Decoding Alexander the Great’s gastrointestinal cause of death using artificial wisdom: An artificial intelligence-human inquiry into a medical mystery

Abstract

BACKGROUND

ChatGPT was developed in November 2022 with studies showing its impressive performance in academic examinations, serving as a promising tool to answer questions even on controversial topics. Artificial intelligence (AI) achieving surface-level performance does not necessarily equate to a deep understanding of human cognition. The development of artificial wisdom, therefore, necessitates a shift from simply mimicking intelligent behavior to modeling the underlying mechanisms of human wisdom, including emotional understanding, ethical considerations, and contextual awareness. Several theories exist behind the death of Alexander the Great, but no definitive conclusion has been made.

AIM

To evaluate whether a hybrid approach, combining generative AI (ChatGPT) with human clinical judgment, can meaningfully reassess the cause of death of Alexander the Great.

METHODS

This is a cross-sectional study using ChatGPT (version 4 Pro). A search was performed with search terms describing the symptoms experienced by Alexander the Great and possible causes of his death: West Nile virus (WNV) encephalitis, poisoning, acute pancreatitis due to excessive alcohol consumption, typhoid fever, and malaria. The historical data and symptomatology were analyzed, weighing evidence and context in a manner akin to human wisdom.

RESULTS

The most likely cause of death of Alexander the Great, as generated by ChatGPT, was typhoid fever complicated by Guillain-Barré syndrome (GBS). The hypothesis was based on the alignment between Alexander’s reported symptoms, such as prolonged high fever, severe abdominal pain, neurological decline, and the known clinical presentation of typhoid fever. However, after carefully reviewing the sources mentioned by ChatGPT, many did not back up the idea that typhoid caused GBS and instead pointed to Campylobacter jejuni as the more likely trigger. Other possible causes of death suggested by ChatGPT including acute pancreatitis from excessive alcohol consumption, infectious causes (WNV encephalitis, malaria), and poisoning were less likely.

CONCLUSION

While ChatGPT initially concluded typhoid fever with GBS as the most plausible cause of death, expert reappraisal of the sources and pathophysiology suggested that C. jejuni-associated GBS was more likely. This study exemplifies how incorporating AI’s pattern recognition with human scrutiny can yield responsible interpretations of historical records.

Key Words: Artificial intelligence; Artificial wisdom; Gastrointestinal; History; Alexander; Typhoid; Campylobacter jejuni

Core Tip: While ChatGPT is an impressive tool, its conclusions still require human validation. Although ChatGPT identified typhoid fever with Guillain-Barré syndrome (GBS) as the most likely cause of Alexander the Great’s death, cross-checking both the available literature and the sources cited by ChatGPT revealed that Campylobacter jejuni, a more common and well-established cause of GBS, better fits the clinical and historical presentation. This study shows how artificial wisdom, which combines artificial intelligence with human judgment, can support more contextually accurate interpretations in historical medical investigations.

INTRODUCTION

Several fields, including automotives, medicine, education, and finance, have applied artificial intelligence (AI) sporadically over the years[1]. In recent years, AI has become increasingly accessible to individual users, particularly following the release of ChatGPT in 2022, which has since been widely used for problem-solving, organizing daily tasks, and content generation. In the medical context, ChatGPT has shown promise in areas such as streamlining workflows, providing accessible health education, generating diagnostic algorithms, and even improving healthcare documentation[2]. While AI’s potential has garnered much attention, critics have raised concerns about the dangers of relying blindly on generative AI for everyday decision-making and information retrieval, including the risk of accepting inaccurate outputs due to biases and stereotypes embedded in AI training data. Moreover, AI remains limited in its ability to feel empathy or compassion, engage in self-reflection, take perspective, or understand and appropriately respond to emotionally or morally complex situations factors, which may lead to inappropriate or context-insensitive conclusions[3-5]. In a field like medicine, where decisions often involve uncertainty and require context-sensitive interpretation, current AI systems may not be able to fully replicate human adaptability and contextual judgment. Additionally, ChatGPT’s ability to seem reasonable on the surface often misses deeper critical thinking about sources; therefore, it is essential to combine AI results with expert knowledge in the field. This combination leads to what is called “artificial wisdom” (AW), a mixed thinking approach that is becoming more important in looking back at diagnoses and investigating past events[6].

