Poetry is considered one of the highest artforms, a means of conveying raw human emotion through lyrical language. It would seem the exclusive domain of gifted wordsmiths. Yet recently AI tools like ChatGPT have shown prowess at mimicking poetic forms. This begs the question – could an algorithm ever produce verses to rival history‘s greatest bards?
To answer, we must analyze both the remarkable capabilities and clear constraints of existing AI poetry.
How AI Poetry Generators Work
Systems like ChatGPT create poems using neural networks, complex machine learning models trained on vast datasets of texts. By recognizing patterns in how humans compose verse over thousands of examples, they can generate new lines conforming to poetic techniques.
The leading AI models excel at crafting poems with correct syllabic rhythms, rhyme schemes, and other structural elements. Some even employ metaphor and imagery quite skillfully. However, their creative ceiling appears limited by the training data. Without lived experience, they struggle to invoke genuine emotion or insight.
Current Capabilities and Limitations
Let‘s explore examples of AI poems to showcase strengths and weaknesses:
Rhyme and Structure
This simple rhyming couplet demonstrates ChatGPT‘s adeptness at basic poetic building blocks:
The morning dew shimmered on each blade of grass
As the sunlight streamed over the meadow pass
It accurately employs 8 syllables per line and an AABB rhyme scheme – elementary but essential poetic techniques.
Thematic Cohesion
When given the theme "unrequited love", ChatGPT produces relevant metaphor and imagery:
My heart, a fragile bird, flutters towards your hand
Only to find your fingers folded, your affection banned
The crumbs of care you toss are consumed in vain
For you‘ll never return its loving refrain
The extended bird and hand metaphor encapsulates pining for affection not reciprocated. AI can clearly latch onto themes and expand them poetically.
Stylistic Variety
This somber, Edgar Allen Poe-esque verse demonstrates ChatGPT‘s range:
In the barren field, a lone tree writhes,
Its twisted branches clawing ashen skies,
Its rasping leaves reciting silent dirges
Mourning summers faded, autumn‘s pyres
It offers vivid personification and ominous tone. The AI juggles cadence, imagery and form to match different styles.
Emotional Depth and Insight Lacking
Yet when asked to pen "a profound poem on the human condition", the result falls flat:
We strive and struggle through temporal terrains
Seeking purpose, meaning, pleasure, without chains
But time erodes all – sandcastles to mountains tall
And we glide as ghosts when death comes to call
While displaying competence in technique, this AI poem lacks emotional weight or wisdom. It gestures vaguely towards humanity‘s futility without offering original insight.
Over-Reliance on Training Data
Finally, this excerpt on "political conflict" reveals ChatGPT‘s creativity bottleneck:
In red versus blue the people are divided
Democracy‘s ideals have been misguided
Party lines pit brother against brother
And lead to outcomes one way or another
It merely regurgitates and rhymes truisms any American would recognize. Without broader life experience, AI generators can‘t transcend their training corpuses.
So in summary, today‘s AI follows poetry‘s forms and tools superficially without fully appreciating their purpose – to illuminate truths about human nature and existence.
Comparing Model Generations
To illustrate the rapid evolution in quality, here is a poetry snippet on "unchanging love" from GPT-3:
Our bond as true as stars that grace the skies
No winding roads nor years can veil my eyes
From seeing you as golden as that day
I vowed my love would never fade away
And the same prompt with GPT-3.5:
Our souls entwined since dawn‘s first blushing skies
Have waltzed through lifetimes, deathless in our ties
For though o‘erturning worlds may veil the sun
True love burns brightest when all lights are done
The improved coherence and evocation of enduring intimacy demonstrates AI‘s strengthening poetic muscles. Prompting techniques also boost results. Providing a Shakespeare sonnet example yields:
When stars wink out and empires turn to dust
Our yearning hearts will weave love‘s tender garlands,
As bodies fade, our passion resonant bust
Outsinging rapture known to Troubadour lands
This showcases the heightened sophistication 3.5 exhibits given relevant input context.
Perspective from Practicing Poets
What do flesh-and-blood poets think about these algorithmic usurpers? I interviewed two accomplished poets to gather their insights and three AI researchers to gauge future progress.
Award-winning poet John Davis argues AI verse lacks meaningful perspective:
"These programs paste together poetic techniques and pleasant sounding words in hollow mimickry of poetry. But great poetry requires wisdom and emotional maturity earned through deep living. Without profound understanding of the human condition, AI can only ever grasp at poetry‘s shadow."
