Ethics and AI in Humanitarian Crisis Management: Saving Lives or Automating Indifference?
In a refugee camp, an algorithm decides who eats. Is it efficiency or inhumanity? Artificial Intelligence is transforming humanitarian crisis management, from f
Imagine a refugee camp in Yemen or on the Sudanese border. Water resources are scarce, food is rationed by the gram, and thousands of people press at the gates, fleeing conflicts or climate disasters. In this apocalyptic scenario, who decides who gets the last bag of rice or the last available tent? Until yesterday, this agonizing decision fell to an exhausted humanitarian worker, influenced by fatigue, stress, and their own inevitable unconscious biases. Today, more and more often, this decision is delegated to an algorithm. A system that has calculated the "vulnerability score" of each family based on biometric data, iris scans, and satellite analysis.
Artificial Intelligence promises to bring efficiency to the ungovernable chaos of humanitarian crises. It promises to predict famines months before crops fail, to clear contaminated land without risking human lives, and to optimize aid logistics like a global, ultra-high-precision supply chain. But when the "goods" in question are not a postal package but human survival, algorithmic efficiency violently clashes with millennia-old ethical principles. How can we ensure that AI does not turn human rights into variables in an equation? How do we prevent data collection from becoming a form of predatory surveillance?
In this "pillar" article for the Ethics and Society column, we will explore the thin line between aid and control, analyzing the ethical frameworks of the Red Cross, the risks of lethal bias, the phenomenon of "data colonialism," and the imperative need to maintain a "Human-in-the-Loop" when decisions are literally life or death.
1. The Technological Promise: Predicting the Unpredictable and Optimizing Hope
Humanitarian organizations operate in environments defined by uncertainty and a scarcity of reliable data ("data scarcity"). AI offers the revolutionary possibility of transforming fragments of messy data into life-saving preventive action.
Anticipating Disasters: "Forecast-based Financing"
The traditional aid model is reactive: a disaster happens, funds are raised, aid is sent. Often, it's too late. As reported by the Italian Red Cross (CRI) and discussed at the Soochow Forum (cri.it), AI is enabling Forecast-based Financing (FbF). Machine Learning algorithms analyze complex weather patterns, historical rainfall data, and local market prices to predict floods or droughts weeks, sometimes months, in advance. This allows funds to be automatically released before the disaster strikes. Veterinary kits are distributed to save livestock before a drought, or embankments are reinforced before a flood. It's not magic; it's statistics applied to survival, transforming aid from posthumous charity into a preventive investment.
AI Agents and Crisis "Digital Twins"
An even more advanced frontier, described by the UNU-CPR (United Nations University Centre for Policy Research) (unu.edu), is the use of "AI Agents" to simulate complex crisis scenarios. By creating "digital twins" of a refugee camp or an affected region, operators can populate these virtual worlds with digital personas that simulate the behavior of real refugees (based on anonymized data). Operators can then test crucial questions: "What happens if we distribute food only to female heads of household?", "How do migration flows change if we close this crossing?". This allows for virtual failure and strategy correction without causing real harm to already vulnerable people.
Demining and Computer Vision
In post-conflict zones, landmines remain a threat for decades. AI is drastically accelerating demining operations. Drones equipped with thermal and hyperspectral cameras fly over vast areas, and Computer Vision algorithms identify anomalies in the ground that indicate the presence of unexploded ordnance. The achieved accuracy surpasses that of the tired human eye, allowing "red zones" to be mapped and arable land to be returned to communities in record time.
2. The "Data Colonialism" Dilemma: Consent and Coercion
If AI is a powerful tool, it is also voracious. It feeds on data. And in a humanitarian context, the extraction of this data raises ethical questions that go far beyond standard privacy.
The Paradox of Informed Consent
Giving Compass (givingcompass.org) raises a crucial point by analyzing cases like that of Yemen. Does it make sense to talk about "informed consent" when the alternative is starvation? If a humanitarian agency says: "To receive your monthly food ration, you must allow us to scan your iris and enter you into our global biometric database," the refugee has no real choice. They will do it to survive. This dynamic transforms data collection into a form of coercion. The bodies of vulnerable people become data deposits to be extracted, creating an extremely unbalanced power relationship that many scholars define as Digital Colonialism: data is extracted in the "Global South," processed and monetized by technology companies in the "Global North," and used to make decisions that fall back on the South, often without any algorithmic transparency.
The Biometric Data "Honeypot"
The centralization of biometric data (iris, fingerprints, face) of millions of refugees creates an irresistible target (a "honeypot") for malicious actors. Wilton Park (wiltonpark.org.uk) warns that in contexts of civil war or ethnic persecution, anonymity is often the only defense left. If a humanitarian database is hacked or, worse, requisitioned by a hostile government or militia, that list of beneficiaries instantly turns into a "kill list." AI, with its ability to re-identify people by cross-referencing seemingly anonymous data (e.g., cell phone location and social media posts), further erodes this protective shield. The principle of "Do No Harm" is severely tested by the very technology that is supposed to help.
3. Algorithmic Bias: When Mathematics Discriminates
As we often explore on La Bussola, Algorithmic Biases are not simple technical errors ("bugs"), but mirrors of social and historical inequalities. In a humanitarian crisis, a bias doesn't deny you a bank loan; it denies you survival.
