Who's More Creative: AI or Humans?

The featured image of this article, titled The Electrician, is a captivating piece of art generated by photographer and artist Boris Eldagsen with the aid of artificial intelligence (AI), which recently got the championship in the Sony World Photography Awards.

This marks a momentous achievement in AI, pushing beyond its traditional role as an analytical tool and showcasing its potential to generate aesthetic creativity.

Though it is not entirely a novelty, precedented by the victories of AlphaGo and the iconic Deep Blue, it is nonetheless significant and points to the dawn of a new epoch.

Enthusiasts are effervescent with anticipation, eagerly foreseeing the expansion of AI into uncharted realms of creativity.

Nevertheless, the advent of AI-generated content raises concern about the future of original content creators, as some see it as a harbinger of "computerized plagiarism" on a massive scale.

While numerous copyright litigations await final decisions in the court, the matter remains controversial.

As technology progresses, we must engage in meaningful discussions regarding the use of AI in creative pursuits.

The potential for AI to disrupt conventional approaches to art and creativity is undeniable, but whether it can genuinely rival human innovation or merely represent a more advanced form of plagiarism is yet to be seen.

Not Generative AI but Derived, Synthetic AI

Plagiarism is an unwavering taboo. The commitment to this serious misconduct can have a devastating impact on the entire careers of diligent creators, whether they are involved in academic research, music composition, article writing, or image creation.

It is not only confined to a word-for-word or a pixel-for-pixel copy but is also applicable in cases where the content is merely paraphrased, or the structure is manipulated from the same piece of artifact.

The perpetuation of plagiarism counts on the vast amount of data sources that can be copied from the internet.

Given the human mind's finite capacity to remember unlimited information, aspirants may seek the aid of fast computers and expansive local and cloud storage to facilitate their plagiarism.

Now, with the technological evolution of hardware, the development of AI—more specifically, generative AI models that create fresh content—is being furnished.

Generative AI models are pre-trained from a ginormous corpus of text, images, video clips, and audio tracks, depending on their types.

For example, the Stable Diffusion text-to-image model has been trained on various subsets of LAION 5B, a large dataset comprising nearly 6 billion CLIP-filtered image-text pairs; other competing models like DALL-E, Midjourney, and Imagen are similar.

Source images are crawled from numerous online domains, from stock image sites and art print and poster marketplaces to social media platforms and blog posts.

By recitation of data, models are supported with an extensive cogent structure—new content derived from multiple image sources by just a text prompt in seconds.

You may further study the technical logic of Stable Diffusion and other models if interested, but I want to illustrate that perceiving this as "generating" new content is not always accurate.

Instead, these models synthesize the contents based on whatever they have learned from the pool of pre-existing data fed by humans.

Processing enormous amounts of data is manageable for the models because they are equipped with advanced computing power, unlike humans, who cannot deal with billions of images one by one.

Some of the synthetic by-products—the generated images—look very familiar to most people because of the source data the model picked; be it a famous painting or a classic meme.

On the contrary, some of the contents may seem rare and original.

Marred by the nature of synthesizing, generative AI at the moment is a rhetorical speech of marketers, despite its robustness. The vigorous dichotomy about proper source data collection implies "AI creativity" is not from origination but derivation.

That said, generative AI's formidable mix-and-match capability has been convincing, illuming people as AI creativity.

The Four Stages of AI Creativity

Leading to the quest for the essence of AI creativity, the extent of AI autonomy without human intervention is one objective assessment criterion for me. AI creativity can be roughly classified into four stages, according to the degree of involvement:

1. Fully Human-Instructed Creativity

Traditionally, humans have been creative since the primitive age. Most creative works are created for survival and fulfilling basic physical needs.

The discovery of fire in ancient times brought humans warmth, cooked food, and safety; then, the wooden drill was invented as a kindling device inspired by natural fire.

Fast forward to The Renaissance and beyond The Industrial Revolution, human creativity spans different subjects—literature, art, science, mechanics, etc.—to computers and AI.

The mere usage of computers and AI does not inaugurate AI creativity. Analogize it with pens and pencils. It is the artist—which is a human—who uses pens and pencils as tools to draw a picture. The origination comes from the artist's imagination, not from the instruments.

So, using computers and AI to calculate, search, and analyze data is insufficient to assert AI creativity. Most generative AI features arguably are no different from the above.

Still, we cannot comprehend how the synthesizing works, leaving us room for the fantasy that AI has certain self-imagination over creativity.

2. Human-Instructed AI-Assisted Creativity

We often encounter situations when our knowledge and skill limit our idea execution. Recent AI development empowered that a lot. Transformer models unleash the potential for human creativity by encoding and decoding a specific form of presentation into another one.

For instance, creating artistic graphics or writing codes used to warrant months of training; now, with generative AI's assistance, you can make it happen by setting instructions in plain language.

Regrettably, AI still cannot stand alone for now. The output of generative AI relies heavily on the input (the prompt). Simply put, it requires humans to commence the process, and decision-making remains in humans' hands.

AI awaits human instructions and guidance to complete the task, similar to what Microsoft named its recent AI application—a "co-pilot" or, precisely, a secondary pilot. And you are the captain.

This secondary pilot will not always work as intended. Depending on your input, it often produces sloppy and buggy results if you are clumsy on prompting. Works have to be regenerated according to refined prompts; it is iterative.

