(Photo by iStock/gorodenkoff)
At the dawn of the age of artificial intelligence, many human beings are haunted by the specter of AI, rendering human intelligence obsolete. This fear, while understandable, is fortunately misplaced. Far from making human intelligence and creativity redundant, AI won’t be able to advance without human-generated data.
To be useful at all, AI must offer solutions that work in the real world, and that means using real-world data to train AI. AI can process data in quantities and at speeds far beyond an individual human being or team of human beings. But the data must come from somewhere—and to train AI models effectively, it must exhibit “ground truth,” meaning it accurately represents reality.
Human beings are the source of this data. And human beings judge the value of AI’s work product. This human review takes place first in data centers as researchers seek to improve the functionality of AI. It takes place next in the assessment of AI use cases as a subject of investment in further research and development. Finally, human review takes place in the marketplace: Consumers evaluate the usability of generative AI output, and businesses assess the ability of AI solutions to improve efficiency, cut costs, and increase customer satisfaction. Throughout this process—improving, applying, and using AI—the human element is essential.
That’s why, in public policy circles, we must nurture and reward innovation in as well as from AI. As has been the case with technological progress generally since the founding of the United States, that means protecting all aspects of creative human work now manifesting itself through AI. As former heads of the US Patent and Trademark Office (USPTO) under presidents of both parties, we see vast potential for economic and social development stemming from intellectual property (IP) protection for all human creativity connected with AI. We need policies that support a robust system of IP rights in order to develop successful AI models. But we must also avoid restricting the use of AI to an extent that inhibits creativity and innovative output, which means that agencies like the USPTO should refrain from overly regulating this technology until we can fully understand it.
Human Creativity and AI
At its core, AI is an adaptive learning system capable of generating insights from vast amounts of human data. AI is not merely mimicking human intelligence but augmenting and extending it in unprecedented ways. When a language model crafts nuanced text or an image generator produces awe-inspiring visuals, these systems are doing more than just recombining existing content—they extend human creativity to new works.
The biggest promise of AI lies in its ability to help humans uncover hidden connections and possibilities that might elude even the most brilliant minds on their own. In health care, AI won’t just assist in taking notes and suggesting diagnoses. We’re already seeing the beginnings of hyper-personalized medicine that employs AI models to analyze individual patients’ genomic data and develop personalized treatment plans for deadly diseases. In climate science, AI models are processing and interpreting massive datasets, such as the data tracked by the World Environment Situation Room on climate indicators like atmospheric CO2 concentration and sea level rise. Even in the arts, AI is not simply imitating existing styles but collaborating with human artists to enable new forms of creative expression, as seen in Refik Anadol’s 2023 AI art display at the Museum of Modern Art.
These are paradigm shifts in how we approach some of humanity’s most pressing challenges and loftiest aspirations. Yet, crucially, these feats are only possible because AI has been created by humans for particular human purposes and trained on diverse, high-quality datasets, including medical journals, market reports, and musical compositions, all of them products of human creativity and expertise. It’s this symbiosis between human creativity and artificial intelligence that can propel us into new frontiers of innovation and discovery.
But that symbiosis—and AI’s effectiveness—isn’t a guarantee. When the bedrock of diverse, abundant, and high-quality data is exhausted, AI falters. Limitations in training data or processes can lead to AI systems that stumble in new scenarios or produce unreliable—even dangerous—outputs. In essence, an AI model is only as good as the data it’s fed. A poor “data diet” produces poor results.
This data diet problem manifests in various ways. One such issue is “overfitting,” where an AI model becomes too attuned to its training data, losing the ability to generalize and adapt to new situations or prompts. Imagine an AI trained to recognize cats but only using images of white cats. When presented with a black cat, it might fail to identify the cat correctly. This occurs when the training data lacks diversity or when the model is overly complex relative to the available data. Fortunately, the solution is simple: More varied, human-created data that captures the full spectrum of possibilities and ongoing human scrutiny and assessment of AI results for accuracy.
Another challenge emerges with the increasing use of “synthetic data”—artificially generated information used to train AI models when real-world data is scarce or restricted. While synthetic data can be useful, overreliance on it can lead to a second problem: a lack of nuance and real-world complexity in AI outputs. For instance, an AI trained primarily on synthetic medical data might ignore rare, but critically important, symptoms a human doctor would catch. Here, too, the remedy lies in incentivizing the creation of more diverse, real-world data through human expertise, lived experience, and creativity.
Perhaps most concerning is the specter of “model collapse.” Model collapse occurs when AI systems are repeatedly trained on data generated by other AI systems, creating a closed loop that gradually degrades the quality and diversity of outputs. It’s akin to a game of telephone, where the message becomes increasingly garbled with each retelling, or like making copies of copies on a photocopier, where each new copy is a degraded version of the last. To avoid this outcome, we need a constant influx of fresh human-generated content to keep AI models grounded in the richness and unpredictability of the real world.
