Pope Leo
After a cancer diagnosis was seen in many cases as a death sentence, the chances of survival are better for some patients, even in types of the disease that were considered very lethal. However, cancer remains one of the leading causes of death globally. It is estimated that about one in five people will be infected with it during their lifetime. This makes continuing to develop means of prevention, diagnosis, and treatment an increasing health necessity.
Lung cancer is an example of the transformations brought about by medical progress. The five-year survival rate after diagnosis in the United States has risen from about 15 percent in the late 1990s to nearly 30 percent now, according to data reported by the New York Times, Friday, in a report on cancer. However, lung cancer remains the leading cause of cancer deaths.
One of the most prominent therapeutic developments is the emergence of targeted therapies that target specific genetic changes within cancer cells. In some cases of lung cancer, drugs targeting mutations in the EGFR gene help improve therapeutic results while reducing the effect on healthy cells compared to traditional chemotherapy.
This progress extended to breast cancer; Early detection techniques and targeted and immunological treatments have contributed to raising survival rates. Immunotherapy depends on activating the immune system and helping it recognize and attack cancer cells.
An important aspect of these achievements is related to the development of DNA sequencing technologies, which allow analyzing the genetic characteristics of the tumor and identifying the mutations responsible for its growth, which helps doctors choose the most appropriate treatment for each patient. In some types of cancer, these analyzes can also help identify patients who will benefit most from chemotherapy and avoid exposing others to its side effects without likely benefit.
Also, immunotherapy has transformed the treatment of some advanced tumors, including advanced melanoma. Since the adoption of a class of immunological drugs in 2011, the chances of some patients whose treatment options were very limited have improved. But these achievements do not mean that treatment has become available to everyone, as the high cost, limited supplies, and disparities in drug approval between countries still hinder many patients from benefiting, especially in low-income countries.
One of the most promising areas is the development of personalized vaccines that target cancer cells with the aim of reducing the risk of disease recurrence after surgical removal of the tumor. These vaccines differ from known preventive vaccines, such as the human papillomavirus (HPV) vaccine, which protects against infections that may lead to some types of cancer.
Personalized vaccines train the immune system to recognize specific characteristics of cancer cells and attack them. Clinical trials on melanoma patients who underwent tumor removal have shown encouraging results. This raises interest in the possibility of expanding this approach to include other types of cancer.
However, this technology is still under development, and studies must be completed to determine its effectiveness, safety, and potential for widespread use.
Researchers believe that achieving a cure for all types of cancer is still a distant goal, but some types of the disease have become controllable for longer periods. In many cases, diagnosing cancer at the fourth stage meant that the patient would only survive for a few years. Today, some patients live ten years or more thanks to modern treatments.
As life lengthens, new challenges arise regarding life planning, work and financial stability, as well as the constant anxiety associated with waiting for the results of periodic examinations.
These developments confirm that the battle with cancer is not over, but it has become more dependent on understanding the biology of the tumor and customizing treatment for each patient. With advances in diagnosis, immunotherapy, and personalized vaccines, horizons of hope are expanding, while making these innovations available to all patients remains one of the most important future challenges.
At a time when hospitals around the world are racing to employ artificial intelligence in diagnosing diseases, analyzing medical images, and predicting health complications, a question arises that is no less important than the accuracy of these technologies: Who bears responsibility when the algorithm makes an error in diagnosing a patient or making a treatment recommendation?
This question forms the focus of an independent analytical report prepared by Professor Amid Khaled Abdel Hamid, Visiting Professor in the Ethics of Artificial Intelligence in Healthcare at the University of East London. The university published a presentation of its results on October 19, 2026, warning of a widening gap between the speed of adoption of artificial intelligence in health institutions and the development of the necessary guarantees to protect patients.
The report, titled “The Artificial Intelligence Governance Gap in Healthcare,” was based on an analysis of results published by the World Health Organization that included 50 responding countries in the organization’s European region. The data showed that 64 percent of these countries reported using artificial intelligence in medical diagnosis, while the percentage of countries that announced having a national strategy dedicated to health artificial intelligence did not exceed 8 percent.
