https://doi.org/10.53656/str2024-1-4-gen

Резюме. This article deals with the concept of “ultimate history” and the role of present-day AI language models. On the one hand, there stands an old idea that someday, with the development of science, it will become possible to create one final narrative of history. With the fast burgeoning of the AI this question becomes even more imperative for historians all over the world. What would mean to be a historian in a world with a machine that knows more than yourself about history, and what would be the new role of a historian, if AI manages to crack the “ultimate history” once and for all? Those are some of the questions that this paper tries to answer.

Ключови думи: ultimate history; AI and history narratives; fake history; ethics in history; obligation of historians

At the end of the 19 th century European historians were celebrating the huge development of historical knowledge that has been accumulated in the past one hundred years, allowing them to recreate and rewrite the human past with greater precision and accuracy than ever before. Big narratives were established. It was believed that with the advance of science and with the buildup of literature and knowledge there might come a day to achieve the ultimate truth about the past, to reach the level of “ultimate history” (Ward & others 1902, v). Such a notion sounds daring even today, at the dawn of the first commercially used AI algorithms, that shook the traditional ideas of education and learning as never before1. But historians of the late Victorian society were people of their time, products of an immense belief in the inevitable success of tomorrow.

But then tomorrow only presented itself with war, devastation, and unprecedented suffering. Two world wars made most researchers in the humanities think twice about the idea of inevitable success. In the minds of the scholars of the first half of the 20 th century there was rather a doubt for a bright future. The 1950’s and 60’s were fueled with cynicism, and the following two decades social and political sciences, as well as history, were infumed by some depressive concepts for the future being of humanity. After the fall of the Berlin Wall, however, an old hope reemerged and faith for universal success and endless prosperity once again became the dominant spirit in the humanities. The bright prospects felt in a way inevitable, just a matter of time. History never dealt much with the notion of future, although history teachers and professors really like to bring out to students the old clich, how if one learns the past, will understand the present, and thus foresee the future. Yet, historians almost seem to be the worst at their prognosis for the future. But that may have to do with the fact that we as human beings are way too limited for a prognosis based on time continuity to make any sense. A single historian, or any scholar for that matter, cannot reliably predict what comes next, since the variables are too great to calculate. With the virtually limitless resource of computational power, however, that might be not such a difficult task.

But if there is a way for AI to see the future, it would at least require to see the past as clearly as it views the present, or in other words, for the AI to tell what is upcoming, it would need to crack the “ultimate history”. Learning what the human kind told about the past wouldn’t be enough, because of the subjective element of any research. The AI may in a sense rewrite history, which sounds a fantastically easy job for a machine, that has access to more data than not only any historian that has ever lived, but also more than any team of historical researchers, multiplying that with the unlimited knowledge of different languages. But the current data shows otherwise.

It is interesting to think if AI would make the typical divisions of history, as we are used to now. Or maybe it will follow the pattern that exists and it will name some centuries “Dark”, while framing others as time of Enlightenment. Historians usually divide history into Ancient, Medieval and Modern times, and this concept is maintained by the AI operating language models for chat service. But in other arias, where AI was already in use for almost a decade, it has different approach. In computer games AI tends to develop its own approach to the game and it does that with a lot of success against very competitive human players. Chess is a good example, where the AI platform “AlphaZero” crushes every human player, even the best Grandmasters of the game. Since it approaches the game differently, learning it from the basics by itself, it is stronger than older computer chess engines, using conventional algorithms. But chess and other similar games are called “closed systems” – they are perfectly fitted for a machine (Bolter 1984, p. 2). For better or for worse, real life tends to get more complicated than a board game with 64 squares and 6 types of pieces. Regardless, it is more than likely AI to try its own approach at some time. It is especially challenging for historians today to establish a common ground for the introduction of new borders in what is most widely called “contemporary history”. Constructing periods such as “postmodern” or “postindustrial” had been a problematic issue from the start (Mazlish 1988, p. 390). There is an even greater emphasis about this absence of universal understanding about newer periods in history in the context of AI.

AI original interpretation of periodization of history could revolutionize the way we see former times, or at least the way AI looks at past events. It is also worth asking if AI would even care about what was what in days of old. Or will it decide that the only thing of relevance exists from the time of its own birth, not before the information age, not the bygone times of mere innocent and unaware analog existence? Does it really matter what happened before this superbeing, the AI, was created? Was the history professor just making himself important and selling his studies to the public, overestimating the worth of history, just for the sake of his salary? One thing is certain AI will be important to history and history will matter to it. Secondly, historian’s work will become really important for this process, so things would not go out of hands.

