Pages

28 March 2023

How to Describe the Future? Large-Language Models and the Future of Military Decision Making

Bryce Johnston

The innovation of the smokeless rifle may not be as flashy as that of the aircraft and armored vehicles, but describing its impacts on the future of war required more than four-hundred pages of technical analysis in Jean de Bloch’s The Future of War.[1] The difficulty in forecasting the effects of new weapon systems stems from the fact that technology is not additive. A smokeless rifle not only increases visibility on the battlefield, but it also opens new possibilities for tactics and strategy. With smokeless rifles, soldiers could fire upon their enemies without giving away their exact location while large formations could no longer hide under a cloud of smoke as they sent volleys across the battlefield. These changes required armies to reorganize their units to become nimbler and more maneuverable leading to deadlier clashes between opposing forces. States had to contend with deadlier conflicts by increasing their investments in new technologies while taking political stances that would maintain the will of the people in the face of high casualties.

Leaders across the world are seeing the early effects of another transformational technology: widely available large-language models.

The example of the smokeless rifle shows that relatively unexciting technologies can have transformational effects on warfare and society. Today, leaders across the world are seeing the early effects of another transformational technology: widely available large-language models. Viewed as the first step in true artificial general intelligence, large-language models incorporate massive amounts of data from books and articles into training sets that allow them to recognize patterns between words and images. These models appear to be able to answer many questions by generating coherent responses in seconds and can perform menial tasks like summarizing articles or cleaning unstructured data. Across the internet, early adopters have demonstrated novel uses for OpenAI’s chatbot, ChatGPT, that range from editing code to writing essays for college courses.[2] These models are well regarded in the private sector, but they have not received much attention in military circles largely because they do not appear to have direct applications for combat. Large-language models will likely have a larger impact on the battlefield than autonomous drones due to their ability to automate the many aspects of staff work that prevent military leaders from focusing on tactics and strategy.
Large-language models will likely have a larger impact on the battlefield than autonomous drones due to their ability to automate the many aspects of staff work that prevent military leaders from focusing on tactics and strategy.

As with all technologies, understanding the true effects of these models is difficult, but there are some techniques analysts could use to describe how war may change once these tools become widespread. Among science fiction authors, William Gibson stands out for his ability to determine which “boring” technologies today will have huge effects on society. The prolific writer coined the term cyberspace years before the internet became a cultural phenomenon and foresaw how the rise of microcelebrities online would shape the world.[3] Gibson owes his predictive ability to his outlook on the world: “The future is already here — it’s just not very evenly distributed.” Gibson’s statement is not just about technological trends, but their effects on society.

Over half a century before Gibson’s seminal work Neuromancer described the future of the internet, Bloch took a similar approach when evaluating technologies such as the smokeless rifle. He argued that further adoption of industrial technologies would make war so destructive as to be impractical. Likewise, Gibson saw that early hacker culture would come to shape society at large in the near future. Both men recognized that they stood at the precipice of a new world shaped by advancing technologies.

Gibson’s technique should be familiar to most military analysts. Analysts observe a localized conflict and attempt to generalize lessons based on their use of new tactics and weapon systems. Though decades apart, military professionals have viewed the armored warfare of the Spanish Civil War or the drone fights over Nagorno-Karabakh as pockets of the future that could be used to understand the next iteration of warfighting.[4] This technique takes an immense amount of creativity, which is why many modern attempts at conducting these same predictions come in the form of speculative fiction written by military analysts.

I attempted this same methodology with Dr. Oscar Jonsson in a paper last year. Using examples from the private sector, we conducted a technical analysis of how AI might revolutionize Russian intelligence.[5] While we are several years out from knowing if our analysis is correct, some of our predictions have already been undermined by Russia’s invasion of Ukraine and the release of commercially available large language models. Even Gibson admits the limits of his method, stating that the internet turned out to be nothing like he imagined.[6]

These events show that technology does not exist in a vacuum. Instead, it affects and is affected by the society around it. Gibson’s technique can provide insights into the broad strokes of the impact of new technologies on society, but it is not meant as a precise instrument for understanding the future. Bloch attempted to get around this by quantifying society as much as he could given the limited data at the time.

The strengths and weaknesses of this method are best exemplified in his work The Future of War: In Its Technical, Economic, and Political Relations. Published in 1899, Bloch believed that history alone did not provide an accurate way to assess the future. He argued that any judgment on the future of war must account for technological advancements as well as the economic and political effects of the conflict.[7] By applying this technique to lessons from the regional conflicts in the 19th century, he sought to determine the possibility of warfare between the great powers of Europe given such significant advances.

Any precise cost estimation of a future conflict is bound to be incorrect, but the magnitude of the difference between Bloch’s prediction and reality displays the weakness with all forecasts.

