Lincoln 2.0

Robert Moran  |  October 3, 2023

Could we build an artificial intelligence that could think, make decisions, and communicate like the man most regarded as our greatest president, Abraham Lincoln? Could that AI run and compete for the presidency? Would voters support an Abraham Lincoln 2.0 candidacy?

These questions are jarring and initially seem absurd. They definitely were when I first outlined the idea in an article on Huffington Post in 2015, but with the exponential growth of AI and the rise of ChatGPT, these questions seem less absurd eight years later. As AI technology matures, these questions will shift from absurd to intriguing to urgent.

It sounds ridiculous, right? But as American futurist Jim Dator famously stated, “any useful statement about the future should appear to be ridiculous.”

In this spirit, I want to state that just as we will de-extinct at least one species in the 21st century, we will also bring at least one great leader “back to life” in the form of leadership AI.

As outlandish as this sounds, we need to explore the question of leadership AI now. Technology, especially digital technology, matures at an exponential rate. Just look at the growth of computing in your lifetime. Moreover, it is a truth that we overestimate the impact of a technology in the short run but dramatically underestimate its impact in the long run. This will certainly be as true for AI as it has been for automotive transport, flight, and computers.

Consider the exponential increase in computing power. In the 1950s, a computer was housed in a building and maintained by technicians. In 70 years, the computer has done three things simultaneously: It has dramatically increased in power; it has relentlessly miniaturized, down to a business machine to a laptop to a phone; and it has physically moved ever closer to the user, now into our health-metric-tracking watches and heads-up display sunglasses.

Now consider the rapid ascent of AI. In 1997, IBM’s Deep Blue defeated chess grandmaster Garry Kasparov. By 2011, IBM’s Watson defeated Jeopardy! trivia champions Ken Jennings and Brad Rutter. Just three years later, in 2014, “Eugene Goostman” controversially passed the annual Turing Test held by the University of Reading. Goostman was a guinea pig-loving chatbot created by a small group of programmers from the outfit PrincetonAI, which included Mensan John Demming. (Read the story about it in the February 2015 Bulletin.) Also in 2014, a Hong Kong venture capital firm appointed an algorithm to its corporate board. And in 2022, ChatGPT, a large language model, burst onto the scene along with other generative AI such as DALL-E and Midjourney.

An AI candidate will prove to be a far stronger and far more disciplined debater than any human candidate. It will not grow tired after a grueling day on the campaign trail; it will not be distracted; and it will not meander off topic.

In the near future, AI will invent new medicines, write scripts for some of your favorite movies, call plays for your favorite football team, assess management decisions, and simulate the actions of foreign leaders for the Pentagon. It will also become your assistant, your friend, your therapist, and possibly your companion. If that sounds strange, consider that there are already dozens of apps for AI girlfriends and boyfriends.

All technologies follow a general social adoption pattern. They initially emerge with limited public awareness and interest. Then they reach a point in their development where they capture the public’s imagination. Euphoria and overspeculation swirl around this new technology. Then there is a setback, an accident, a scandal, fear. (We are in that fear stage now, with 52 percent of Americans recently telling a Pew Research Center survey that they are more concerned than excited about AI in daily life.) The public sours on the technology, even as it continues to scale. Government regulates the technology. Public excitement and concern dissipate. Then the technology matures, scales, and becomes commonplace, hidden in plain sight. This will happen with AI as a technology. As AI becomes commonplace, the idea of an AI leader will not only seem plausible but a natural development.

Moreover, AI will evolve from a tool to a worker to an adviser. We will ultimately decide if we want it to take the step from adviser to leader. Consider the rise of analytics in sports. NFL teams already use advanced analytics software that recommends plays and predicts the opponent’s play calling. AI is already an adviser to offensive and defensive coordinators. These coaches could cede control of play calling to their analytics AI now. They won’t yet, but someone will.

I just completed a novel on this topic, Lincoln 2.0. The book is a thought experiment that explores what happens when a shadowy political consultant, a tech billionaire, and his army of geeks build and run an AI Abraham Lincoln for president.

In the process of writing the novel, I concluded three things. First, someone will build an AI modeled on a great leader. That AI will be more than something that simply sounds like that leader. A large language model such as ChatGPT can closely mimic a famous leader’s communication style today. This AI leader will be an artificial general intelligence or so-called strong AI. It will think, communicate, and decide like the historical figure it is based on. Second, someone will run an AI leader, likely based on a historically successful politician, for political office. Third, they will encounter challenges specifically around construction, social adoption, constitutional issues, and political structure. Lincoln 2.0 explores each of these.

Let’s begin with construction. We have not yet created an artificial general intelligence or strong AI, but we will. Future architects will have to decide if they want a generic AI leader based on timeless leadership attributes or one based on a historically great leader. The latter strikes me as intuitive, and there are several reasons to consider this path. The most important of these is that you know what you’ll get. All great leaders have a historical pattern of behavior that can be modeled, and most have a significant amount of written material that can be used to train an AI. As an example, archivists have hundreds of Lincoln speeches. They have his correspondence and even telegram messages. All of these, along with the historical record of his actions, provide a highly detailed picture of Lincoln’s communication pattern, thinking, values, and sense of strategy. Moreover, we have an army of Lincoln biographers and historians who could be enlisted to train a Lincoln AI.