In 2020, Jeste et al[6] introduced the AW concept, a form of institutional wisdom. In this model, wisdom does not arise from a single AI system or human agent alone, but from a collaborative system that integrates AI’s ability to rapidly synthesize vast amounts of information with the human capacity for ethical reasoning, emotional intelligence, and contextual understanding.

Rather than offering answers, AW offers perspectives, which hold particular value in processing ambiguous or biased data, weighing the credibility of sources, and drawing context-sensitive, ethically restrained conclusions. This makes AW especially applicable to areas such as forensic medicine and retrospective historical investigations, where unresolved or unnatural deaths often involve complex and conflicting evidence. In such cases, conclusions must be drawn with care and humility. Figure 1 illustrates a conceptual flow diagram explaining how AI and human insight interplay to derive AW.

Figure 1 Conceptual flow of artificial wisdom. AI: Artificial intelligence.

A relevant case where AW may be applied is the death of Alexander the Great, the ancient Macedonian ruler who became king at the age of 20 and famously never lost a battle. He died unexpectedly in 323 B.C. at the age of 32, marking the fall of his empire and the beginning of the Hellenistic period, during which Greek influence persisted[7]. To this day, the exact cause of his death remains medically unconfirmed and has sparked ongoing debate, fueled by conflicting historical accounts, the passage of time, and the absence of concrete evidence.

Historians such as Plutarch, Arrian, Adrienne Mayor, Richard Stoneman, and toxicologist Leo Schep have proposed various theories to explain his demise, ranging from poisoning and infectious diseases to congenital conditions[8]. Among modern hypotheses, Marr and Calisher[7] have suggested West Nile virus (WNV) encephalitis, based on epidemiological clues and symptomatology. However, this too remains speculative.

This study applies the framework of AW to a medical-historical investigation: The gastrointestinal cause of death of Alexander the Great. By combining ChatGPT’s outputs with expert clinical reasoning, we critically reassessed competing etiologies like typhoid, Campylobacter jejuni, malaria, pancreatitis, and poisoning in light of both symptomatology and source reliability. While ChatGPT excels at surface-level synthesis and pattern recognition, it lacks higher-order cognitive capabilities such as ethical reflection, contextual awareness, emotional nuance, and epistemic humility[3,4]. Relying solely on ChatGPT for such research questions can therefore be problematic, as their persuasive tone can create a false sense of credibility when they could be confidently wrong, leading to the uncritical acceptance of speculative claims. For instance, ChatGPT has been documented to have cited non-existent sources, misquoted references or even made biased overgeneralizations in attempts to summarize and simplify complex concepts to end users[3,5].

As noted by Jeste et al[6], a wise system should “learn from experience, integrate multiple perspectives, and learn from its mistakes.” In line with this principle, ChatGPT was used to generate and rank possible gastrointestinal diagnoses based on symptom patterns. These outputs were then subjected to human clinical discernment to scrutinize conclusions, identify overgeneralizations, and refine them in light of available historical and medical evidence. This process illustrates how the integration of AI and expert insight (termed AW) can enable a more responsible and context-sensitive approach to historical medical analysis.

MATERIALS AND METHODS

This was a qualitative, cross-sectional simulation study applying ChatGPT (version 4 Pro, accessed June 3, 2025) to assess differential diagnoses based on Alexander the Great’s reported symptoms. Prompts were crafted to simulate clinical reasoning by integrating historical data with medical context. ChatGPT’s outputs were evaluated through a structured framework simulating the principles of AW.

Rather than simply retrieving or listing causes, the prompts were designed to encourage contextual reasoning and the weighing of evidence, in alignment with the principles of AW. Factors such as the plausibility of each hypothesis, the historical and medical context, and the need to balance conflicting information were incorporated into the prompt structure. References and justifications supporting each proposed diagnosis were embedded in the inputs to guide ChatGPT’s analysis. The full search prompts, in verbatim, are appended in Supplementary Table 1.