Invisible Discrimination in Triage
Imagine an algorithm designed for medical triage in an overcrowded field hospital. If the AI was trained on historical Western health data or on local data reflecting past discrimination (e.g., less access to care for a certain ethnic minority), the model might learn that this ethnic group has "lower survival rates" or "less chance of recovery." Consequently, the algorithm might assign that group a lower priority score, recommending not to waste scarce resources on them. The AI is just statistically maximizing efficiency, but ethically it is automating eugenics. As highlighted by MOAS (moas.eu), without constant auditing of training data, we risk encoding racism and social exclusion directly into the relief software.
The Digital Divide and the Invisibility of the Poor
AI relies on digital data. But who generates digital data in a crisis? Those who have a smartphone, those who are connected. This creates a huge representation bias. Women, the elderly, people with disabilities, and the poorest rural populations are often "invisible" to digital sensors. An algorithm mapping needs based on cell phone signals or social media posts will only see the wealthier, male part of the affected population, completely ignoring the most vulnerable. AI risks directing aid towards those with a digital voice, leaving the analog to die.
4. Human-in-the-Loop: The Necessity of Human Judgment
Faced with these existential risks, the international and scientific community agrees: AI cannot and must not decide alone.
Life-or-Death Decisions
The European Commission's Scientific Advice Mechanism (scientificadvice.eu) establishes an inviolable principle: in the acute phases of a crisis, critical decisions ("life-and-death decisions") must always have significant human supervision (Human Oversight). AI can and must process data, identify patterns, and suggest options (e.g., "Option A saves more people but is riskier for staff; option B is safer but slower"). But the final choice must belong to a human being capable of moral responsibility and understanding the non-codifiable context (e.g., an unofficial truce just verbally negotiated, which the algorithm cannot know).
Accountability and "Black Box AI"
Who is responsible if an autonomous drone delivering medicine crashes into a house, or if a food distribution algorithm mistakenly excludes an entire village? The ICRC (International Committee of the Red Cross) (international-review.icrc.org) works tirelessly to anchor the use of AI to International Humanitarian Law. Autonomous systems do not operate in a legal vacuum. The problem of the "Black Box" – i.e., Deep Learning algorithms so complex that not even their creators can explain how they reach a conclusion – is unacceptable in the humanitarian sector. Organizations must be able to explain to a mother why her family was excluded from aid. The answer "the computer decided" violates human dignity.
This need for transparency brings us back to the theme of the democratic governance of technology, which we explore in detail in our article on AI and Governance: Between Utopia and Dystopia.
5. Real Cases and Best Practices: Towards Responsible AI
Despite the shadows, there are lights. There are concrete examples of how ethics can be integrated into the code from the start ("Ethics by Design").
Fighting the Infodemic with AI
During health crises (like Ebola or COVID-19) or conflicts, fake news kills as much as viruses or bullets. The Umma Foundation (ummafoundation.org) cites UN projects (like Global Pulse) that use NLP to monitor social media and local radios in real time. The goal is not censorship, but to detect dangerous rumors (e.g., "the vaccine is poison," "humanitarian workers bring disease") to allow organizations to intervene quickly with targeted, corrective information campaigns. In this case, AI acts as a shield for factual truth. (To explore language analysis mechanisms, see AI and Language: Words and Truth).
Participatory Design
To avoid colonial biases, the solution is to involve affected communities in the design of tools. Instead of parachuting technological solutions from Silicon Valley, the most virtuous projects develop algorithms with refugees and local operators. This approach allows for understanding what the real needs are (perhaps the priority is not facial recognition, but a simple SMS system to report polluted wells) and building trust. AI thus becomes a tool for local empowerment, not external imposition.
Conclusions: Technology Serving Humanity, Not Vice Versa
Artificial Intelligence in humanitarian crises is a powerful drug: it can cure incurable ills, but if overdosed or administered poorly, it becomes poison. It can be the greatest force multiplier for good we have ever invented, capable of bringing eyes and ears where no human can safely go. Or it can be an engine of algorithmic inequality that dehumanizes victims, turning them into datapoints to optimize and surveil.
The difference will not be made by the computing power of GPUs, but by the solidity of our moral compass. We must demand a humanitarian AI that is:
- Transparent: Explainable, verifiable, and open to external audit.
- Fair: Actively designed to mitigate biases and include the invisible.
- Human: Always, rigorously subordinate to human dignity and judgment.
Only then can we say that AI is truly "saving the world," and not just coldly optimizing its management.
Bibliographic References and Further Reading
To ensure maximum authority on such a delicate topic, this article is based on official reports, academic research, and international guidelines:
- Ethical Frameworks and Human Rights:
- ICRC (International Committee of the Red Cross) – AI in humanitarian action and international law. PDF Link, Article Link
- La Bussola dell’IA – Algorithmic bias and discrimination. Link
- Ethical Challenges (Consent, Privacy, Bias):
- Operational Applications and Risks:
- CRI (Italian Red Cross) – Technology and operational risks (Soochow Forum). Link
- UNU-CPR – AI agents for scenario simulation. Link
- Umma Foundation – Fact-checking and aid distribution. La Bussola dell'IA · Articoli · Rubriche