The reason for this is that if AI starts rewriting history, but gets it wrong, there is no one else who can prevent fake knowledge become the main narrative, except for trained professional historians. As AI consultant James Manyika believes “getting AI to understand science and mathematics and to theorize and develop novel concepts remain grand challenges for AI” (Manyika 2022, p. 12). Or in other words, AI cannot at this point create new contributory publications, including in the field of history (or any other science, in that regard). But it has the potential to flood the internet with fake facts about the past, that are very hard to catch without a trained eye. As shown by experiments I conducted with chat language models, however, AI sometimes creates fake content or just gets some things wrong. In a dystopian scenario AI could create the “ultimate history”, that fake version could become the “new official” history, and who will have the say what is truth and what’s not. The internet is full of fake history and wrong facts already. There are abundant interpretations out there that compliment national histories or political agendas, and they already coexist together with some very adequate materials. Surely AI can contribute to multiplying fake or politically inspired versions of history, that don’t serve society, but interested parties.

Up until now when a heated historical debate is taking place, we tend to listen to the best authorities. When British historian David Irving sued Penguin Books and Deborah Lipstadt, experts were called, in that case the renowned historians Richard J. Evans and Christopher Browning. The testimony of Browning and Evans was vital for the defense of Lipstadt and her claim, that Irving was in fact distorting historical evidence to deny the Holocaust in the Third Reich. This example shows the importance of professional historians whose dedicated work serves society in times of doubt or attempts of misrepresenting historical facts for various reasons. Admittedly, this task seems harder today with the fast spread of fake or manipulative articles or materials in the internet, or occasionally in published papers and monographs. Professional historians just can’t keep up with the scale and speed of disinformation. And in the case of David Irving, a professional and a very capable historian as himself was the one responsible for the manipulations. Ethics and professionalism don’t always go hand in hand.

History should be taught with ethical consideration and historians generally agree to a code of ethics, that is a part of the schooling of every young historian. The problem with such concepts is that they vary greatly, and though universally centered around the notion of objectiveness and basing facts on primary sources, cultural, political and social implications may influence the assessment of a historical topic, process, person, event or result. There is no common ground for many historical issues, even for things that happened a generation ago, where a rich variety of sources and documentation is available. It seems even more complicated when tracing the steps of our ancestors in more ancient periods, when sometimes the only sources come from archeology. Interdisciplinary approach made it fashionable, though maybe not so useful as one might imagine, to use biologists and medical doctors to examine DNA, and chemical solutions for dating evaluation in archeology are standard equipment for many years now. As much as history is being enriched by new and innovative research though, it looks even more perplexing to explain the past wie es eigentlich gewesen ist2, if we want to embrace the old, but still valid principle of Leopold von Ranke. More modern way of seeing things is presented by some American historians, who believe that what Ranke meant was simply telling the essence by conscientiously working with the primary sources. This seems a job not so hard for an AI.

For the preparation of this paper multiple experiments with commercially available AI of different companies have been conducted. In the wake of the new “AI Spring” (as opposed to the “AI Winters of the past) 3 of the last two or three years, the belief that a machine can answer all questions has become more widespread than ever. There have been other observations regarding the use of this AI in the field of historical knowledge, but what I have discovered can be summarized as follows: Some well-known historical facts are adequately presented by the AI, other, however, are not. Even very basic knowledge, like, for example, which is the political party of the British Prime Minister Lord Palmerston (1784-1865), the AI answer was wrong – putting him in the conservative party, a blunt mistake, since Palmerston was a liberal during the time of his premiership and for nearly thirty years before that. But he was also a Tory in the beginning of his career, so this might be the reason the AI made a mistake. Asking the AI to perform a little bit more difficult tasks, albeit very regular for any historian – a historical comparison of events, for example, is where the AI seems to lack depth or analytical approach, rather it presents crude facts about the subjects in question, without being able to actually make a real comparison. AI is capable to produce summaries and its conclusions on historical issues are sometimes very accurate, informative and synthesized, tough short and surface-leveled. But the fact the AI has the strange tendency to fabricate unexisting facts and present them as the real answer, makes it an unreliable tool at best, and a dangerous manipulator at worst. This has been observed by other researchers and in other fields of knowledge, not only in the field of history (Minow, Dwork 2022). But one would wonder why a platform for AI-chatbot would distort historical facts, so readily available at Wikipedia or other internet encyclopedias. There might be very good technical explanations for this and surely those problems can be addressed in the future development of AI. However, the reality now is, that AI is basically producing fake facts on occasions, and which is even more frustrating, this happens only to every 1 out of 7 questions I asked, based on my limited experience with it.