The input provided to ChatGPT was derived from multiple academic and historical references, simulating a reasoning process similar to human clinical and historical analysis integrating evidence, evaluating probabilities, and acknowledging uncertainty, which are hallmarks of AW-based decision-making.

ChatGPT was then prompted to cite sources to support its claims. All sources mentioned by ChatGPT were reviewed by two doctors who rated them on a scale from 0 to 3 scale (0 = irrelevant, 1 = partial, 2 = relevant, 3 = strongly supporting) across four domains: (1) Clinical plausibility; (2) Source reliability; (3) Cultural/historical contextual fit; and (4) Direct relevance to determining the cause of Alexander’s death. Finally, the sources were evaluated qualitatively, and conclusions were drawn based on a combination of input by ChatGPT and human discretion on the part of the two reviewers. Inter-reviewer agreement in each domain was calculated using Cohen’s weighted κ statistic with linear weights, and major discrepancies were resolved by consensus. Source misapplication, hallucinated citations, and speculative logic were coded as “critical weaknesses” and tabulated separately. Misapplied sources were still qualitatively reviewed by both clinicians to assess whether they might support alternative pathologies, whereas hallucinated citations were automatically assigned a score of ‘0’ in each evaluation domain.

RESULTS

Using the search terms described in the methodology, ChatGPT’s response was obtained. The complete response is in Supplementary material (ChatGPT’s in-verbatim output using Supplementary Table 1 prompt [3/6/25]), and following the rules of AW, the AI’s thinking process included looking at different pieces of evidence, taking into account historical and medical background, and weighing the chances in its evaluation.

Based on our search entry into ChatGPT, ChatGPT determined the most likely cause of death of Alexander the Great to be typhoid fever with Guillain-Barré syndrome (GBS). Several factors, reflecting a nuanced interpretation of symptoms and relevant medical knowledge, supported this conclusion. Specifically, the symptoms exhibited by Alexander such as high fever, severe abdominal pain, chills, restlessness and delirium closely aligned with the clinical presentation of typhoid fever, possibly also involving enteric perforation as a complication in view of the abdominal pain and fever. Furthermore, the mention of paralysis and inability to speak was linked to GBS, a rare complication associated with typhoid, illustrating an integrative analysis that considers less common but plausible neurological sequelae. Reports of Alexander’s body allegedly not decaying after death may also be attributable to GBS-induced paralysis, leading observers to mistakenly believe he was dead initially. Typhoid fever also fits the historical context as it was prevalent during Alexander’s time and would have been lethal without the advent of antibiotics.

WNV encephalitis was described to be a less likely cause of death due to “historical and epidemiological inconsistencies regarding the presence of WNV in that era,” as the symptoms he exhibited did not match the typical earlier onset of neurological symptoms of WNV. This demonstrates AW’s capacity to incorporate historical context and evaluate the likelihood of hypotheses based on scientific plausibility. Table 1 summarizes the reasons provided by ChatGPT on why other possible causes of death were less likely. The detailed explanation against other etiologies as possible causes of this death is shown in verbatim under ChatGPT’s generated output in Supplementary material (ChatGPT’s in-verbatim output using Supplementary Table 1 prompt [3/6/25]).

Table 1 ChatGPT’s reasoning for rejection of alternative diagnoses in Alexander the Great’s death.

Cause of death	Explanation
Poisoning theories	The symptoms of the poisons speculated do not match as closely with the full range of symptoms Alexander exhibited, especially the neurological ones. Most poisons would have caused a more rapid deterioration and death, unlike the 12 days of illness Alexander endured
Acute pancreatitis from alcohol consumption	This cause is plausible given Alexander’s history of heavy drinking. However, the progression of symptoms and his ultimate neurological signs do not align as well with pancreatitis, which typically would not lead to paralysis or a ‘locked-in’ state
Malaria	While malaria is a strong contender due to the endemic nature of the disease in the region and his symptoms, the lack of specific mention of periodic fever cycles and symptom onset typical of malaria slightly weakens this argument compared to the detailed matching of typhoid fever symptoms
WNV encephalitis	This is less likely due to the historical and epidemiological inconsistencies regarding the presence of WNV in that era. Furthermore, Alexander’s symptoms do not entirely match the typical clinical presentations of WNV

WNV: West Nile virus.

