Jimmy Song began his presentation at the University of Tennessee at Chattanooga with a simple request: Think of a number between one and 10.
After giving the audience a few seconds to make their selections, he asked those who had chosen one, two, nine or 10 to raise their hands. About a dozen responded. Then he asked how many had chosen seven. Nearly the same number of hands went up.
Song explained that the results reflected a well-known human tendency. When asked to select a number between one and 10, people disproportionately choose seven while avoiding numbers near either end of the range.
Then he asked the audience to repeat the experiment, this time using their smartphones. Participants opened their preferred artificial intelligence chatbots, whether ChatGPT, Claude, Gemini or another model, and asked them to choose a number between one and 10.
When Song asked who had received a number other than seven, only a few hands went up. Nearly everyone else’s chatbot had chosen seven.
The demonstration illustrated a phenomenon Song said poses a greater danger to humanity than the apocalyptic scenarios that dominate much of the public conversation about AI: conformity.
During his Nov. 10 presentation, titled “The Economic Impact of Artificial Intelligence,” Song argued that the greatest threat posed by AI isn’t that machines will become smarter than humans, seize control of society or eliminate the need for human labor. Rather, he said the growing reliance on AI could gradually narrow the range of ideas people encounter, reinforcing conventional thinking at the expense of originality.
“The real danger is that AI creates a false consensus,” Song said. “We gravitate toward the middle, not because the best ideas rise to the top, but because the same ideas are repeated over and over.”
The consequences, he warned, could extend far beyond the technology itself, influencing everything from creative expression to economic competition and the way people think.
The danger of thinking alike
Song, an author, programmer and Bitcoin educator whose books include “Thank God for Bitcoin,” was invited to UTC by the Probasco Chair of Free Enterprise. Claudia Kramer, who holds the chair, introduced the presentation as an opportunity to explore how technological change is reshaping the economy and to encourage discussion of challenging ideas.
Song began by acknowledging the fears commonly associated with AI, including the possibility of artificial general intelligence surpassing human capabilities and turning against its creators.
One familiar scenario resembles SkyNet, the fictional computer system in the “Terminator” movies that decides humanity is a threat and attempts to destroy it. Another imagines an AI instructed to manufacture paper clips pursuing that objective so relentlessly that it consumes the world’s resources or enslaves humanity to accomplish its task.
Song considers these scenarios distractions from a more immediate concern.
To explain why, he described how large language models, or LLMs, process information. These systems are trained on enormous collections of text and other material, identifying statistical relationships that allow them to generate responses to questions and instructions.
Song compared the process to data compression. A model cannot preserve every detail of the material used to train it, so some information is inevitably lost as it learns to represent patterns in a more compact form.
That process, he argued, favors common responses while potentially diminishing the influence of unusual or less frequently represented ideas.
The number-seven experiment offered a simple illustration. If humans already favor seven when asked to choose a number, an AI trained on examples of human responses may reproduce that preference with even greater consistency.
Song worries that as people rely more heavily on AI to write, communicate and make decisions, the technology’s preference for familiar patterns could begin to influence human thinking as well.
“We’re being trained by the AIs as much as the AIs are being trained by us,” he said.
A world of diminishing originality
Song argued that the drift toward conformity did not begin with artificial intelligence. In his view, AI is accelerating a process already visible throughout the economy and popular culture.
He displayed images of automobiles from different manufacturers, observing how similar many contemporary vehicles have become compared with the distinctive designs of earlier decades.
Corporate branding provided another example. Song showed how technology companies and fashion houses have abandoned distinctive logos in favor of simpler designs, often using similar sans-serif lettering.
The industries are different, but Song said the underlying incentives are much the same.
As companies grow, they must appeal to broader audiences. Unconventional products and ideas become harder to justify because they carry greater financial risks. Businesses gravitate toward choices that satisfy the largest number of customers, even if those choices sacrifice originality.
“Centralization turns variety into conformity,” Song said.
In his view, AI could accelerate that tendency because a relatively small number of widely used models have a growing influence on how people write, communicate and solve problems.
He pointed to the recognizable language of AI-generated emails and other content, arguing that the technology already produces a distinctive sameness. As people adopt those patterns in their own writing, the distinction between human and machine expression may become less pronounced.
The larger concern is that people could begin to mistake the most frequently repeated answer for the best one.