Bloch displayed an impressive knowledge of the weapons’ technical capabilities that the great powers employed on land and sea. He applied this knowledge to data from recent battles to predict the cost of human life in the next major conflict between European powers, predicting catastrophic losses for all sides in the conflict due mainly to the introduction of more accurate rifles that did not emit smoke when fired. The introduction of these weapons would disrupt the typical use of large formations and give an advantage to riflemen hidden in the terrain or dug into protected positions. For this reason, he proclaimed that future wars would be long struggles for fortified positions where whole nations would be called to arms. The result, he predicted, would impose an economic and human cost that would make war prohibitively expensive.

The most interesting aspect of Bloch’s work is that his theories were tested fifteen years later during World War I. The grand strokes of his predictions hold up well. The Great War was a defensive slog that required the participants to mobilize the whole of their economies. Interestingly, Bloch’s weakness was that he did not go far enough in his predictions. He underestimated the cost of a great war, projecting that France would need to spend $1.8 billion annually to finance a conflict between Germany and Austria. The actual cost was more than $6 billion a year.

Any precise cost estimation of a future conflict is bound to be incorrect, but the magnitude of the difference between his prediction and reality displays the weakness with all forecasts.

Despite his attempt to update the lessons from regional conflicts with technical and economic data, his model could not account for all the political factors that would lead countries to go to war even when the decision guaranteed mass destruction. His assessment also failed to leave room for the possibility of further evolutions in military technology. Two years after his death, the Wright Brothers successfully launched the first heavier-than-air flight. Airplanes would not be used in conflict until the Italo-Turkish in 1911, which would have made it difficult for Bloch to recognize their importance even if he had lived to see them fly. Even with these weaknesses, Bloch showed that something as simple as a smokeless rifle can have enormous effects on how states finance future conflicts.

Leaders in the defense realm should take a page from Gibson and Bloch by studying early adopters of this technology to see how political and economic considerations may change to support AI-augmented military staff sections.

Bloch’s assessment shows the limits of trying to quantify the future. The purpose of his book, however, was not to accurately predict the future of warfare but to convince others to abandon the idea of war altogether to avoid such harmful new technologies. On that accord, he failed miserably, but his effort still shows that examining history for lessons on future warfare best reveals how the introduction of a technology can affect the entire system of warfare.

Leaders today are in need of this kind of direction. The release of large-language models to the public have brought the world to a new precipice. These technologies have enraptured the public and the media alike. The interfaces released to the general population feel like a form of magic, but the large-language models that drive them are more like the steam engine or electricity. The large-language model is a general purpose technology that will automate many of the rote tasks performed by office employees. While military leaders could be content in knowing that their staffs will run more efficiently once they integrate these models into their workflow, they should look beyond this effect to understand how greater efficiency in staff work may affect the economic and political calculations behind the next war.

Much of today’s strategy is formed by large staffs that function like any other modern office place. The introduction of ubiquitous AI will have far-reaching consequences. Talk of AI in defense circles has been rooted in the dream of fully autonomous robots, but the real transformation will occur in the offices that litter the headquarters of every echelon of the military.

While tools to increase staff productivity are not as flashy as pilotless helicopters, they will likely contribute more to the transformation of warfare by decreasing the amount of time it takes leaders to generate and analyze courses of action.

The general trends of AI integration into staff functions like intelligence is inevitable. In looking at current uses for large-language models, it seems that these models will help automate many of the bureaucratic tasks that bog down staff sections during the planning process and provide prompts for common battlefield problems. The fear that these models will completely pull humans out of the loop are unfounded; rather, they will create decision-making space by lessening the load of other tasks. These models will automatically create a common operating picture that will keep different staff sections synched even when they are not in face-to-face meetings. The timeline for this integration may be shorter than most analysts may think. Microsoft, who owns a large share of OpenAI, has already started integrating it into Microsoft Teams, the main program that staff sections in the United States use to synchronize their files.[8]

While the immediate effects are clear, the second and third order effects will take time to forecast. Leaders in the defense realm should take a page from Gibson and Bloch by studying early adopters of this technology to see how political and economic considerations may change to support AI-augmented military staff sections. While tools to increase staff productivity are not as flashy as pilotless helicopters, they will likely contribute more to the transformation of warfare by decreasing the amount of time it takes leaders to generate and analyze courses of action. Like Bloch’s smokeless rifle, it seems likely that the adoption of large-language models by military staff will make war more deadly in ways that may be hard to predict. Yet despite this difficulty, it is necessary to take the time to try to describe the second-order effects of a technology that the world does not quite understand yet.

No comments:

Post a Comment