The idea that machines could someday mimic human creativity has long captivated our imaginations. From science fiction on depictions of androids painting masterpieces or composing symphonies to modern-day fears that AI will automate creative jobs, the notion persists that artificial intelligence may one day match, or even exceed, the human capacity for creative thought.
But what does it mean for AI to be truly creative? And how close are we to developing artificially intelligent systems with their creative spark?
To explore these questions, let’s examine how AI produces original art, music, writing, and other creative output.
Defining Creative Intelligence
First, we must define what exactly is meant by creativity. Psychologists describe creativity as the ability to develop ideas or products that are novel and appropriate to the task or problem. Creative intelligence involves divergent thinking, intuition, and the capacity to make new connections between concepts or ideas that may not relate.
So, for AI to be considered creative, it would need to fulfill both conditions:
1. Generate original output, not just duplicate or recombine existing works
2. Produce output that meets quality standards and has value to human observers.
This also implies emotional engagement – truly creative works often stimulate an emotional reaction in the viewer, reader, or listener.
The Current State of Creative AI
AI has made impressive advances in replicating human creativity in recent years thanks to machine learning techniques like neural networks and generative adversarial networks (GANs). These methods allow AIs to generate new outputs after “learning” the patterns in huge datasets of existing art, literature, or music works.
AI musical tools like Aiva can compose original scores and melodies based on the styles and techniques used by famous historical composers.
Users specify the instruments, genre, mood, and length, and the AI generates an audio file of a new song meeting the criteria. Companies like Anthropic and Meta have also developed AIs that can compose music and improvise in real-time in response to human musicians.
Again, though, while pleasing to the ear, some question if algorithmically combining musical theory qualifies as creativity in the human sense. Can creativity exist without a conscious experience behind the output?
AI Writing and Poetry
After learning the patterns in immense writing volumes, tools like Jasper and Shortly Read can craft blogs, articles, stories, and even poetry on the text-generation front.
For example, Shortly Read uses GPT-3 to generate poetry based on a title and author name. While the poems seem well-formed, they are more imitative patchwork than works of imagination.
Some issues around AI writing include incoherent narratives, offensive outputs, and the potential to automate journalism jobs. But others argue creative tools augment creativity rather than replace it – AI still requires human prompting and curation.
Assessing Creative AI
From the output of today’s state-of-the-art creative AIs, we can observe:
They display combinational creativity – intelligent recombination of known elements and styles in new ways. But true creative imagination – inventing unprecedented ideas and making meaning through self-expression, emotion, and beliefs – remains lacking.
Outputs can correctly follow domain patterns like brush strokes, musical chord progressions, etc. Yet they lack global coherence – a unifying vision or message. Generators cannot maintain an artistic style or storyline throughout an image, song, or story.
So, by current benchmarks of creativity – originality, cohesion, emotional resonance – AI still falls short of human creativity. Yet rapid progress hints these deficits may not persist.
Pathways to True Creative AI
If AI hopes to attain human levels of creative intelligence – or even surpass it – continued breakthroughs are needed in a few domains:
- Conceptual Understanding: While AI can statistically analyze patterns, truly understanding art and music requires grappling with abstract ideas like emotion, significance, beauty, etc. Advances in modeling symbolism, causality, emotions, and building common sense knowledge will help move towards this goal.
- Agency and Self-Supervision: We hypothesize creative people have a drive to create for intrinsic motivations inside themselves. AI today mostly optimizes outputs based on external rewards and specifications from humans. Developing AI that can learn, experiment, and define creative goals for itself may uncover more profound creativity.
- Embodied Models: Intelligence originates from experiences and sensorimotor processing in our embodied brains. Disembodied AI today misses crucial contextual understanding permeating human creativity. Continued progress in simulations and transferring AI to robotics may provide missing pieces.
- Collaborative Creativity: Creativity often emerges through the interplay between multiple minds via collaboration, criticism, competition, etc. Building collective and social AI with feedback loops may yield new creative explosions.
The Future of Creative AI
While AI has already displayed narrow combinational creativity, the coming decades may reveal far more human-like or alien creative potential from artificial neural networks. Such advances promise to disrupt many creative sectors – transforming how art, music, stories, design, and more are produced. We will witness the birth of new creative industries producing AI art, characters, symphonies, literature, and interfaces with little human input needed.
The Bottom Line
While AI today has limited creativity compared to humans, rapid advances in machine learning are quickly bridging that gap. AI art and music tools now showcase rudimentary combinational creativity, though they still lack deeper meaning, emotion, and vision. It’s unclear if computational creativity will rival human imagination. Still, the pace of progress suggests we’ll one day coexist with machines capable of generating novel, evocative art reflecting their alien creative spark.