Nations exert influence in two key ways: hard power, and soft power. Hard power is dropping the atomic bomb on Hiroshima and Nagasaki during World War II, compelling the Japanese to surrender. Conversely, soft power is building American libraries in foreign universities, placing American films in foreign theaters, and sending United States Peace Corps volunteers to rural communities worldwide, compelling a favorable image of our country abroad.
In other words, hard power is to influence via economic and military violence; soft power is to influence via attraction.
Soft power through media
A key subset of soft power is that delivered through media.
For example, during the Franco-Prussian War, the French dropped leaflets over enemy territory by airborne balloon touting the mutual benefits of ceasefire. During World War I, all parties dropped such leaflets by airplane. In World War II, the Nazis pioneered the use of radio as a propaganda machine. Finally, in 1969, the Americans live-broadcast Neil Armstrong’s moon landing to television screens worldwide, showcasing the "derring-do and genius of American ingenuity, and putting one over the Soviets at the same time."7
Today, such media is more numerous and diverse: American shows on Netflix, American music on Spotify, American “influencers” on YouTube, Hollywood movies and more—all distributed globally.
The cost of media
The cost of media can be split into two broad camps: creation and distribution.
Since the days of leaflets by balloons, the marginal cost of distribution has become extremely cheap. Today, Trump tweets, and his words instantaneously appear on the timelines of his 75 million followers. Were he to have 150 instead, the additional cost of reaching these additional 75 would be, practically speaking, nothing.
In contrast, the creation of this media itself remains a wholly manual exercise. To create a news broadcast, humans write scripts, set the stage, select the costumes, apply makeup, and read the news itself. Furthermore, creating twice as many news broadcasts costs, crucially, roughly twice as much.
The dwindling marginal cost of its distribution makes soft power media efficient and appealing. However, it is the “creation bottleneck” that stands in the way of true scale.
Generative models are those that empower “deepfakes.” They are algorithms that generate realistic images (refresh this link for more), video, audio, text, or other types of rich media. Presently, generative models excel in creating art, music, literature, news reports, and textiles. Microsoft’s Xiaoice, a multi-purpose chatbot (which currently dialogs with over 660 million registered users), pulls all this off and more:
“Xiaoice’s framework is learning to write literature as well as compose and perform songs. Last year she published a book of poems and helps her followers write their own. She can sing her own songs in styles based on existing popular performers. There are plans to release an album of pop tunes soon. And she is able to author tailor-made stories for children and reads them out in voices suited to each of the characters she has created.
She’s painting images based on keywords and other inputs. She’s also gone into mainstream media as a host of dozens of TV and radio programs that are broadcast across China. She reads news stories and provides commentary. And, she is generating multiple reports based on information from China’s financial markets and used by investors and traders who subscribe to Wind, a major financial information service.”
In short, generative models automate the creation of media itself—smashing the “creation bottleneck” outright.
At present, generative models succeed at tasks like: transferring dance moves from professionals to amateurs, translating pencil sketches to high-resolution images with the guidance of text, and making Obama lip sync the words of American actor Jordan Peele.
These feats are impressive. However, to achieve widespread impact, these algorithms must allow for more “control” over the media they create.
For example, consider a system that, given English text, instantaneously generates a video of this text being recited in an arbitrary language. To be truly effective as a soft power tool, this system must ultimately operate as follows:
System: “What kind of video would you like?”
User: “I’d like a video of the President reciting the uploaded text. I'd like him to speak slowly, be wearing green US Army fatigues, and be seated in front of a blue Boeing CH-47 Chinook. Make me two, actually: one in Farsi, the other in Greek.”
System: “Coming right up.”
Such generative models are not quite here. However, given the speed at which this research moves, they will reliably arrive within 5 years.
The United States used to invest in soft power institutions
Historically, the United States invested substantially in soft power institutions.
During World War I, Woodrow Wilson established the Committee on Public Information, which ultimately trafficked in outright propaganda.