In other words, AI-chatbot works fine most of the times, but can mislead you when you just start trusting its otherwise impressive ability to be competent about everything and produce elaborate answers to various topics almost immediately on your screen. Another interesting occurrence during the trial of different chatbots with AI was another tendency – when there is something wrong with the answer (wrong fact, etc.) asking the AI to try once more may result in producing a different response, and even admittance for the previous mistake. The follow up, however, wouldn’t be necessarily the right answer, rather another wrong one, occasionally even absurd reply.

All things considered, AI-chatbots of the most famous commercial types available in the internet, are quite impressive tools for checking historical facts. When the AI gets it right, which is in most cases, and depending on the difficulty of the question, of course, it performs amazingly well, having in mind it was not created for a historical database in the first place. It can be concluded, that future versions and continuing developments will ensure even better performance to complicated historical issues. One thing remains certain, though. At this stage of development AI has no potential in rewriting big narratives, nor does it possess any authority or credibility over historical topics, general or specific. The short answers on very complicated matters are an attempt to summarize huge problems into small generalized statements, which is not enough to generate anything close to an “ultimate history”. And this is a satisfying thought for most historians. But an important concern should be to value and truly appreciate the work of historians. As in other cases in the past, the only way to deal with misrepresenting history, is to rely that historians would do their duty to defend the truth. Weaponizing AI to create fake content is a serious issue, that needs to be addressed. It can be damaging to any science. History is very vulnerable to manipulations anyhow, so with this new tool it can become even more important to underline the role of the professional historian in our society, who will be the only trained to properly deal with the coming challenges, the first and last wall of defense before the approaching wave of fake “ultimate histories”, that may flood the internet, if no measures are taken. It is also true that in a world of high risks in defense, economy and finance, medical care and other much more pressing issues than history, we might overlook the danger of putting the machine in control of our past, since we have so much concern for the present. But history is our memory, collective, but also individual. And like in a computer where the memory is one of the indicators of power, in a society collective and historical memory is something that matters a lot. Historians, now more than ever, have the obligation to be that memory.

Acknowledgments & Funding

This study is financed by European Union - NextGenerationEU, through the National Recovery and Resilience Plan of the Republic of Bulgaria, project № BG-RRP-2.004-0008-C01. The publication itself is a result of the work on a sub-project of Sofia University “St. Kliment Ohridski” on the topic: “Diplomacy in the New Times. Between Ideology and Practice” (contract No. 70-123-485/28.06.2023)”.

NOTES

1. More general about AI see: Simova, A. 2023. History of civilization. The great technological inventions. Sofia: Millenium. ISBN 9789545156182, pp. 286 – 290.

2. “How things actually were”.

3. AI Newsletter, 1 January 2005. https://www.ainewsletter.com/newsletters/ aix_0501/#w last visited 20 October 2023.

REFERENCE

BOLTER, J. D., 1984. Artificial Intelligence. Daedalus, Summer, 1984, vol. 113, no. 3. pp. 1 – 18. The MIT Press on behalf of American Academy of Arts & Sciences.

MANYIKA, J., 2022. Getting AI Right: Introductory Notes on AI & Society. Daedalus, Spring, 2022, vol. 151, no. 2, AI & Society, pp. 5 – 27, The MIT Press on behalf of American Academy of Arts & Sciences.

MAZLISH, B., 1998. Comparing Global History to World History. The Journal of Interdisciplinary History, Winter, 1998, vol. 28, no. 3, pp. 385 – 395. The MIT Press.

MINOW, M.; DWORK, C., 2022. Distrust of Artificial Intelligence: Sources & Responses from Computer Science & Law. Daedalus, Spring, 2022, vol. 151, no. 2, AI & Society, pp. 309 – 321. The MIT Press on behalf of American Academy of Arts & Sciences.

SIMOVA, A. 2023. History of civilization. The great technological inventions. Sofia: Millenium. ISBN 9789545156182. (In Bulgarian).

WARD, A. W.; PROTHERO, G. W. & LEATHES, S. (eds.), 1902. The Cambridge Modern History. V. I. The Renaissance. Cambridge University Press.