These challenges all underscore a critical point: The “intelligence” in artificial intelligence is, in essence, a sophisticated human tool that depends on human knowledge and creativity. The question then becomes: How do we encourage this essential human input? The answer lies in strong and effective intellectual property protection.
Protecting Creators
Strong IP protections serve as a powerful incentive for human creativity and innovation. When creators know their work will be protected, they will invest time and resources into developing new ideas, technologies, and content. This isn’t just about protecting the next blockbuster movie or breakthrough drug. It’s about fostering the diverse ecosystem of human-generated data that’s needed to make AI systems as successful as possible, including scientific and academic research, creative works, images, and more.
Consider the realm of scientific research. Strong patent protections encourage scientists and institutions to undertake risky new research and bring resulting products to market. Of course, filing a patent necessarily entails disclosing the precise nature of one’s discovery. Over time, this process of research, protection, and disclosure contributes to the broader pool of human knowledge. Published research can become invaluable training data for AI systems working on everything from drug discovery to climate modeling.
Similarly, in the creative arts, copyright protections incentivize the production and distribution of diverse content—books, music, visual art—that form the backbone of training data for generative AI models. Without these protections, we risk a chilling effect on creative output, ultimately limiting the raw material available for AI to learn from and synthesize.
Strong and effective intellectual property protection isn’t just a matter of encouraging creators to generate the new data required to make AI as useful as possible. It’s also about securing IP for improvements to the function of AI. For example, it’s true that detecting and analyzing patterns in data is something the human mind can do and that AI can do. And it’s true that, in general, activities undertaken in the human mind—detecting, measuring, analyzing, among many other things—are not eligible for patenting as such. But while such processing as undertaken in the human mind isn’t patent-eligible, in many cases where accomplished by AI, it should be.
Currently, however, USPTO guidance concerning human-created enhancements in the ability of machines to detect and analyze using AI is too restrictive. The Office only permits patenting when the use of AI is supplemented with “enough” human contribution—an arbitrary standard not applied to any other tool used by inventors. Current guidance also equates the outputs of AI machines with “abstract” thought processes generated by humans, which are normally not patent-eligible. These rules, based on clumsy generalizations, will only slow further research into AI and should be rescinded. Our laws should treat AI like any other tool: defaulting to patent eligibility until and unless compelling evidence becomes available to dictate some other policy direction.
Of course, strengthening IP protections for both data inputs and human-created enhancements won’t just increase the social and economic utility of AI. A more robust and reliable system of IP rights will also directly benefit content creators. Stronger copyright protections for creative works—whether in visual art, photography, songwriting, graphic design, or some other field—will allow more innovators to make a full or partial living from their artistic endeavors. The opportunity to earn a financial return will attract more investment into established artistic genres. Even more importantly, making engagement with AI “worth it” for more creators could facilitate the creation of entirely new genres we haven’t yet conceived of.
To be clear, we must be diligent in protecting the artistic community from the potential harms of AI. The ability of generative AI models to accurately mimic the style and content of human creators’ visual and written work has justifiably renewed artists’ fears of copyright theft. However, AI is not inherently a threat to art—it represents a significant opportunity to expand the field. It’s crucial for government agencies like the USPTO and the Copyright Office to craft thorough and thoughtful policies that safeguard creators and their works from infringing uses of AI without stifling artistic progress or preventing artists from harnessing the beneficial power of this technology.
We must also make sure AI isn’t used in ways that discourage people from creating their own work. Tolerating AI counterfeits is counterproductive since it will ultimately lead to a decrease in human-generated data, increasing the risk of problems like overfitting and model collapse. On a more fundamental level, no artist should have to compete with an AI clone of their own voice or vision. To address these challenges, Congress is considering the NO FAKES Act, which aims to protect individuals from unauthorized AI manipulations like “deepfakes” while safeguarding legitimate uses under the First Amendment.
The path forward is both clear and exhilarating: To unlock AI’s transformative potential across fields, we must incentivize the human creativity that fuels it. Strong IP protections are key to this process. By safeguarding human ingenuity, we can accelerate AI capacity in areas from energy to personalized medicine, education, climate change solutions, and artistic expression.
As we navigate the AI revolution, we should view human and artificial intelligence as complementary forces, each enhancing the other’s potential. By fortifying IP protections—through legislation such as the NO FAKES Act and modified USPTO guidance that clarifies the patent eligibility of inventions made with AI—we’re not merely safeguarding human creativity. We’re catalyzing a new era of innovation where AI amplifies our problem-solving capabilities and expands the horizons of human achievement.
Read more stories by Andrei Iancu & David Kappos.