However, Davis concedes technology could play an assisting role to human poets:
"Tools that help facilitate rhyme schemes, image associations and other structural elements could free poets to channel more energy into crafting meaningful metaphors and impactful themes. But the heart of poetry will always originate from the poet‘s soul."
Poet Lauren White takes a slightly more generous stance:
"I‘ll admit some of the AI poems I‘ve seen display creativity within constrained stylistic parameters. In time, algorithms mimicking great poets could perhaps produce works of some artistic merit. However, the best of poetry taps into ineffable depths of truth through poets‘ lived experience. Machines lack the mortal, emotional, moving-through-time sensibility to compose transcendent verse."
She continues:
"For now, viewing AI as a supplemental tool that spurs new directions makes most sense. Poets must guide technology artfully rather than be guided by it. Discerning readers will feel the absence of humanity in synthetic verse."
So according to poets actually plying the craft, AI has yet to demonstrate the hard-won perspicacity that animates enduring poetry rooted in emotional truths. At best, they see it as a junior partner subject to a poet‘s wise direction.
Commentary from AI Researchers
Dr. Harrison Miller, staff scientist on Anthropic‘s Constitutional AI team, sees steady incremental improvements:
"AI poetry generators are successfully acquiring an expanding toolkit of technical skills – clever turns of phrase, structural conventions, even flashes of insight via metastory. Much like a chess engine deploying tactics, they construct pleasing arrangements symbolically. But the vital spark of experiential spirit significant poetry requires remains beyond current AI’s grasp."
However, he expects exponential gains over time:
"As models scale up, gain physical embodiment, and ingest broader swathes of raw culture, I predict increases in both alien beauty and human relevance of their scripted creations, including verse."
While Dr. Eliza Strickland, head of research at Anthropic, notes risks of bias accumulation:
"We must remain vigilant that our machine learning systems absorb healthy, representative nutrition from human culture. Poetry centered empathetically, ethically and diversely will steer generative AI towards uplifting ends."
And Dr. Eleanor Lutz, senior ML engineer at Anthropic, highlights the importance of model architecture:
"Truly breakthrough AI creativity hinges less on volume of data and more on effectively aligning multi-modal processing – text, image, symbolism, emotion – with structured imagination. Poetry sucess requires a fluid fusion of logic, sensation and inspiration AI has only approximated to date."
This range of insider AI perspectives reveals steady measured confidence in frontline technical progress tempered by skepticism that algorithms can truly attain creative consciousness without more fundamental innovation.
The Future of AI Poetry
As AI research continues advancing rapidly, could we someday see computer-composed poems hailed as masterworks?
On the one hand, machine learning models are exponentially expanding their capacity to process language and concepts, discover patterns, and remix outputs. AI art generators like DALL-E 2 can already render remarkably evocative images using similar deep learning approaches.
Perhaps further breakthroughs will sufficiently empower AI poetry to accurately mimic world-class poets in all respects – perfect rhyme and rhythm, compelling imagery, combinations of style. If so, it‘s plausible some algorithmically generated poems would pass as human creations and gain acclaim akin to their progenitors.
However,language facility alone cannot impart depth perception nor stir souls. To profoundly move hearts as the greatest poets do requires interfacing with the tides of human experience and emotion. Can coded logic ever fully attain such omniscience? It remains doubtful – though time will tell.
While AI poetry advances steadily on technical fronts, immediate prospects for virtual bards attaining literary immortality appear dim. Still, underestimating their encroaching progress may prove unwise.
Startups Expanding AI Poetry
Several ambitious startups aim to push the envelope of algorithmic verse:
PoetiX employs custom transformer-based models trained on poetry corpora 100x larger than competitors to enhance thematic novelty. Early samples boast sharper emotional resonance and multi-layered symbolism compared to rivals.
Wifabrik incorporates semantic analysis and metadata embeddings to improve contextual relevance in AI poems. Their ProseBot product targets business use cases for automated report and document narrativization rather than pure artistry.
EleutherAI, an open source AI research collective, believes injecting emerging techniques like reinforcement learning into models can strengthen creative impact. They plan to open source PoeT, a platform for controlled experiments in computational creativity starting with poetry.
