Generative AI and Creativity: The First Year After ChatGPT and a Look to the Future

Welcome to the new issue of Turtle's AI newsletter. Please follow me, Duke Rem ?? for daily news, insights and guides about AI and technology.

It has been just about a year since ChatGPT took the world by storm and brought generative AI into the mainstream. In that short time, these technologies have already begun transforming many facets of society, from how we work and create to how we socialize and spend our leisure time. However, the rise of generative AI has also prompted vigorous debates about its implications, especially regarding human creativity. As we reflect on the first year of this new era and look ahead to the next, a central question emerges: how will the continued development of generative AI like ChatGPT impact human creativity?

In this issue of Turtle's AI newsletter, I will examine where we stand today, one year into the ChatGPT revolution. Drawing on scientific studies, philosophical perspectives, and thoughts from artists, I will analyze how generative AI has thus far influenced creative expression. I will then explore where we may be headed in the next 365 days as these technologies advance. Ultimately, I argue that while generative AI poses certain risks, it also presents immense opportunities to augment and enhance human creativity. Adopting an epistemologically balanced perspective, we can harness these tools to expand the horizons of imagination while cultivating the uniquely human aspects of creative expression.

The First Year: Generative AI Finds Its Footing

When ChatGPT launched in November 2022, it demonstrated a level of human-like conversation ability unmatched by previous AI systems. Based on a cutting-edge natural language processing model called GPT-3.5, ChatGPT dazzled users with its articulate, nuanced responses on virtually any topic. It was a watershed moment for generative AI, a category of AI focused on creating novel content.

In the 12 months since, ChatGPT has been joined by a slew of new generative models (Claude, Bard, LLaMa, Palm just to name a few of these). Adoption has skyrocketed; enthusiasm has been accompanied by apprehension, especially around how these technologies could impact human creativity and jobs, something that we have already faced in previous issues of this same newsletter.

The feverish growth of generative AI reflects a technology catching fire with users, but the first year has also seen these systems expose their limitations. While capable of producing original content, current systems often lack contextual awareness (at least to a certain degree), can occasionally respond incoherently or with misinformation (the so called "hallucinations"), and perpetuate societal biases in their training data. Most experts agree the technology remains years from matching human capabilities. Progress has been swifter than expected, but generative AI remains primitive compared to the nuance and complexity of the human mind.

This first year has seen vigorous debate about the implications of these technologies for human creativity. Nicholas Carr, author of The Shallows, has argued generative AI could atrophy our imaginative capacities in the same way calculators weakened math skills. Others counter it simply represents a creativity-enhancing tool. The philosopher Jean Baudrillard would possibly characterize generative AI as simulacra incapable of true originality since their outputs derive from human-generated training data (Baudrillard, 1994). But the artist Refik Anadol, for example, sees the technology as a new medium with which to expand the boundaries of human and machine creativity. These debates illuminate vital questions about the symbiotic relationship emerging between humans and creative AI.

The Impact on Human Creativity Thus Far

Let us turn to the key question: one year in, how has the rise of generative AI impacted human creativity so far? Several recent studies provide scientific data to assess its effects. A Nature study found that - on a divergent thinking task - the best AI systems produced more creative responses on average than the typical human participant. However, top human responses exceeded even advanced AI. The authors conclude generative AI can now match average human creative performance but still falls short of the most creative individuals (Koivisto & Grassini, 2023).

Another study, published in Harvard Business Review, evaluated generative AI’s impact when people were given access to AI-generated ideas on a story writing task. They found writers produced more creative stories after viewing AI ideas, with benefits largest for less creative individuals. However, the AI-assisted stories were more similar to each other, suggesting possible homogenization if AI writing assistants were to become widespread. This highlights an important debate - will increased use of AI lead to more creative outputs overall, even if the outputs themselves are more similar?

In the last year, rapid advances have been made in generative AI systems across modalities including text, images, audio, video, and 3D models. Systems like DALL-E, Stable Diffusion and others demonstrate an astonishing ability to create realistic, coherent visual outputs based on natural language prompts.

These advances build on the core strengths of deep learning - the ability to recognize patterns and reproduce content by training neural networks on vast datasets. Unlike rule-based expert systems, deep learning models do not rely on explicitly programmed knowledge. Instead, parameters are adjusted through exposure to optimize performance on a task. This data-driven approach is key to the rapid improvements in generative AI.

But does this mean AI systems are truly creative? Human creativity is complex, drawing on cognition, emotion, culture, and life experiences. As psychologists Robert Sternberg and Todd Lubart emphasized, creativity requires a confluence of factors including knowledge, intellectual abilities, thinking styles, motivation, and environment. Machines may lack this richness.

Philosopher David Novitz provocatively argued that “computers neither understand, nor create, nor feel.” From this view, AI has no inner mental life from which novel ideas spontaneously emerge. Its outputs merely reconstitute elements derived from training data. Although results may appear creative, AI is just optimizing statistical relationships between inputs and outputs.