The sources that ChatGPT used to generate its responses fell broadly into three categories: Historical accounts from books, governmental or organizational health guidelines, and peer-reviewed journal articles. However, while the list was extensive, not all sources were appropriately applied to support the claims made by ChatGPT. In some cases, the cited references did not adequately address the specific claims or failed to consider the cultural and historical context. In addition, 13% of ChatGPT’s 24 cited sources were inaccessible, and 8% were misapplied or quoted out of context. Table 2 demonstrates the degree of citation misapplication or irrelevance that reduced the epistemic value of ChatGPT’s output. These limitations undermined the validity of ChatGPT’s conclusions and required additional human input. To quantify these limitations, a structured review of the sources cited by ChatGPT was performed by human reviewers based on reliability, clinical relevance, contextual fit, and clinical plausibility, with the full results detailed in Supplementary Table 2. Several references, particularly for the typhoid + GBS hypothesis, were inaccessible, misapplied, or lacked historical-contextual alignment, as demonstrated.

Table 2 Evaluation of sources cited by ChatGPT: Accuracy, contextual fit, and misapplication[7,10,18,19,38,50,59-73].

No.	Source	Ref.	Source type	ChatGPT’s summary of source	Average score1	Remarks
1	Did Alexander the Great die from Guillain-Barré syndrome? In: Howe T, Anson E, Balmaceda C, Fronda M, Hollander D, McAuley A, Muller S, Vanderspoel J, Wheatley P, Dunn C, editors. Ancient History Bulletin, 2018: 106-128	Hall[18]	Book	Alexander the Great and the Guillain-Barré syndrome hypothesis	11	This source actually argues that Campylobacter jejuni is the cause of the GBS rather than typhoid while ChatGPT uses it to argue for typhoid. It is a source that was taken out of context to support the argument that typhoid caused GBS, when the paper argued the opposite
2	Typhoid fever. Lancet 2015; 385: 1136-1145	Wain et al[19]	Journal	Clinical progression and symptoms of typhoid fever	8.5	No mention of typhoid causing GBS. It is also wrongly cited as the paper was in the journal lancet not nature
3	Infectious causes of Guillain-Barré syndrome. Curr Opin Neurol 2016; 29: 419-425	Willke A, Ergonul O (fictitious reference)	Journal	Supported the possibility of Guillain-Barré syndrome as a complication of infections like typhoid	7.5	Unable to locate source article. Citation likely to be erroneous
4	Alexander the Great: Murder in Babylon. London: Virgin Digital, 2012	Phillips[59]	Book	Introduced strychnine poisoning theory involving Roxanne	4.5
5	Historical principles and perspectives in toxicology. Toxicol Pathol 1999; 27: 129-138	Wax PM (fictitious reference)	Journal	General toxicological background relevant to ancient poisoning theories	0	Unable to locate source article. Citation likely to be erroneous
6	Calicheamicin gamma 1I: An antitumor antibiotic that cleaves double-stranded DNA site specifically. Science 1988; 240: 1198-1201	Zein et al[60]	Journal	Describes the chemical properties of calicheamicin, related to the Styx river poisoning theory	5
7	Was the death of Alexander the Great due to poisoning? Was it Veratrum album? Clin Toxicol (Phila) 2014; 52: 72-77	Schep et al[50]	Journal	Evaluates the white hellebore poisoning theory	9
8	Alexander of Macedon, 356–323 B.C.: A Historical Biography. University of California Press, 2013	Green[38]	Book	Provided background and narrative of Alexander’s final days	4.5
9	Acute pancreatitis. N Engl J Med 2016; 375: 1972-1981	Forsmark et al[61]	Journal	Supported acute pancreatitis hypothesis from excessive alcohol use	9.5	Authors’ names wrongly cited
10	Malaria. [cited October 17, 2025]. Available from: https://www.who.int/news-room/fact-sheets/detail/malaria	World Health Organization[62]	Website	Described malaria symptoms and prevalence	8	Unlikely, but still a possibility as malaria is always an infectious consideration
11	Pathogenesis, clinical features, and neurological outcome of cerebral malaria. Lancet Neurol 2005; 4: 827–840	Idro et al[63]	Journal	Provided evidence supporting cerebral malaria as a cause of neurological symptoms	3.5
12	Conquest and Empire: The Reign of Alexander the Great. Cambridge University Press, 1993	Bosworth[64]	Book	Historical reference for Alexander’s campaigns and death context	6
13	Alexander the Great and West Nile virus encephalitis. Emerg Infect Dis 2003; 9: 1599-1603	Marr and Calisher[7]	Journal	Proposed West Nile virus encephalitis theory	5
14	West Nile virus: A primer for the clinician. Ann Intern Med 2002; 137: 173-179	Petersen and Marfin[65]	Journal	Clinical background on WNV symptoms	5.5
15	Lives, Volume VII: Demosthenes and Cicero. Alexander and Caesar. Loeb Classical Library 99. Cambridge, MA: Harvard University Press, 1919	Plutarch[10]	Book (translated pri source)	Primary historical source detailing events of Alexander’s death	8
16	The Campaigns of Alexander. London: Penguin Classics, 1976	Arrian[66]	Book (translated pri source)	Primary source documenting Alexander’s military and health history	7.5
17	Ancient Medicine. 2nd ed. London: Routledge; 2012	Nutton[67]	Book	Context for medical practices and understandings in antiquity	6
18	Greek Fire, Poison Arrows & Scorpion Bombs: Biological and Chemical Warfare in the Ancient World. New York: Overlook Press; 2003	Mayor[68]	Book	Insight into ancient biochemical warfare and poison use	4.5
19	Alexander the Great: The Hunt for a New Past. London: Overlook Press, 2004	Cartledge[69]	Book	Reinterprets Alexander’s life and legacy, including his final illness	6.5
20	Alexander the Great: Man and God. 1st ed. London: Longman, 2004	Worthington[70]	Book	Provides historical analysis including character traits possibly relevant to health behavior	3.5
21	Fiction as History: Nero to Julian. 1st ed. Berkeley: University of California Press, 1995	Bowersock[71]	Book	Used to contextualize propaganda and myth in ancient historical narratives	5
22	Who’s Who in the Age of Alexander the Great: Prosopography of Alexander’s Empire. 1st ed. Oxford: Wiley-Blackwell, 2006	Heckel[72]	Book	Provides biographical insights into Antipater and others possibly involved in poisoning	3.5
23	Guillain-Barré Syndrome. Nat Rev Neurol 2020; 16: 10-20	Engel N (fictitious reference)	Journal	Contemporary clinical reference for GBS symptoms and progression	0	Unable to locate source article. Citation likely to be erroneous
24	The Death of Alexander the Great: Rumour and Propaganda. Classical Quarterly 1971; 21: 112-136	Bosworth[73]	Book	Explores the use of rumor and misinformation surrounding Alexander’s death	5.5	Author wrong