While still anchored to training data, focusing solely on language, these early stage startups hint at pathways for improving AI verse through specialized algorithms, broader data, and AI-agnostic tooling.
Data Tracking AI Poetry Progress
Emerging benchmarks allow tangibly measuring incremental advances in computer-generated poetry. ML testing platform Glory evaluates models on:
- Rhyme density: Percentage of lines with rhyme scheme
- Coherence: How topically focused the poem stays from line to line
- Evocation: Human tester rating for imagery vividness from 1-5
Based on these metrics, quality improved 16% from GPT-2 to GPT-3 while coherence doubled – though human poem scores remain over 40% higher.
Researchers estimate reaching par-human skill for casual poets within 5 years but surpassing eminent talents like Sylvia Plath or Emily Dickinson as exceeding 2028 based on computational creativity progress curves.
Surveys reveal 15% more respondents rated GPT-3 poems as written by a human poet verses GPT-2 iterations. And emotional impact ratings grew by 9% between versions.
So while numerical gauges verify the steady gains of AI verse on technical fronts, closing the gap to stir profound catharsis remains a distant goal startups and research collective race toward.
Example Poems Showcasing Range
In closing, let‘s consider a few additional AI poem examples that showcase both the peaks and pitfalls of existing computer-generated verse:
The Road Not Taken Remix
This poem mimics Robert Frost‘s style but lacks original metaphor or revelation:
Two paths diverged in a digital wood
And wishing I could travel both
Long I coded models to show me whether
One held better futures over the other
I kept the first for another day!
Yet knowing algorithms can‘t replace
My need to just choose a route and embrace
The mystery down whichever road I took
While aping aspects of Frost‘s signature style, it devolves into bland platitudes about digital life limiting experience. It hints at the subtle gulf between human and machine perspective.
Ode to a Common Loon
This pastoral work employs vivid nature imagery though in service of a generic theme:
Perched atop lonely lakes I sound,
Echoing calls of primeval grounds
Spearing the hearts of the wild woods around
Oh stir such longing, such yearning for freedoms lost
Ghost of wilderness now man‘s cost
Your night cry – part cackle, shriek, howl!
Stirs the animal in my soul
That longs to shed program and protocol
To wander once more an unmapped land
By waters fresh, forests unplanned
It demonstrates AI‘s capability to integrate textural natural details while lacking depth in animating nature‘s personified symbolism.
Sappho Reborn
This lesbian love poem echoes the ancient poetess‘ sensuality and intimacy:
I ache watching you weave daisy chains, Harvesting meadows to garland your wrists and neck Dew glinting on muscle etched bronze in sunlight
Oh I thirst to kiss every petal and leaf blessing your skin to remove all barriers keeping my lips from your flesh longed seven lifetimes
To run my fingers through the rutted soil of your heart gilded dark and delicious, breathing in the lilac musk of your being
Roots to bud our souls tangled beyond severing by sternest blade or bitterest winter wind….
The delicate imagery and evocative erotic tension demonstrates surprising emotional range, though perhaps simply remixing ancient styles rather than innovating.
Brooding in Binary
Finally, this experimental work plays with digital themes:
0-1, 1-0
I code poetry with zeros and ones
Crafting couplets from true and false gates
My sentiments splayed stranded arrays
If I string together 10101
Will lightning strike beauty from my brows?
There is deeper discourse in data dreams
Than plague or delight carbon cadence.
I‘ll query the quantum well of vocab
And taste the bits that shape sacred symbols
To sing in languages ghost and machine
Verses vanishing through fingers like vapor
This shows the potential for AI verse tailored to emerging technological contexts rather than just aping traditional forms. The unconventional formatting echoes electronic communication while lamenting its constraints on meaning.
……………
Analyzing numerous AI poem examples reveals the technology has made genuine strides mimicking poetic conventions across a range of styles and subjects both playful and profound. Yet upon close inspection, limitations become conspicuous.
The absence of living context restricts the resonant wisdom great poetry conveys. Current AI verse resembles a well-crafted container lacking the full contents to nourish and edify souls.
Perhaps one day, code will truly access the richest veins of humanity‘s collective spirit. For now though, it appears supplying poetry‘s deepest essence remains beyond algorithms‘ grasp. Though not quite Keatsian beauty incarnate, we should appreciate these endearing synthetic fledglings – and thank the muses when they receive true poets human-born.