Song described this as a form of false consensus, in which ideas acquire apparent authority through repetition rather than careful examination.
Who controls the answers?
For Song, the danger becomes more consequential when a handful of companies control the AI systems on which millions of people rely.
Song argued that large technology companies have an incentive to discourage competition, including by supporting AI safety regulations that smaller developers might struggle to satisfy.
Although such regulations are intended to address legitimate concerns about the technology, Song suggested they could also protect established companies from emerging competitors, further concentrating control of AI in the hands of a few powerful corporations.
That concentration, he warned, could have consequences beyond the marketplace. If a relatively small number of AI models become the principal sources of information for millions of users, what happens to ideas that fall outside those systems’ prevailing patterns?
Song says unconventional perspectives could receive less attention, not necessarily because they have been disproven, but because they are less prominent in the information the models reproduce.
The solution, he argued, involves encouraging a more decentralized AI ecosystem, with smaller, specialized models drawing on different sources of information and serving more narrowly defined needs.
Rather than relying exclusively on a few systems designed to answer nearly every conceivable question, developers could create models tailored to particular professions, communities or specialized fields.
Such systems, Song suggested, might preserve information and perspectives that larger models overlook.
The human advantage
Despite his warnings, Song did not end his presentation with a prediction of technological decline. Instead, he offered an argument for why humans could become more important as AI grows more capable.
“Human creativity is what makes AI useful, and it’s what AI will continue to need,” Song said.
Song said the technology’s limitations create an opportunity. Although AI can perform an expanding range of tasks, he argued that humans remain essential to identifying problems, imagining possibilities and directing the technology toward meaningful goals.
He returned to that distinction during a question-and-answer session when an audience member challenged his concerns about conformity by pointing to AI’s contributions to mathematical research and the development of new medicines.
Song acknowledged the usefulness of those applications but emphasized the role of the people directing them.
In mathematics, he said, researchers can use AI to explore possible approaches to a difficult proof, investigate existing literature and perform work that might otherwise require considerable time.
The technology can be an extraordinarily powerful assistant, but Song said the creative initiative often begins with the mathematician who recognizes a promising direction.
“You still need someone to provide direction,” Song said. “Someone who has an idea and a vision for how to make it work.”
The same principle, he suggested, applies to scientific discovery and other fields. AI can help researchers investigate possibilities, but people must still decide which questions are worth asking.
“If you treat AI as a tool, then it can be extremely useful,” he said.
That distinction forms the basis of Song’s optimism. Rather than viewing AI solely as a competitor that will replace human abilities, he sees the possibility of a technology that magnifies the impact of original thinking.
Creativity as the next frontier
If creativity is what distinguishes human contributions from those of AI, Song believes it could also become one of the most valuable skills in an economy shaped by the technology.
Near the end of the Q&A, an audience member asked Song what he would do if he were a college student preparing to enter today’s economy.
His response brought the presentation back to its central theme.
“Creativity is the next frontier of human value, especially as AI takes over much of the routine work,” Song said.
As AI becomes more capable of performing routine tasks, he argued, the ability to define meaningful work could become more valuable than the ability to carry it out.
Song envisioned a future in which entrepreneurs use AI to develop products and services that previously would have required large teams of specialists.
Someone with an idea for a household robot, for example, might use AI to identify components, develop software, research suppliers and plan production.
Tasks that once required extensive technical knowledge or considerable financial resources could become accessible to individuals willing to pursue an idea.
The opportunity, however, comes with a responsibility that Song said traditional education does not always encourage: taking initiative.
Students spend years learning to follow instructions, satisfy requirements and work within established systems, he observed. An AI-driven economy may reward people who can move beyond those expectations and identify opportunities for themselves.
“AI requires more creativity, not less,” he said. “And the more creative you are, the more leverage you’ll have.”
Song said this could produce a proliferation of small businesses and independent creators capable of solving specific problems. With AI handling much of the technical work, individuals could pursue ideas without assembling the large organizations that once made such ventures possible.
The result, he suggested, could be a more entrepreneurial economy in which the willingness to imagine something new becomes a valuable asset.
That possibility also reveals the tension at the heart of Song’s argument: The technology he says could encourage conformity might simultaneously provide individuals with unprecedented opportunities to challenge it.
Song’s opening experiment suggested how readily machines can converge on the same answer. His closing remarks challenged the audience to resist doing the same.