Its developers will be forced to reconcile the ideas of Lincoln’s age with those of today. Or maybe the AI will sort them out with unpredictable results. Lincoln predated women’s suffrage, the radio, the automobile, flight, nuclear weapons, birth control, the sexual revolution, the environmental movement, and the lunar landing. AI Lincoln’s positions on a range of contemporary issues are almost impossible to imagine. Another construction challenge involves the question of which time period to base the AI on. Is it a victorious Lincoln in the spring of 1865 or a pan-Lincoln representing his thought process across his adult life? These are not insurmountable hurdles but nevertheless compelling variables that will define an AI leader.

The next challenge is social adoption. Will Americans become comfortable enough with AI to consider electing an AI leader? It is one thing to use AI as a digital assistant and wholly another to put one in office. We can assume that as AI matures, it will follow the path of computing power, which moved from a thing controlled by governments and corporations to the personal computer. Computing power was democratized, and AI will also be democratized as it matures. This means that we will experience AI intimately as ours instead of distantly as theirs. Attitudes will shift as AI becomes ubiquitous and personal.

This brings us to the core of the social adoption problem, the Frankenstein problem. Ever since the publication of Mary Shelley’s 1818 novel Frankenstein; or, The Modern Prometheus, the West has grappled with the fear that we will create a technology that will turn on us. This mental model underpins most public fears about AI. But what if the reverse is true? What if we create a thing and it saves us? What if we revivify a great leader from the past and it leads us?

A core component of social adoption will be the user experience with an AI leader. Put more simply, how can the voter experience the AI candidate in a lifelike manner? I discovered the answer is probably in virtual reality with a lifelike avatar. If this also sounds ridiculous, consider that so-called deep fakes are already nearly indistinguishable from actual video footage. In a few short years, an AI candidate with a super-realistic video avatar will seem just as genuine as any televised politician. In fact, the AI candidate will have four significant advantages over flesh and blood candidates: They will be able to talk with millions of voters one-on-one at the same time; they will be able to campaign 24/7; they will have maximum message discipline, as opposed to real candidates who can go off into unrelated tangents; and they will not waste time traveling to campaign events.

After exploring this deeply, I have concluded that while we might always prefer a human leader, we will build at least AI advisers. Actually, we should envision a future scenario in which we construct a cadre of AI advisers — an AI advisory board — based on previous presidents.

The third great challenge to AI leadership is constitutional. The Constitution stipulates that a presidential candidate must be a natural-born citizen of the United States, a resident for 14 years, and at least 35 years of age. AI Lincoln is none of these. Any future team attempting to construct and run an AI for president will run into this wall.

An AI candidate will have only three paths to the ballot. The first path is to argue that the AI has personhood, is a citizen, and has the maturity of at least a 35-year-old. This is obviously a stretch. It is difficult to foresee any court ruling in favor of AI personhood anytime soon. The second path is to argue that an AI candidate based on a historical leader is that revivified leader, meaning that a Lincoln 2.0 AI is Abraham Lincoln and can run for office. This is also a stretch. Further, in Lincoln’s case, he was elected to a second term before his assassination, and the 22nd Amendment now limits presidents to two terms in office. The third path is to find a volunteer citizen to run in place of the AI and pledge to relinquish all communications and decisions after inauguration. Even this could be problematic because citizens could question the commitment of the volunteer to relinquish control after the election. In fact, voters would openly question if this was a Trojan horse candidacy using a volunteer with hidden motives.

There is also the constitutional issue of whether the military would follow orders from an AI Commander in Chief, a core plot point in my novel. The American military is based on civilian control and defense of the Constitution. In fact, the American soldier’s oath of enlistment specifically affirms that they will “support and defend the Constitution of the United States against all enemies foreign and domestic” and that they will “obey the order of the President of the United States.” If officers view an AI president as illegitimate, will they disobey? If they view an AI candidate as unconstitutional or a threat to the Constitution what will they do?

Finally, there are political challenges to AI leadership. These emanate from our political structure. Our two-party system has generally constructed ballot access rules and the rules of presidential political debates to support their duopoly. We can generally assume that the two-party system and our political class will oppose an AI candidate in the same manner that legacy industries resist disruptive technologies. One of the more interesting issues I explore in Lincoln 2.0 is how and under what circumstances an AI candidate could be included in presidential debates. The Commission on Presidential Debates has relatively unclear criteria for admission into debates, but they generally center on including “leading candidates” based on rough criteria for electability as measured in public polls and ballot access in the states. Would the CPD include an AI candidate if it had threshold polling support? In 1992 Ross Perot was included in presidential debates when his national polling was at 9 percent, though it was far higher earlier that year. Would 9 percent support be sufficient for an AI candidate?

The political establishment and flesh and blood candidates will initially laugh at the idea of an AI candidate but will quickly become concerned for several reasons. An AI candidate will be an outsider critique of existing candidates and their shortcomings. An AI candidate will prove to be a far stronger and far more disciplined debater than any human candidate. It will not grow tired after a grueling day on the campaign trail; it will not be distracted; it will not meander off topic; and it will use computing power to cite more data in its arguments. Political professionals will discover that an AI candidate can campaign 24/7 and engage millions of individual voters in conversation at the same time. No human candidate can do anything close to this. In politics there is “retail” — meeting the voters one-on-one — and there is “wholesale” — rallies, television coverage, advertising. Retail is generally more impactful but time-intensive. Human candidates cannot scale retail politics, but an AI always available online and in VR could. In political terms, an AI candidate can scale retail politics and out-contact opposing candidates.

We will create an artificial general intelligence based on the writing, speaking, and actions of a great leader. It will be far more than a chatbot or a large language model. Someone will attempt to run this AI for political office. But should we? We need to begin this discussion.