Patient Attitudes Toward Artificial Intelligence in Jordanian Healthcare: A Cross-Sectional Survey Study

A cross-sectional survey of 500 patients in Jordan reveals that while there is conditional optimism regarding AI in healthcare, acceptance is heavily dependent on maintaining human physician involvement, ensuring transparency and privacy, and addressing disparities linked to education and digital literacy.

The Gist

Imagine the healthcare system as a bustling, high-stakes kitchen. For years, the chefs (doctors) have cooked every meal from scratch, relying entirely on their own taste, experience, and hands. Now, a new tool has arrived: a super-smart, robotic sous-chef (Artificial Intelligence) that can chop vegetables in a millisecond, calculate the perfect cooking temperature, and even suggest new recipes.

But before the kitchen can start using this robot, the people waiting for their meals (the patients) need to decide: Do we trust this robot? Do we want it in our kitchen? And how much should we let it cook?

This study is like a survey sent out to 500 hungry customers in three major cities in Jordan (Amman, Balqa, and Irbid) to ask exactly those questions. Here is what they found, translated into everyday language:

1. The "Human-in-the-Loop" Rule

The most important finding is that no one wants the robot to take over the kitchen entirely.

When asked, "Would you rather the robot cook alone, or work alongside the chef?" almost everyone shouted, "Work alongside!"

• The Analogy: Think of AI as a GPS in a car. You love the GPS for telling you about traffic jams and the fastest route (it's useful!), but you still want to keep your hands on the steering wheel. You want the doctor to be the driver, with the AI as the co-pilot giving advice.
• The Result: Patients felt strongly that human empathy, a warm hand on the shoulder, and a doctor's personal touch are things a robot simply cannot replace.

2. The "Trust vs. Fear" Balancing Act

The patients were generally optimistic, but they had a few reservations, like a cautious driver checking the rearview mirror.

• The Good News: They believed the robot could actually help. They thought it might make diagnoses more accurate and save time.
• The Worry: They were a bit nervous about their personal data. "If this robot knows my medical history, is it safe? Will my secrets be leaked?"
• The Verdict: The fear of privacy leaks didn't stop them from wanting to use the tech, as long as they trusted the doctors managing the robot. If the doctor says, "I'm watching the robot, and your data is safe," the patients were willing to give it a try.

3. The "Digital Divide" (Who is ready?)

Here is where the study found a crack in the road. Not everyone is ready to hop in the car with the GPS at the same speed.

• The Education Gap: People with less formal education felt less ready to use AI. It's like trying to use a complex smartphone app when you've only ever used a basic flip phone. They felt a bit lost and unsure.
• The Skill Gap: People who rated their own "digital skills" as high (people who are comfortable with apps, websites, and gadgets) were much more excited and ready to use AI.
• The Experience Factor: People who had already tried a health app or a smart device before were more trusting. It's the difference between someone who has never driven a car and someone who has driven a few times; the experienced driver is less scared of the new features.

4. The "Robot" Isn't Everywhere Yet

Interestingly, even though everyone was talking about AI, most of the patients hadn't actually seen it in action yet. Their exposure was low.

• The Analogy: It's like everyone talking about flying cars. People have opinions about them, they think they sound cool, but very few have actually sat in one. Because they haven't used it much, their opinions are based on what they hear and imagine, not on real experience.

The Big Takeaway: How to Serve the Meal

The authors suggest that if Jordan (and other places like it) wants to introduce AI into hospitals, they shouldn't just throw the technology at the patients and hope for the best. Instead, they should:

1. Keep the Chef in Charge: Always make it clear that the AI is just an assistant, not the boss. The human doctor must remain the face of the care.
2. Be Transparent: Don't hide how the robot works. Explain it simply. "This computer helps us double-check our work."
3. Teach the Customers: Run workshops to help people who aren't tech-savvy feel more comfortable. If you teach someone how to use a smartphone, they stop fearing it.
4. Protect the Secrets: Build a fortress around patient data so people know their privacy is safe.

In short: The patients in Jordan are saying, "We love the idea of a smart assistant helping our doctors, but we want to hold the steering wheel, we want to know our secrets are safe, and we need a little help learning how to use the new dashboard."

Technical Summary

1. Problem Statement

While Artificial Intelligence (AI) is increasingly integrated into global healthcare systems for diagnosis, triage, and treatment, patient acceptance in resource-constrained, middle-income settings remains poorly characterized. Existing literature often focuses on Western contexts or healthcare professionals/students rather than patients. There is a critical gap in understanding how patients in the Arab region, specifically Jordan, perceive AI-supported care. Key unknowns include:

• The balance between perceived usefulness and concerns regarding privacy, transparency, and the loss of human empathy.
• How demographic factors (education, age) and digital literacy influence the "readiness" to accept AI.
• Whether patients view AI as a replacement for clinicians or a supportive tool.

2. Methodology

Study Design & Setting:

• Type: Cross-sectional survey.
• Location: Seven tertiary public and academic teaching hospitals across three Jordanian governorates: Amman, Balqa, and Irbid.
• Timeline: Data collection occurred between July and October 2025.
• Sample: A convenience sample of 500 adult patients (aged ≥18) capable of informed consent.