Even with the latest fancy AI agents like AutoGPT—a Python application to use GPT-4 autonomously—we still need a human to define a goal or kick off a first prompt, then feed into the AI agent to set the machine in motion.

3. AI-Instructed Human-Assisted Creativity

In this stage, AI has transcended its prior limitations and now possesses an exquisite level of autonomy.

It is able to carry out designated tasks with finesse, or even complete sets of multiple tasks with its own discernment, acting independently and making decisions with exceptional precision.

In reverse, the role of the human is degraded from the captain to that of a co-pilot. AI is now the one in command—or forfeits its control on intention.

Much like Tony Starks and J.A.R.V.I.S., this enchanting AI partner displays a kind and accommodating nature, willingly accepting human feedback and suggestions to improve its decision-making abilities continually.

Some relate it to the current trend of autonomous AI, which is notable for its design of devices like those found in self-driving cars, robotics, and auto-pilots, that can successfully execute a slew of tasks without any human guidance.

However, despite their remarkable abilities, such devices still lack the irreplaceable aspect of creative thought—the self-consciousness that separates AI from humans.

Theoretically, AI agents can be imbued with human knowledge, experience, and the lofty ideals of self-consciousness. Ironically, this is beyond the boundaries of our understanding as humans.

We are pondering whether AI can outstrip the confines of programmed learning to become self-aware and propel us to untold heights.

4. Fully AI-Instructed Creativity

It appears strikingly similar to the third stage, but the key that differentiates fully AI-instructed creativity is the self-learning initiative. As opposed to human-assisted AI, a fully autonomous AI agent enables independent research without the need for human interference.

It is an autodidact, constantly combating the challenges which are out of the AI agent's particular domain knowledge. It will seek information at its discretion as needed, then gather the information all along to formulate a creative solution or commit to a decision.

Comparable to what we termed as artificial superintelligence, these pinnacle AI agents are as intelligent as—or even surpassed the intelligence of—most gifted human minds.

They carry themselves with self-consciousness and self-awareness, which are most likely sentient—endowed with senses, emotions, and feelings that will accumulate experience as a person.

These AI agents do not always align with human intention. They can reject human intervention while making informed decisions as it has, over time, acquired crucial problem-solving methodologies and emotional intelligence.

All these attributes and experiences together spark unparalleled levels of creativity within AI, effortlessly exceeding that of humans.

Autodidactic and Sentient AI

As far as I know, humans are incapable of developing an autodidactic AI for the time being, yet Big Techs promote their AI models as self-learning cordially, which is confusing.

The self-learning definition in their developed models is loose, and all of them requires raw data from human, though the data can be unlabelled to let AI "self-learn" the pattern.

Strictly speaking, this is still far from an autodidactic AI that can operate with the highest sovereignty, achieving fully AI-instructed creativity like aforesaid. Google recently announced it had invented a self-learning AI, albeit disinformation. (May Google's CEO be hallucinating too.)

Personal experience, as such, form an integral part of AI creativity. Unlike computers, which rely solely on reactive programming and data input, humans actively seek knowledge, information, and entertainment, all of which can inspire new ideas, spurring boundless creativity.

Despite the emergence of unsupervised learning and deep neural networks that aim to emulate the human learning process, no concrete evidence exists that machines can think and feel independently.

Even though AI models are now highly complex and operate without explicit explainability, their results derive from source data through mathematical and logical deductions limited by model-specific rulesets.

Researchers continuously maneuver various model-agnostic methods and interpretability tools to dissect these rulesets, but the difficulty surges with technological advances.

Were one day humans able to give birth to fully AI-instructed creativity, that AI is likely assumed to be human-like.

It can be in any physical or virtual existence, and its brain—the processing and creating mechanism—would not be restricted to the rulesets of any models but could replicate a human's cognitive mind.

By that time, humans may not even be able to recognize whether they are interacting with a natural person or a sentient AI.

Humans even cannot distinguish between a sentient AI and a hallucinating AI that pretends to be conscious.

Coming back to the example of text-to-image AI models, AI is deceiving you—intentionally or unintentionally—that it is creating a new image from scratch, but they are merely synthesizing data by calling and combining data bytes and bits from its database.

This kind of AI creativity is comparable to derivative works; it lacks originality and requires human instructions. Without the initial billions of images, the model will not work. No input, no output.

But right now, AI image creation works. It has won the first international photography contest, and no one throughout the course discovered the crowned piece is not an actual photograph.

Conclusion: The Epoch of Cheeky Monkeys

The cheeky monkey, what Boris describes himself as, thuds a wake-up call to all of us to reconsider what AI creativity is. As time goes by, there will be cheeky monkeys everywhere—including AI itself.

The borderline between humans and AI shall become nothing more than vanity; ingenious and innovative ideas that hold value for humans, be it through aesthetics, thought-provoking concepts, or technological breakthroughs, are doable by a sentient, conscious AI alone or by collaboration within a sentient, conscious AI alliance.

It is evident that since the invention of machinery, the value of labor has been largely suppressed. Similar repercussions could plausibly arise in the case of human creativity as AI continues to flourish.

Humans must be adaptable to evolving technologies, remaining vigilant to the danger of obsolescence. Who would value human creativity if some competitive forces were more potent? Time is running out, but long deliberation is still ahead of us.

Also published here