¹Average score: Average total score given by two reviewers for clinical plausibility (0-3), source reliability (0-3), cultural context fit (0-3), relevance (0-3). Maximum score of 12, with full details in Supplementary Table 2.

GBS: Guillain-Barré syndrome; WNV: West Nile virus.

Two independent reviewers assessed the sources to verify their relevance, and accuracy, to provide context-based interpretation in determining the plausibility of the AI-generated claims, and verified the sources to be more supportive of C. jejuni infection as a cause of death after assessment of the sources and cross-checking with relevant literature. Between the two raters, the Cohen’s weighted kappa statistic demonstrated good strength of agreement in assessing the sources, with values of κ = 0.77 for clinical plausibility, κ = 0.76 for source reliability, κ = 0.76 for cultural and historical contextual fit, and κ = 0.61 for relevance to determining the cause of Alexander’s death, respectively. The comparative analysis revealed key discrepancies in diagnostic weighting and citation fidelity. ChatGPT exhibited “anchoring bias,” favoring typhoid despite conflicting citations. By contrast, expert reviewers applied an epistemically grounded approach, weighing the neurological trajectory (ascending paralysis) against the clinical epidemiology (Campylobacter vs typhoid incidence). This led to the re-ranking of C. jejuni as the most plausible etiological agent. A full list of these sources is provided in Supplementary material, along with reviewer ratings for clinical plausibility, source reliability, cultural-context fit, relevance to the claim, and accompanying comments in Supplementary Table 2. Table 3 shows the differential diagnosis matrix using ChatGPT and human expert assessment, and the comparative plausibility of each proposed diagnosis as ranked by ChatGPT vs human reviewers. This reflects the corrective function of AW in reappraising overconfident or misinformed AI-generated hypotheses.