Data Collection & Instruments:

• Administration: Trained staff administered an electronic questionnaire (Arabic) via smartphones/tablets.
• Pilot Testing: The instrument was piloted with 44 participants to refine item wording.
• Measures: The survey utilized 25 Likert-type items (1 = Strongly Disagree to 5 = Strongly Agree) grouped into six constructs:
  1. Exposure: Prior experience with AI in healthcare.
  2. Trust & Transparency: Belief in AI explainability and governance.
  3. Perceived Usefulness: Expectation of performance benefits (accuracy, efficiency).
  4. Privacy & Perceived Risks: Concerns about data security and bias.
  5. Empathy & Human Interaction: Preference for human involvement.
  6. Readiness/Behavioral Intention: Willingness to accept AI-enabled care.
• Covariates: Demographics (age, gender, education), prior AI app/device use, and self-rated digital skill.

Statistical Analysis:

• Software: Python (pandas, SciPy, statsmodels).
• Reliability: Cronbach's alpha ($\alpha$) calculated for internal consistency.
• Comparisons: Welch's t-tests (gender, prior use) and One-way ANOVA (age, education, digital skill) with Tukey post-hoc tests.
• Correlations: Pearson correlation coefficients ($r$) to assess relationships between constructs.
• Regression: Multiple linear regression to identify independent predictors of "Readiness," controlling for demographics and digital skills.

3. Key Results

Descriptive Statistics (Attitudinal Domains):

• Highest Endorsement: Empathy & Human Interaction ($M=4.33, SD=0.58$). Patients strongly believe AI should not replace human care.
• Highest Specific Item: "I prefer AI to work alongside the physician rather than relying on it alone" ($M=4.47$).
• Perceived Usefulness: Relatively high ($M=3.97$).
• Trust & Readiness: Moderate to high ($M=3.57$ and $M=3.66$, respectively).
• Privacy Concerns: Moderate ($M=3.51$), indicating salient but not paralyzing fears.
• Exposure: Lowest score ($M=2.57$), indicating limited direct experience with AI in Jordanian clinical settings.

Reliability:

• Trust & Usefulness: Acceptable internal consistency ($\alpha \approx 0.75$).
• Other Constructs: Modest reliability ($\alpha$ ranging from 0.59 to 0.66), suggesting heterogeneity in how patients interpret items related to exposure, privacy, and readiness.

Group Differences:

• Education: Significant predictor. Participants with lower educational attainment (<High School and High School) reported significantly lower Readiness compared to those with Bachelor's degrees.
• Digital Skill: Strong positive correlation. Higher self-rated digital skill predicted higher Readiness, Trust, and Perceived Usefulness.
• Prior AI Use: Those with prior experience reported higher scores across Exposure, Trust, Usefulness, and Readiness compared to non-users.
• Demographics: Gender and Age showed statistically significant but small effect sizes (e.g., females had slightly higher exposure; older adults had lower trust). These were less predictive than education and digital skill.

Correlations & Regression:

• Key Drivers: Trust ($r=0.48$) and Perceived Usefulness ($r=0.44$) were positively correlated with Readiness ($p < .001$).
• Barriers: Privacy concerns were negatively correlated with Trust and Usefulness.
• Regression Model: The model explained 10.1% of the variance ($R^2=0.101$) in Readiness.
  • Significant Predictors: Lower education (negative coefficient) and higher digital skill (positive coefficient).
  • Non-Significant Predictors: Gender, age, and prior AI use (when adjusted for other variables).

4. Key Contributions

1. Context-Specific Evidence: Provides the first patient-centric data on AI acceptance in Jordan, filling a gap in the literature regarding middle-income, resource-constrained settings.
2. Nuanced Acceptance Model: Demonstrates that "conditional optimism" exists; patients are optimistic about utility but demand human-in-the-loop systems. They do not view AI as a replacement for clinicians.
3. Identification of Modifiable Levers: The study shifts the focus from static demographics (age/gender) to modifiable factors: digital literacy and education. It suggests that acceptance gaps are driven by capability and familiarity rather than inherent resistance.
4. Psychometric Insights: Highlights that "oversight and shared control" may be a distinct dimension of AI acceptability that cuts across traditional scales, suggesting a need for refined measurement tools in Arabic-speaking populations.

5. Significance & Implications

• Implementation Strategy: Successful AI integration in Jordan should prioritize human-centered workflows where AI supports, rather than replaces, physicians.
• Trust Building: Interventions must focus on transparent communication and robust data governance to address privacy concerns, which are inversely related to trust.
• Equity & Digital Divide: To prevent widening acceptance gaps, health systems must invest in digital literacy programs and targeted education for lower-educated and lower-skill populations.
• Policy: Regulators should emphasize "clinician-supervised" AI frameworks to align with patient preferences for human oversight.

Limitations Noted by Authors:

• Cross-sectional design prevents causal inference.
• Convenience sampling limits generalizability to the entire Jordanian population.
• Self-reported data may be subject to social desirability bias.
• The regression model explained a relatively small portion of variance, indicating other unmeasured factors (e.g., specific AI use cases, fairness perceptions) influence acceptance.