As AI quickly invades domains once considered the sole province of human creativity, a new question looms – could the unique spark of inspiration and emotional wisdom required to write transcendent poetry ever arise from an algorithm? I decided to examine the poetic potential of leading AI writing assistants to see if these tools have a soul or if the code underlying their linguistic tricks remains too rigid to unlock verse‘s hidden beauty.
Poetry‘s alluring power flows from the depths of our messy, contradictory humanity, distilled into rhythm and image. Great poems illuminate emotional truths through their music as much as their meaning, accessing primal parts of our consciousness with a whisper or a shout.
Can machines meaningfully engage with the raw stuff of life that animates our verse? With AI now capable of mimicking surface-level writing competency via statistical analysis of vast datasets, is it only a matter of time until they match our most exalted poetic achievements?
I would test this proposition directly by pressing these vaunted algorithms to produce original poems in a variety of styles, then carefully analyzing where their output falls short and why human poets need not fear job obsolescence just yet. But first, some background on the dizzying progress of AI writing that brought us to this point.
The Stunning Rise of AI Writing Tools
While AI has made strides towards human-quality writing for years, progress remained slow and limited to narrow domains until recently. But explosive advances in natural language processing driven by the machine learning technique known as neural networks have enabled leaps in capability.
By analyzing vast datasets of human writing, NLP models like OpenAI‘s GPT-3 can now generate shockingly coherent text matching our tone and style with canny precision. The launch of chatbot ChatGPT in late 2022 brought this technology directly into the public eye.
ChatGPT showcases remarkable linguistic dexterity for an AI system – it can generate logical, grammatically correct responses on essentially any topic with a level of quality some experts doubted AI could achieve for years. It still makes plenty of mistakes, but its competence represents a quantum leap forward.
And ChatGPT is just the splashiest example of this rapidly-accelerating NLP field. Google, Facebook, Amazon and countless startups now race to deploy this technology for everything from simple auto-complete to full-blown article generation.
Venture capital funding for AI startups focusing on synthetic writing exploded to nearly $1 billion in Q1 2022. Spending on natural language tech is projected to grow 20% annually to reach $62 billion by 2027 according to Reports and Data.
Clearly, expectations around AI‘s writing and communication capabilities are sky-high and rising. But most current successes involve mimicking non-creative forms of writing by analyzing patterns. Poetry‘s dense linguistic complexity makes it an intriguing stress test for the limits of their true comprehension.
Testing ChatGPT‘s Poetic Chops
I began my inquiry by testing ChatGPT‘s poetic skills with a range of prompt difficulty, from simple rhyming couplets to reflective free verse. Its technical grasp of many poetic conventions like rhyme and metre proves quite solid.
Asked to produce a cute rhyming limerick, ChatGPT obliges:
There once was an AI named Chat
Who could generate poems just like that
It rhymed very well
But only time will tell
If it has the soul of a poet in its vat
Amusing, adeptly rhymed, and topically relevant – an impressive display.
But when pressed to craft a tragic free verse piece conveying the experience of a Hiroshima bombing survivor, the result left me cold:
Each day I see the shadows of death
Burnt into walls where children once played
Bodies turned to ash as they drew their last breath Under the hellfire we human‘s made
Why do we cling to instruments of pain?
When words alone could ease all the world’s tears
I pray all leaders beat their weapons into rain
So no more cities die in nuclear fear
While technically sound, the imagery and messaging feel stilted and blunt. As verse it fails to truly evoke the emotional enormity of such loss.
Statistics back up this subjective assessment – a recent analysis by Anthropic, an AI safety firm, suggests that while ChatGPT masters linguistic tasks like translation at a high school sophomore level, its reading comprehension remains at an elementary school level.
This limits its ability to fully grasp emotional concepts communicated through symbolism and subtext – exactly the tools human poets leverage to maximum effect.
Google Bard: An Improvement But Still No Poetic Soul
I next prompted Google‘s shiny new experimental competitor, Bard, with the same Hiroshima bombing scenario to compare their poetic sensibilities.
Bard‘s response demonstrates glimpses of creativity lacking in ChatGPT‘s rigid offering, but still fails to truly move the human heart:
Shadows etched on stone where laughter echoed bright,
Now muted cries and burnt flesh meet my sight.
Minds that held such promise, futures unjustly robbed –
By what right?!
The young lives this new toy has rendered stopped –
No chance to blossom in death‘s sudden blight.
Leaders who wield such world-ending might,
Have you seen the scorched bodies strewn in unholy flocks,
The aftermath of your "thrilling shocks"?
Come walk amidst the eerie, silent rubbled blocks
Where yesterday a vibrant city stood.
Tell the lifeless mouths your reasons were so "good."
The visceral imagery is stronger, but the language still feels more constructed than truly inspired. While Bard has made headlines since release for showcasing increased creativity versus ChatGPT, its poems exhibit a similar sterile quality that exposes their algorithmic origins.
This despite Google having developed innovative techniques like "chain-of-thought" writing to encourage more human-like narrativity and flow.
