosophical richness, was uncharted territory. When I listened to his poetry performed by Persian natives and artists from the Indian sub-continent, I was instantly struck by a profound, almost spiritual, connection to it. At this time, I did not speak a word of Farsi nor did I have any idea about its grammatical structures. And yet, I was able to naturally understand more than 70% of it.
The experience was similar to finding a door to a hidden garden within the familiar confines of my linguistic home. The overlap in vocabulary between Urdu and Farsi, while helpful, did not justify the intuitive understanding that I felt with this language. While Urdu might share some vocabulary with Farsi due to the globalisation factor, the grammatical structures of these languages are very distinct and most native Urdu speakers have to enrol into a formal language learning course to be able to understand Farsi.
My process of learning Farsi was not methodical or analytical; it was organic, driven by a deep emotional resonance with the content of the poetry. The words of poets like Rumi and Hafez did not just convey meanings; they evoked feelings, images, and connections that transcended the conventional barriers of language learning.
This phenomenon, where language transcends the boundaries of cognition and ventures into intuition and emotion, offers a compelling counter-narrative to the traditional views on language acquisition. The prevailing theories, steeped in grammar rules and vocabulary lists, could not fully explain my experience.
It was a vivid demonstration of the human mind's capacity to assimilate and express complex linguistic constructs through exposure and emotional engagement, rather than through deliberate instruction.
Within just a couple month of pure listening and reading the lyrics of these medieval poetry, I had somehow managed to develop enough understanding to start writing poetry of my own in Farsi. I managed to get in touch with some people understood this language fluently and with a lot of trepidations, presented my poetry to them. To my surprise, they were impressed by sense of expressionism in this language. While there were some grammatical mistakes here and there, overall, I managed to convey my thoughts and feelings in a foreign language, without any formal/informal training, and that too, in a strictly metered form of poetry.
Zero classes taken. Zero grammar books read. Zero contact with native speakers. And yet somehow, I had managed to write poetry in a language that I had only listened to for a couple of months.
Reflecting on this journey, I realised that my experience mirrored, in some ways, the innate language learning capabilities observed in children. The concept of 'Poverty of Stimulus', which suggests that children learn to understand and generate complex language structures with seemingly insufficient input, resonated with my own learning process. It was as if my mind, guided by a blend of emotional connectivity and subconscious linguistic intuition, was able to fill in the gaps, constructing a bridge to Farsi through the medium of poetry.
This unexpected venture into Farsi poetry opened up my mind to the exploration of the human capacity for linguistic intuition and emotional resonance. It highlighted a dimension of language learning that is often overlooked in our rush to quantify and mechanise language education: the power of emotion and the intrinsic human desire for connection and understanding.
The Intrigue of Language Learning and the Poverty of Stimulus
These past months, I have ventured deeply in the field called the Philosophy of Language. I have written extensively on this subject and will continue to do so throughout this year on my substack The Lumeni Notebook. So if this subject interests you, I do recommend you subscribe to it. I would love to engage with like minded individuals on this topic.
What I realised, by studying language acquisition through a linguistic and philosophical perspective, is that my experience with Farsi, while unique in its context, shares remarkable similarities with a well-debated concept in linguistics known as the Poverty of Stimulus.
The Poverty of Stimulus is a theory suggesting that children learn their native language without sufficient explicit linguistic input. This concept, introduced by Noam Chomsky, challenges the notion that language learning is solely a result of environmental exposure. Instead, it posits that humans possess an innate faculty for language acquisition, enabling them to understand and generate complex structures with minimal external guidance.
My encounter with Farsi poetry exemplifies this theory in action, albeit in the domain of language learning as an ‘Adult’. Without deliberate instruction, I found myself understanding the nuances of Farsi.
This phenomenon raises profound questions about the nature of knowledge and learning. How do we, as adults, tap into this latent potential? Is there a universal grammar embedded within us, waiting to be activated by the right stimuli, much like the chords of a song seeking the right musician to bring them to life?