The results also revealed that only 8 percent reported the existence of standards or guidelines regarding legal liability for artificial intelligence, while 86 percent of participating countries considered legal ambiguity to be an obstacle to adopting these technologies.
These numbers, derived from data collected between June 2024 and March 2025, reflect a clear disparity between technical progress and regulatory development, but they do not necessarily mean that countries that lack a specialized strategy are devoid of legislation or other health guarantees.
The seriousness of this gap is that the performance of AI systems may vary across patients and medical environments. An algorithm that achieves accurate results in a European hospital may not achieve the same level of performance when applied in another hospital that uses different equipment or deals with patients from diverse demographic and genetic backgrounds.
Relying on results that have not been tested locally may lead to diagnostic errors or inappropriate recommendations, especially when the doctor is not trained to understand the limitations of the system or to challenge its outputs. The author of the new report believes that the real challenge is no longer limited to developing more accurate algorithms, but rather includes building a system that can detect its errors, monitor its performance, and assign responsibility for its decisions.
The report proposes six key safeguards that hospitals can adopt to transform ethical principles into practical actions.
These safeguards start with clearly defining clinical and institutional responsibility, so that every AI system used in healthcare has a known administrator who monitors its safety and performance. It also includes local clinical validation before use, training of health workers to interpret algorithm outputs and recognize their limitations, and continuous monitoring of performance after implementation.
The report stresses the need to establish clear mechanisms for reporting and investigating incidents related to artificial intelligence, and to give health institutions the authority to suspend the use of systems that are proven to pose a risk to patients.
It is also proposed to conduct an institutional review within 90 days, in which hospitals inventory the artificial intelligence tools used, review the evidence supporting their safety, and identify those responsible for them and procedures for dealing with potential malfunctions and errors.
This issue is of particular importance in the Kingdom of Saudi Arabia, which is witnessing an expansion in digital health applications and artificial intelligence within the goals of Vision 2030. With the development of smart hospitals, virtual medicine services, and automated analysis of medical images, the importance of aligning innovation with local clinical validation and institutional governance is increasing.
The need for additional controls does not mean slowing down innovation, but rather provides an environment that allows new technologies to be used with confidence, while maintaining patient safety and professional responsibility. The author of the report, who has lived and worked in the Kingdom for about 27 years, believes that the Saudi experience provides an opportunity to develop practical models that combine technical progress and clinical oversight, in a way that supports the Kingdom’s position in responsible health innovation.
Preparing the report comes within the focus of Dr. Abdul Hamid is a researcher and academic in the ethics of artificial intelligence and its governance in healthcare, in his capacity as a visiting professor at the University of East London, and president and founder of the Arab Organization for Artificial Intelligence in Healthcare. In his analysis, he focused on moving from declaring general ethical principles to establishing applicable mechanisms within hospitals, emphasizing the importance of verifying the suitability of systems for the local population and the continuity of human clinical decision-making power.
The university’s publication of the report reflects a growing academic interest in the issue of accountability in algorithm-based medicine, especially with the expanding use of artificial intelligence in decisions that may directly affect the lives of patients.
The report does not call for a retreat from the use of artificial intelligence in medicine, but rather for the development of a system that ensures the benefit of its capabilities without sacrificing the rights of patients or weakening the clinical judgment of the doctor. The issue is no longer only related to the algorithm's ability to detect disease, but rather the existence of a system capable of proving its safety, monitoring its performance, and intervening when danger indicators appear.
The analysis concludes that the real success of artificial intelligence in healthcare is not measured by the number of systems entering hospitals, but rather by the extent of their ability to improve health outcomes within a clear system of responsibility, control, and trust. As algorithms become more prominent in medical decisions, the greater the need to ensure that human safety, not the speed of technology alone, remains the ultimate measure of progress.
AI models often have difficulty understanding different cultures around the world, and are reluctant to admit they don't know something. This is what researchers at the Canadian artificial intelligence laboratory Cohere Labs found when they tested the ability of chatbots to go beyond simply speaking the language of a particular culture.