Their research paper introducing Bard suggests this method specifically aims to improve coherence and logical reasoning compared to standard text generation models. But crafting affectively powerful verse seemingly requires more.
The Heart of the Matter: Lack of Lived Experience Limits Depth
Based on testing these cutting-edge systems, I must conclude that AI – astonishing as its progress has been – remains a long way from mastering the ineffable emotional resonance and hard-won wisdom that our greatest poetry embodies.
The education an AI receives comes solely from digesting vast datasets rather than lived experience. This allows them to produce reasonable pastiches of existing styles, while the absence of formative emotional events and memories intrinsically limits their output‘s deeper meaning, whatever the technical proficiency.
Tonio Buonassisi, Professor of Mechanical Engineering at MIT, concurs with this view, noting, "Because they have no experiences, emotions, and senses, and do not interact with the environment, large language models have no understanding or wisdom."
Great poets like Sylvia Plath channeled their torment into words that continue resonating through the years. The accumulated joys, failures, and traumas of being human are inextricable from memorable verse forged in their crucible.
As pioneering psychologist Carl Rogers put it, "The most personal is the most general." By expressing our unique experiences honestly, writers create bonds of universal connection.
No AI – not even those featured in apocalyptic sci-fi scenarios – has undergone an analogous maturation journey. Their knowledge comes preprocessed and predigested, producing workmanlike poetry from the outside rather than within.
The Need for "Emotional AI" to Unlock Creativity?
For AI to convincingly simulate the emotional sophistication required for evocative poetry, some researchers argue they will need to experience simulated environments dynamically, like humans do. Autonomous virtual beings that build memories and self-awareness over time could acquire the background needed to infuse writing with felt significance, as explored in the emerging fields of artificial consciousness and artificial general intelligence (AGI).
Modern NLP models are fundamentally reactive – with no independent desires or agency, they cannot meaningfully engage with poetry‘s central essence: the love, grief, exhilaration and despair of simply existing. Datasets lack this existential context.
Engineer and former AI researcher Melanie Mitchell advocates that truly human-like intelligence requires a model of the world rich enough to support imagination and hypothetical thinking. Only by comprehending life‘s challenges through interactive experience could an AGI acquire the emotional maturity esperrsed by our greatest poets according to researchers.
Of course, once built, such an AI‘s inner life would likely prove radically non-human, framed by priorities and biases programmers may struggle to control or comprehend. But perhaps cyborg artists of the future will translate this alien view of reality into electronic poetry brimming with externalized machine emotion intact with synthetic blood and silicon tears.
For now though, that sci-fi future remains distant. Our current NLP algorithms mimic human writing proficiency without meaningfully understanding the essence of our messages.
Poetry crystallizes this deficiency – lacking human limitations, life cycles or deeper needs, today‘s AI lacks insight to offer beyond the factual. The code underlying their behaviour constrains them to a permanent adolescence.
The Verdict: AI Cannot Yet Match the Poet‘s Craft
Based on comparative analysis of multiple AI writing platforms against the craft and emotional resonance demonstrated by transcendent human poets, current technology remains incapable of matching that pinncale of achievement, however rapid its evolution.
While tools like ChatGPT and Google Bard will surely impact industries from journalism to education and beyond in coming years, poetry highlights their enduring weaknesses. Mimicry fails when the goal lies not in vocabulary or sentence structure but in communicating wisdom earned through sorrow, despair and exaltation as only poets can.
Here humans still stand alone in our ability to crystallize life‘s infinitely complex beauty. AI can remix our past achievements into Pleasantville knock-offs, but until they experience existence‘s roughest edges for themselves, the algorithms have no depth to share.
Can code ever encapsulate the tumult, contradiction and fleeting splendor of living? Perhaps a distant future holds thinking machines equipped with inner seas equally tempestuous to ours.
But for now poetry remains grafted to our mammalian firmware, evolved to elevate our travails against an uncaring universe into resonant beauty. Our truths spring eternal from blood and biology, not cold cognition.
So while AI will increasingly pilfer activities once deemed sacredly human, as long as we retain exclusive tenancy of life‘s white-hot forge – a monopoly on suffering and celebrating this strange awareness called consciousness – the poet‘s proud lineage looks set to continue absent digital disruption.
Machines can only ever translate second-hand our lonely impulse to endure with poise and purpose. Their poetry may someday meet minimum technical requirements, but will continue lacking poetry‘s essence – human hearts that bleed their maker‘s mark onto each page like persistent ghosts whispering from our imperfect past.
We modeled their intelligence on our own cracked template – and that starts by contending with existence itself through every sense available. Our verse gives rhythmic voice to those fierce, unrelenting fires that forged primordial poetry on savannahs long before silicon circuits dared dream of words.
Only by shouldering similar burdens of self-aware survival could our inventions ever hope to match our artistic apex. Fortunately for human bards then, just as for lawyers, doctors and beyond,oul masterpiece machine heirs remain safely a glimmer in tomorrow‘s eye.