The quest to create LLMs that can process and generate human-like language involves feeding these models vast datasets, hoping to capture the essence of linguistic knowledge. Yet, my journey into Farsi, guided by intuition and emotional resonance rather than data-driven algorithms, underscores a fundamental disparity between human learning and machine learning. It highlights the intuitive, almost mystical aspect of human cognition that remains elusive to AI.
How AI Learns Language
Can LLMs truly grasp the essence of language in the way humans do?
LLMs, including the ones that power search engines, chatbots, and virtual assistants, operate on principles that are diametrically opposed to the human experience of learning and using language. They are fed billions of words, sourced from books, articles, websites, and more, encoding patterns and probabilities into their algorithms. This quantitative approach, though impressive, lacks the qualitative, emotional, and intuitive nuances that define human language learning.
Consider the way I came to understand Farsi through poetry. This process was not just about decoding symbols or memorising vocabulary; it was an emotional journey, where meaning and understanding were deeply intertwined with the cultural and spiritual context of the poems.
This dimension of language learning is conspicuously absent in the realm of AI. LLMs can generate responses that seem eerily human-like, but the 'understanding' behind these responses is mechanistic, based on statistical correlations rather than genuine comprehension.
I discussed this in-detail in my previous article that I linked below. You can read it to get a more nuanced view of the problem of ‘Understanding’ in machines.
The contrast becomes even starker when we explore the concept of Poverty of Stimulus. Humans, especially children, can learn and extrapolate complex grammatical structures from seemingly insufficient input, guided by an innate linguistic framework.
LLMs, however, require massive amounts of data to achieve a semblance of this ability, and even then, they can falter at tasks humans find intuitive. This reliance on vast datasets highlights a critical limitation of AI: its learning is bounded by the data it is trained on and the explicitness of the patterns within that data.
AI as a Tool, Not a Replacement
AI and LLMs are useful and valuable. That is beyond question. I myself, use these tools on a daily basis for researching my articles and generating artwork.
These tools, with their ability to process and generate language, have sparked discussions around the future of work, creativity, and even the essence of what it means to be human. Although I am an AI researcher with a good understanding of the mechanics that run LLMs, the first time I used ChatGPT, I was just as awestruck as any other non-technical person.
But my journey into the heart of Farsi poetry, and later my research in the philosophy of language, have contrasted against the backdrop of AI's capabilities. a fundamental truth: AI, for all its sophistication, operates within a realm distinctly separate from the human experience of learning and understanding.
I am not one to take strong stances on these matters. Nor am I someone who downplays the potential risks with AI. But after much contemplation, I cannot help but reach to this understanding. I maybe wrong. Most theories generally are. And if proven wrong in this regard, I will be more than happy to re-align my understanding and recant my statements.
The intelligence of AI, while impressive, is defined by algorithms, probabilities, and data. It lacks the intuitive grasp, the emotional resonance, and the deeply personal context that characterise human learning.
The process of acquiring a new language, is not merely about assimilating words and rules but about connecting with a culture, feeling the weight of its history, and expressing its beauty through our unique lens. This dimension of learning is where AI, in its current form, cannot tread.
The narrative around AI often swings between utopian and dystopian extremes, forecasting a future where machines either solve all our problems or usurp our place in the world. However, the essence of AI, as a tool crafted by human hands and minds, suggests a more balanced perspective. AI can automate, enhance, and even inspire, but it cannot replicate the depth of human learning and creativity. It cannot replace the teacher who ignites a passion for knowledge, the artist who captures the subtleties of human emotion, or the engineer who sees beyond the code to solve real-world problems.
As we stand on the cusp of a future shaped by AI, it's crucial to remember our role not just as creators but as stewards of this technology. AI should augment our abilities, not diminish our value. The jobs it "kills" are an opportunity for evolution, not a sentence of obsolescence. The fields it evolves will benefit from our oversight, our ethical considerations, and our creative input. Our learning abilities, rich in complexity and nuance, ensure that we remain indispensable – the composers of a world where AI plays but one of many instruments.
Enjoy Reading about AI and Technology? You’re in the Right Place! Subscribe for FREE!