Table 3 Differential diagnosis matrix of ChatGPT and human expert assessment.

Diagnosis	ChatGPT rank	Human assessment	Symptom alignment	Source quality	Final verdict
Typhoid + GBS	Most likely	Less likely	Moderate	Weak (some misapplied)	Rejected
Campylobacter jejuni + GBS	Mentioned, not top-ranked	Most likely	Strong	High	Supported
Acute pancreatitis	Moderate	Unlikely	Partial	Good	Rejected
Malaria	Moderate	Plausible	Partial	Good	Inconclusive
West Nile virus encephalitis	Low	Rejected	Weak	Poor	Rejected
Poisoning (various)	Low	Rejected	Weak	Variable	Rejected

GBS: Guillain-Barré syndrome.

DISCUSSION

The past often serves as our most reliable source of information, particularly when the present is evolving and the future remains uncertain[9]. However, historical records are inherently constrained by personal opinions, subjective viewpoints, limited documentation, and genuine uncertainties about the context in which events occurred. In such settings, augmenting human reasoning with AW embodied in tools like ChatGPT offers a novel approach that synthesizes evidence, evaluates plausibility, and balances conflicting narratives. However, AW does not merely validate AI outputs. It explicitly scrutinizes source provenance, assesses the plausibility of mechanistic links and incorporates medical reasoning that large language models (LLMs) cannot autonomously perform. This methodology is particularly suited for reinterpreting historical deaths where primary data are limited. However, AI-generated information must first be critically validated by subject-matter experts before it can be accepted as historical or clinical truth. This integration seeks to emulate a human-like reasoning process, where context, ethical considerations, and uncertainty are acknowledged and carefully weighed.

Born on 20 September 356 B.C., Alexander the Great lived a life of extraordinary accomplishment until his untimely death at age 32. His decline began on 29 May 323 B.C. and progressed over 12 days until his demise in Babylon on 10 June. During this time, he experienced abdominal pain, escalating fever, chills, fatigue, extreme thirst, progressive weakness, paralysis, speech difficulties, and delirium[10]. Possible causes of death can be broadly categorized into: (1) Infectious (e.g., WNV encephalitis, typhoid, malaria); (2) Inflammatory (e.g., acute pancreatitis [AP]); and (3) Toxic (e.g., poisoning).

To explore these possibilities, ChatGPT (GPT-4 Pro) was applied to simulate an AW-guided inquiry. While ChatGPT’s vast training dataset enables fluent synthesis, its reasoning lacks transparency, often presenting speculative links as confident truths. This opacity is exacerbated by its persuasive tone, which risks uncritical acceptance[11,12]. Its performance is especially noteworthy in its ability to generate structured, human-like responses based on a massive parameter base (approximately 100 billion in GPT-3)[13]. Despite its utility, concerns remain about its use of fabricated references[14]. Our study exposes this gap: ChatGPT promoted typhoid-GBS despite citing a source that explicitly argues for C. jejuni instead, a clear example of misrepresentation. Nonetheless, a study by Ayers et al[15] showed ChatGPT’s responses were preferred for quality and empathy in nearly 79% of 585 evaluations, reflecting its potential to mimic aspects of AW, particularly when human oversight is applied.

When prompted, ChatGPT determined WNV encephalitis to be a less likely cause of Alexander’s death. Marr and Calisher[7] proposed this etiology based on features of febrile illness, encephalopathy, and paralysis. However, ChatGPT’s analysis noted misalignment between the historical symptom timeline and the typical presentation of severe WNV encephalitis, which usually involves earlier onset of neurological signs such as dysarthria, nuchal rigidity, seizures, and tremors[16]. Additionally, WNV was only first described in 1937 in Uganda[17] and while its presence in 323 B.C. cannot be ruled out entirely, the epidemiological and clinical inconsistencies reduce its likelihood. ChatGPT’s rejection of WNV as a likely cause aligns with current scientific reasoning, although its sources required human vetting to confirm reliability.