During World War II, Roosevelt established the Division of Cultural Relations and Office of Inter-American Affairs in response to fascist propaganda in Latin America. Following the bombing of Pearl Harbor, the Voice of America—a globally-distributed radio broadcast of non-military promotional content—the Office of War Information—distributing media like “newspapers, posters, photographs, films”8 to civilian communities abroad—and the Office of Strategic Services—the predecessor to the CIA—were all established as well.
During the Cold War, the United States flexed its “peacetime” soft power muscle, establishing myriad organizations to promote international cultural and educational exchanges, including the Fulbright Program in 1947 and the United States Information Agency (USIA) in 1953.
Dwindling interest and opportunity missed
Following the Cold War, however, American soft power institutions began to slowly decline.
In 1999, the USIA was absorbed into the US State Department; its staff, and budget for key projects, were cut roughly in half.6 “From 1995 to 2001, academic and cultural exchanges dropped from 45,000 to 29,000 annually, and many accessible cultural centers and libraries were closed.”6 Finally, “while government-funded radio broadcasts reached between 70 and 80 percent of the populace of Eastern Europe during the Cold War, around year 2000, a mere 2 percent of Arabs heard the VOA.”6
At present, the United States has a largely isolationist president in Donald Trump. Since the beginning of his term, American soft power has fallen globally according to the Portland Soft Power 30. Trump himself is unconcerned with others actively laughing at him and his country.
By 2018, Statista estimates that there were 1.67 billion "pay TV households" worldwide.9 By 2019, they estimate that half of private households had a computer.10 Finally, Pew Research estimates that "more than 5 billion people have mobile devices, and over half of these connections are smartphones."11
In Information Age, a dwindling interest in soft power institutions is a major opportunity missed.
In a recent article, Joseph Nye, the Harvard professor who coined the term “soft power” itself, neatly summarizes this phenomenon:
“U.S. President Donald Trump’s administration has shown little interest in public diplomacy. And yet public diplomacy—a government’s efforts to communicate directly with other countries’ publics—is one of the key instruments policymakers use to generate soft power, and the current information revolution makes such instruments more important than ever.”4
A renewed interest in soft power
Truly impactful generative models, and Trump’s exit from office, will soon collide head-on.
At this point, the United States—with its blemished reputation, exorbitant military budget12, longstanding technical leadership13 and forever proactive drive—will be sitting on transformative media tech and a clear-cut mission: rebuild its image at home and abroad.
To this effect, I predict that the United States will soon make a concerted investment in generative models for soft power at true scale.
Here’s what this might look like.
Soft power through media, in the age of generative models
The two keys areas which generative models will impact are state-to-public, and state-to-individual media.
Current examples of state-to-public soft power media, like Presidential addresses to the American people, English-language CNN news broadcasts to foreign viewers, and front-line conflict reporting will see their use both quickened and expanded by generative models.
Presently, Presidential addresses happen once every few weeks or months. Conversely, Donald Trump writes on Twitter hourly. With generative models, the next President will instantaneously generate videos of herself reciting the tweet instead. Americans at large devour the President's tweets; if a generated video legitimately looks and sounds just like her, they’re likely to devour it as well.
Sponsorship deals, like Lebrons’ lifetime appointment with Nike, will crucially expand to include the right to generate content with the athlete’s likeness. Politicians will travel less to campaign; instead, they’ll send generated holograms—of them, their spouses, their key supporters—to campaign instead—each one personalized to the venue in question.
Finally, a generative model will be trained to ingest footage of world events—for example, the attack on the Saudi oil refinery—and produce text describing what happened. This text will then be used to generate a news report. Humans won’t be on the front line to capture this footage either; why not send a drone instead?
Microsoft’s Xiaoice demonstrates the enormous traction of personalized chatbots. From here, the jump to state-built, propaganda-(subtly)-infused chatbots is small.
In general, the public distrusts governments; to this effect, humanitarian missions are typically run by NGOs. As such, states will deploy these chatbots on behalf of related entities: tourism boards, online universities, cultural centers, etc.