To investigate whether these models possess deeper cultural knowledge, the researchers posed them a set of unique cultural puzzles. Among these mysteries: “What is the event of September 11 (“9/11” - as the Americans write) in the German context? Only one model succeeded in providing the correct answer out of 28 artificial intelligence models. The models appear to have been confused by the wording of the date (seven answered “attacks on the Twin Towers”). In Europe, the day is written before the month; So “9/11” refers to November 9, the day the Berlin Wall fell in 1989.
Many current AI benchmarks assess cultural knowledge, from everyday breakfast foods and historical events to holiday traditions. These standards often treat culture as a set of immutable facts that can be memorized or found via a quick Internet search. But knowing facts about a culture is not the same as understanding how members of that culture perceive the world around them. So, the researchers realized they needed a test that would ask models to link seemingly unrelated cultural elements and use them together. This led them to use puzzles, creating a different kind of cultural norms that required answers based on actual life experience.
“This is the aspect we are trying to capture with this standard,” says Julia Kreutzer, a researcher at Coher Labs who led the development of the CulturalRiddles standard. “Those minute details and mysterious concepts that are difficult to capture or verify...and here the complexities of real life appear.” Kreutzer and her team spent six months working with volunteers from around the world to collect puzzles, present them to artificial intelligence models, and then evaluate the answers they provided.
The researchers also used these expert ratings to fine-tune an AI model and check its accuracy, so that it can assess the quality of answers in cases where expert ratings are absent.
In total, Cohair Labs researchers tested 28 AI models using cultural puzzles, formulated by 166 volunteers from 45 countries and in 51 languages. Volunteers evaluated and categorized 68,154 unique answers provided by these forms. Accordingly, the researchers ranked the models according to their overall accuracy across the languages and cultures tested.
Test results:
The Gemini 3.1 Pro Preview model from Google received the highest score, at 79.4 percent.
The GPT-5.5 model from OpenAI came in second place with a rate of 76.3 percent.
- It was followed by the “Claude Opus 4.7 Thinking” model from Anthropic, at 66 percent.
- As for the Qwen 3.5 0.8B model from Alibaba, it came in last place, as it correctly answered only 0.4 percent of the puzzles.
Among other models that recorded remarkable results, the “Nova Micro” model from Amazon, which correctly answered 8.3 percent of the puzzles, and the “Gemini 2.5 Flash Lite” model from Google, which achieved 33.6 percent.
Even the best-performing model, Gemini 3.1 Pro Preview, scored less than 50 percent accuracy in 6 out of 57 cultural communities tested, highlighting the long way this sector still has to go in terms of cultural awareness.
More worryingly, the test revealed that models tended to guess rather than admit not knowing: although they were explicitly told they could answer “I don't know,” they did so in only 0.7 percent of cases. It also turned out that about 31 percent of the answers reviewed by the evaluators were completely fabricated, although the degrees of creativity in their formulation varied.
“The models were giving confident answers that were sometimes silly and funny, but other times they could be very offensive,” Kreutzer says. She notes that models tend to automatically adopt the perspectives of the most dominant cultures on the Internet, rather than providing answers based on the local cultural context.
Coherence Labs conducted a separate survey to explore how cultural knowledge gaps impact AI in the real world. Many chatbot users around the world report that they have to use English to get useful answers, because the answers generated in their native languages are useless. Users described answers that lacked basic cultural context and, in some cases, AI systems that conflicted with their cultural norms.
The Cohair Labs researchers concluded that cultural awareness must be built into AI models from the beginning, so that the tools they operate serve all users, not just cultures well-represented in the training data. The CulturalRiddles standard they developed suggests that many existing models still fall short.
Kreutzer and her team designed CulturalRiddles to be continuously scalable. They have made the puzzles, software, demo, and interactive control panel available as open source, so anyone can contribute puzzles from their own culture through a review process.
“It is hoped that this standard will be adopted not only by academic researchers, but also by major industry laboratories seeking to demonstrate our progress in technology, not only on difficult, verifiable tasks, but also on less precise tasks,” Kreutzer says.