However, AI researcher Margaret Boden contends that creativity does not require human-like consciousness. She proposes a useful distinction between psychological creativity (novel ideas that are surprising given the creator's mindset) and historical creativity (ideas that are unprecedented in human history). An idea can be computationally creative if it is novel with respect to the system's previous outputs, even if it is not historically creative.

By this standard, we may consider AI generative if it consistently produces outputs distinct from its training data. But critics emphasize that AI lacks intention and agency in framing problems and conceptualizing solutions. The generative power arises from engineers providing datasets, formulating prompts, and iterating systems - not autonomous machine creativity.

I personally think that generative systems can clearly assist human creativity in many ways. multiple studies show that tools like DALL-E and ChatGPT help creators overcome fixation, stimulate new associations, and improve ideation.

For instance, the video that was recently appreciated here on LinkedIn was made possible, for me, by GenAI’s amazing support:

https://www.dhirubhai.net/posts/duke-rem-%F0%9F%90%A2-793616267_genai-animation-turtlesai-activity-7115780402157338624-CV2D

However, the best human creations still equal or surpass AI. In divergent thinking tests, humans imagine more varied uses for objects and generate story ideas exceeding the originality of AI. But unsophisticated human responses lag behind AI. With the sheer scale of data it assimilates, AI covers immense ground.

A comprehensive view suggests that current AI has narrow, bounded creativity. Its originality depends wholly on training by humans. But properly directed, AI can enhance creative discovery. The appropriate role may be as a power tool abetting (and augmenting) human imagination, not an autonomous creative entity.

This balance is tenuous, though. Several concerns arise about over-reliance on AI creativity tools:

• Homogenization of ideas: As the study of AI-assisted stories showed, outputs tend to resemble each other more. Overuse of tools like DALL-E could lead to stylistic convergence rather than exploding diversity.
• Unrealized potential: If creators lean too heavily on AI, they may not develop their own latent abilities. AI that appears to excel could stunt human creativity.
• Loss of meaning: AI might excel at novelty but lack deeper meaning. Text and art may be technically proficient but hollow, missing the symbolism and cultural resonance of human creations.
• Excessive expectations: When AI overpromises on creativity, people get frustrated with imperfections and limitations. Unmet expectations could undermine confidence in collaborating with AI. alienation: Over-automating creative acts could estrange people from the process. Creativity could become impersonal rather than an expression of human experiences.

Concerns vs advantages

These concerns are partly offset by AI's advantages in expanding access to tools for creativity. Those with limited time, resources, training, or natural aptitude can benefit from AI in developing their ideas. But experts warn about an over-reliance on machine creativity. The appropriate role for AI is to complement human creativity, not substitute for it.

This balanced perspective is supported by a sociological view on the interplay between technology and creativity. Historically, new tools that augment human capacities but do not replace them have driven expansions of creativity. For instance, the invention of photography did not end painting. Rather, it freed painters to explore more stylistic directions without the constraint of realistic representation.

AI can similarly liberate creators from repetitive, analytic tasks and provide stimuli to trigger new insights. But the essential act of creativity remains profoundly human. To go back to the example of the video that I mentioned before, the basic idea, the script and, in the end, the finishing operations I did them myself: technology shapes the forms creativity takes but should not supplant the role of human background experiences, judgment, and choices.

AI may eventually cross the threshold to become independently creative. But currently, it lacks creative intention, does not autonomously frame problems, and cannot critically evaluate its own ideas. Full-fledged creativity requires a capacity for self-reflection and grasping meanings that AI does not yet possess.

Looking ahead, responsible development of AI creativity tools should:

• Maintain human agency and control
• Make AI collaborators, not competitors
• Require transparency about AI contributions
• Avoid exacerbating biases in systems
• Prioritize human flourishing over productivity

The upcoming future?

The trajectory of the next year will further illuminate if AI can move beyond bounded creativity. Key developments to monitor include:

• More sophisticated prompting: Better techniques for framing prompts could make AI more versatile and reactive. Two-way dialogue may enhance interactivity.
• Self-learning and imagination: Next-generation AI could show glimmers of self-directed learning, seeking out new knowledge and spontaneously imagining.
• Cross-domain creativity: Current AI excels in domains with vast training data like images and text. Creativity across music, visual art, poetry, and more could emerge.
• Evaluation and critical judgment: AI may get better at reviewing its own creations, selecting its best ideas, and recognizing flaws.
• Transparency and ethics: Researchers may improve techniques to detect AI content and ensure ethical transparency. Public debates will likely intensify.
• Specialized creative AI: AI tailored for specific creative fields could demonstrate genius exceeding general models. Expert communities will scrutinize results.

I would like to close this essay stating that, whatever GenAI can do, humans possess inexhaustible creative potential. AI can uplift that potential if carefully directed. But we must remain vigilant that technology augments rather than automates creativity. The path forward requires nurturing human talents while thoughtfully integrating AI's gifts where they complement our own. Responsible development of AI creativity tools can unlock new horizons of imagination while keeping human creativity at the forefront.

Duke Rem ?? and Turtle's AI team.