ChatGPT attributed typhoid fever complicated by GBS as the most probable cause of death. This conclusion was based on symptoms including prolonged fever, abdominal pain, and neurological decline[18-20]. However, a review of the sources revealed inconsistencies: Several did not mention typhoid at all, or identified other pathogens, notably C. jejuni, as the most probable precipitant of GBS[18]. C. jejuni is now a well-established cause of bacterial gastroenteritis and one of the most common infectious triggers for GBS, with an incidence of 30.4 per 100000 infections[21,22]. First identified in 1913 in livestock and initially classified under Vibrio, Campylobacter was reclassified in 1973 and only recognized as a major human pathogen in the 1980s when culture techniques advanced. Similarly, GBS was first described in 1859 as “acute ascending paralysis” and formally characterized during World War I in 1916 in two French soldiers. GBS as a complication of C. jejuni may also explain Alexander’s profound paralysis, delayed decomposition, and apparent death. GBS can result in locked-in syndrome a state where a person is conscious but unable to move or speak, potentially mistaken for death in ancient times[23-25].

By contrast, typhoid fever, while common both historically and today, has only a handful of documented associations with GBS. A Boolean search on PubMed using “Guillain-Barré and typhoid” yielded only 11 case reports, mostly in pediatric patients[26-36]. This suggests that typhoid-induced GBS in an adult male especially without modern treatment is unlikely. As the adage goes, “when you hear the sound of hooves, think horses, not zebras” and given Alexander’s symptoms, a C. jejuni gastroenteritis precipitating GBS emerges as a more clinically plausible explanation.

ChatGPT’s ranking of AP as a less likely cause was based on a lack of typical historical features associated with severe or necrotizing AP. While Alexander’s history of excessive alcohol use is noted particularly during his final illness, such as in the instance where he fatally harmed Clitus, the co-commander of the aristocratic cavalry at the end of a long drunken quarrel[37,38], the evidence leans more toward episodic intoxication than chronic alcoholism[39], which limits its relevance as a trigger for AP. Furthermore, the availability of laboratory or radiological data at the time was necessary for severity stratification. Most AP cases are mild and self-limiting, and only 12%-20% result in severe complications[40]. Thus, while alcohol-induced or biliary AP cannot be entirely ruled out, it is a less likely primary cause of death.

Regarding malaria, ChatGPT acknowledged its plausibility due to endemicity and symptoms like fever, chills, and fatigue. However, the absence of intermittent fever patterns, characteristic of malaria caused by Plasmodium vivax, P. ovale, and P. falciparum, reduced its likelihood[41-43]. By contrast, typhoid fever with its sustained fever pattern caused by the release of endogenous pyrogens[44] aligns more closely with the recorded symptomatology, while C. jejuni bacteremia can also present with high fevers as seen in Alexander’s case[45]. Cerebral malaria could also indeed cause a coma-like state as seen in Alexander; however, its course is often marked by cerebral swelling and more overt signs of terminal illness[46].

ChatGPT also reviewed several poisoning hypotheses. Strychnine was rejected due to its rapid onset typically within 15-60 minutes, and characteristic convulsions which were absent in historical accounts[47]. The Styx river water poisoning theory based on calicheamicin lacks both historical and chemical evidence[48,49]. Meanwhile, the White hellebore hypothesis proposed by Schep L is plausible due to its slow-acting nature and overlap with Alexander’s symptoms[50,51], but it still lacks direct supporting evidence and depends on speculative assumptions about alcohol fermentation and toxin delivery[38,52].