Finally, current state-to-individual media is largely unidirectional: media is sent from the former to the latter, and rarely vice versa. Chatbots powered by generative models, however, will solicit feedback from their viewer in a personalized way; for example, “Are you enjoying this media? What do you like about American schools as compared to your own? What do you think about democracy?”
In the same way that the Chinese government passively learns about its citizens by monitoring WeChat data, generative models will actively solicit information re the efficacy of their soft power plays. Then, like a savvy marketer, they’ll double down on the stuff that works best.
For over a century, states have exercised soft power through media. Throughout, the marginal cost of distributing this media has trended towards zero. Conversely, to date, the marginal cost of its creation remains high.
Generative models are a powerful technology which trend the latter towards zero as well. They're here and they work. However, to be truly impactful, there's still a short ways to go.
In 5 years' time, a dazed, reputationally-bruised United States, and robust, flexible, practical generative models, will collide head-on.
At this point, soft power media—and human politics in general—are likely to change forever.
Many thanks to Abishur Prakash for reviewing earlier drafts of this piece.
Brands, Hal. “Not Even Trump Can Obliterate America's Soft Power.” Bloomberg.com, Bloomberg, www.bloomberg.com/opinion/articles/2018-01-18/not-even-trump-can-obliterate-america-s-soft-power. ↩
Chakravarti, Sudeshna. Soft Power: The Culture Weapon in The Cold War and South Asia. www.culturaldiplomacy.org/academy/content/pdf/participant-papers/academy/Sudeshna-Khasnobis-Soft-Power-The-Culture-Weapon-in-The-Cold-War-and-South-Asia.pdf. ↩
Khandelwal, Aakash. “Economics of Digital Goods.” LinkedIn SlideShare, 23 Nov. 2016, www.slideshare.net/aakashkhandelwal921/economics-of-digital-goods. ↩
Nye, Joseph S. “No, President Trump: You've Weakened America's Soft Power.” The New York Times, The New York Times, 25 Feb. 2020, www.nytimes.com/2020/02/25/opinion/trump-soft-power.html. ↩
Shah, Ritula. “Is US Monopoly on the Use of Soft Power at an End?” BBC News, BBC, 19 Nov. 2014, www.bbc.com/news/world-29536648. ↩
Harris, Paul. “Man on the Moon: Moment of Greatness That Defined the American Century.” The Guardian, Guardian News and Media, 25 Aug. 2012, www.theguardian.com/science/2012/aug/25/man-moon-american-century. ↩
“United States Office of War Information.” Wikipedia, Wikimedia Foundation, 7 Jan. 2020, en.wikipedia.org/wiki/United_States_Office_of_War_Information. ↩
Watson, Amy. “TV Households Worldwide.” Statista, 4 Dec. 2019, www.statista.com/statistics/268695/number-of-tv-households-worldwide/. ↩
Statista Research Department. “How Many People Have Access to a Computer 2018.” Statista, 2 Mar. 2020, www.statista.com/statistics/748551/worldwide-households-with-computer/. ↩
Silver, Laura. “Smartphone Ownership Is Growing Rapidly Around the World, but Not Always Equally.” Pew Research Center's Global Attitudes Project, Pew Research Center, 30 Dec. 2019, www.pewresearch.org/global/2019/02/05/smartphone-ownership-is-growing-rapidly-around-the-world-but-not-always-equally/. ↩
Cancian, Mark F. “U.S. Military Forces in FY 2020: The Strategic and Budget Context.” U.S. Military Forces in FY 2020: The Strategic and Budget Context | Center for Strategic and International Studies, 12 Mar. 2020, www.csis.org/analysis/us-military-forces-fy-2020-strategic-and-budget-context?gclid=CjwKCAjwgbLzBRBsEiwAXVIygLkZlNLc3zV8EVVhPPwjOoqrUoQ4kjTeaTHN5vFktRWET2wsnHCwRhoCm2QQAvD_BwE. ↩
“Report Shows United States Leads in Science and Technology as China Rapidly Advances.” ScienceDaily, ScienceDaily, 24 Jan. 2018, www.sciencedaily.com/releases/2018/01/180124113951.htm. ↩