Despite its strengths, ChatGPT exhibits well-known limitations. Its sources often include content from health education websites, which lack peer review and are prone to updates. Issues like hallucinated references, as reported by Meyer et al[14], and a high incidence of fabricated citations (69% in one study)[53], highlight its unreliability. These issues represent persistent impurities in generative AI, where hallucinations, false claims, and fabricated references may be presented with unwarranted confidence. Furthermore, ChatGPT functions as a black box as its internal reasoning is opaque[54], limiting the trustworthiness of its outputs. As Starke et al[54] suggest, trust in AI requires reliability, competence, and ethical intent. Moreover, ChatGPT cannot interpret sensory clinical signs (for example, “toxic-looking appearance” or “board-like abdomen”), which limits its diagnostic accuracy without human interpretation. This study highlights a core epistemic lesson: Plausibility without responsibility is insufficient. ChatGPT generated a superficially credible explanation but failed to apply the evidentiary burden required in clinical reasoning. In contrast, the AW framework imposed epistemic discipline by triangulating source credibility, pathophysiological fit, and logical consistency. As LLMs become increasingly embedded in medical reasoning, such guardrails are essential to preserve trust and accuracy.

This study had notable strengths. It is one of the first to explore ChatGPT’s role in investigating a historical medical mystery using a framework rooted in AW. Each hypothesis was cross-referenced against scientific literature to assess the validity of ChatGPT’s conclusions. This study proposes a replicable AW framework for AI-assisted historical diagnostics, introduces a novel causative hypothesis and demonstrates how source triangulation can refine LLM outputs. Future work could include blinded expert panels, benchmarking with other LLMs, and developing a formal AW scoring protocol. Nonetheless, limitations remain. The principle of “garbage in, garbage out”[55] underscores the importance of high-quality input hence, we curated references from peer-reviewed and credible sources. Worryingly, our study found that ChatGPT had the potential to generate fictitious sources and quote sources out of context as discussed in the results segment. Despite the small sample size, the fact that ChatGPT generated these citations and passed them off as contributory towards its conclusions greatly diminishes its reliability, and these discrepancies demonstrate the need for human vetting to validate AI outputs. The simulation relied on publicly available ChatGPT-4, which may behave differently from domain-specific medical LLMs. The assessment of citations and diagnoses involved expert subjective interpretation, though inter-rater agreement was high. The historical records of Alexander’s illness are inherently incomplete, limiting definitive conclusions. Lastly, LLMs are vulnerable to algorithmic bias, especially when user prompts subtly guide responses[56]. Beyond our findings, the growing body of literature on generative AI reflects widespread concerns about its reliability in healthcare. As of July 2025, a PubMed search using the term “ChatGPT” retrieves over 11000 publications, many of which examine its clinical utility and limitations. For example, a recent study by Steele et al[57] compared AI-generated and authentic personal statements in emergency medicine applications and found no statistically significant differences in perceived quality or influence on selection decisions, underscoring the potential for AI to mimic human reasoning even in evaluative domains without necessarily ensuring factual or contextual integrity[57]. These findings reinforce the importance of validating AI-generated content with expert clinical judgment, particularly when such outputs may be persuasive yet flawed. A recent comparative study evaluated ChatGPT-4 alongside other advanced LLMs such as DeepSeek-R1 and Google Gemini in the context of medical education. All three models achieved comparable accuracy exceeding 80% in both basic and clinical medical sciences, underscoring their expanding capabilities in knowledge processing and test performance[58]. However, such promising performance further amplifies the necessity of human oversight when AI tools are applied beyond controlled settings, particularly in clinical or historically ambiguous cases. These discrepancies underscore the need for “AW” models where LLMs serve not as definitive experts, but as co-analysts in a rigorously supervised environment. In sum, this exercise demonstrates that ChatGPT, while generative, is not inherently discerning. The AW framework transforms AI output into responsibly reasoned insight by embedding source vetting, clinical logic, and epistemic humility. This methodology is not only useful for historical forensics but also offers a model for safe AI integration in contemporary diagnostic practice.

CONCLUSION

This study demonstrates that while ChatGPT can generate plausible hypotheses regarding historical medical events, it remains vulnerable to citation errors, logical overreach, and source misapplication. By applying AW, which relies on human clinical judgment to critically assess AI outputs, C. jejuni-induced GBS was determined as a more credible cause of Alexander the Great’s death than typhoid fever. AW offers a novel pathway for fusing computational power with contextual nuance, especially in settings where medical, historical, and ethical ambiguity intersect.