Exploring the Science Behind Biological Age Assessment: Can AI Really Measure Our Age?

Imagine possessing a magical device that could reveal your remaining lifespan by simply analyzing your facial features. A new AI tool named FaceAge claims to provide insights into how quickly our bodies are aging and how this might correlate with our longevity. Intrigued and somewhat anxious, I submitted my photograph to FaceAge, awaiting its findings.

Before diving deeper into my results, I spoke with one of the FaceAge creators, Raymond Mak, an associate professor at Harvard Medical School. He admitted his fear of confronting his own mortality through the machine he helped develop.

“Were you apprehensive?” I inquired. It’s difficult to reconcile that fear with his scientific credentials, but as our conversation continued, my unease grew.

FaceAge has demonstrated in clinical trials that how we appear—specifically in terms of perceived age—can predict our mortality. Although the analysis of my photo involves a brief wait as it travels to Harvard and back, in reality, this AI can make assessments in milliseconds using just a selfie. By moving beyond superficial indicators like wrinkles or gray hair, FaceAge evaluates subtle physiological changes that indicate aging.

Mak and his co-founder, Hugo Aerts, who leads the Artificial Intelligence in Medicine program at Mass General Brigham, believe this technology could be used positively as a new vital sign, aiding in medical assessments. The implications are vast, potentially allowing us to gauge the life expectancy of anyone—from a partner to a political candidate—whether they consent or not.

“It’s reminiscent of something from Gattaca,” Mak reflected during our video call, drawing parallels to the film where individuals are judged by their genetic predispositions. He expressed his personal reluctance to know such information.

In today’s aesthetics-driven society, where youthfulness is idolized through social media and cosmetic procedures, FaceAge exemplifies our obsession with external appearances. As Oscar Wilde articulated, the mystery of life often lies in the visible rather than the invisible.

According to Aerts, “Your perceived age is significant. If you look older than those around you, you’re likely to have a shorter lifespan. This correlation had not been proven until now.”

The development of FaceAge stemmed from Mak and Aerts’s prior collaboration aimed at refining machine learning to enhance the accuracy of tumor detection in medical scans. They recognized an opportunity to improve a commonly relied upon method in medicine known as the eyeball test, where physicians estimate patient conditions based on physical appearance.

In a striking study, eight oncologists were shown photographs of 100 terminally ill patients to predict their survival over the next six months. Typically, they achieved only a 60% accuracy rate, performing barely better than random chance. However, when given access to the patients’ medical histories, their accuracy improved to 70%. This demonstrated a significant level of uncertainty in visual assessments of health and mortality. Could algorithms outperform human judgment?

Motivated by this potential, Aerts and Mak set out to train an AI system that could assess biological age based on observable traits not typically factored into human assessments. Unlike age estimation apps that focus on the youthful appearance of teenagers, FaceAge aims to identify signs indicating deteriorating cellular vitality.

In their study, researchers applied FaceAge to over 6,000 cancer patients before they began radiological treatment, utilizing standard images captured by healthcare staff. The AI had been trained on a dataset of nearly 60,000 healthy individuals, a number that has now expanded to 40 million.

They found that cancer patients typically appeared five years older than they were, an early indication that FaceAge might hold genuine relevance. A clear pattern emerged: the more advanced the cancer, the older the patients looked, and crucially, the AI’s estimation of biological age was predictive of survival outcomes.

“This was a significant realization,” stated Mak. “It indicated a clinically useful signal; the older you appear, the less time you may have.”

The distinctions made by FaceAge are not limited to traditional markers of aging, unaffected by cosmetic interventions. It evaluates often overlooked features, such as the size of the temporalis muscle, which is linked to overall strength and vitality. Additionally, varying characteristics of the eyes and nose structure contribute to its assessments, enabling it to generate accurate predictions of mortality.

When tasked with predicting which patients might succumb within six months based solely on photographs, FaceAge achieved over 80% accuracy—surpassing seasoned medical professionals who had full access to medical data.

Should healthcare practitioners be cautious about solely relying on visual assessments? “Definitely,” Mak warned. While some might dismiss the need for AI in gauging age, utilizing a tool to refine clinical judgments could enhance patient care.

Is the day approaching when AI programs could be integrated into everyday devices, providing health assessments from our reflections?

“Absolutely,” Mak affirmed. The project has shown that tracking visual changes over time yields better accuracy than a single image. A dramatic shift in appearance could signal concerning health trends.

Despite the exciting prospects, Mak recognizes the extensive research required before FaceAge can be deemed a clinical biomarker. He envisions a future where it serves as an early alert system, prompting individuals to consult their healthcare providers for potential health issues.

However, the simplicity of a photograph poses ethical dilemmas that have yet to be fully explored. “Our team harbors concerns regarding potential misuse,” Aerts admitted.

While a photograph is less intrusive than a blood test or MRI, its implications can be profound—ones that could affect employment opportunities and healthcare decisions based purely on perceived biological age derived from publicly available images.

“Imagine entering a store and undergoing an instantaneous health evaluation based solely on your photo. That is plausible and far from a novelty,” Aerts cautioned. “We could ascertain someone’s biological age and its implications for mortality. It’s a double-edged sword that necessitates careful consideration of its benefits and pitfalls.”

Curious about its capabilities, I asked Aerts if he had tested the AI on notable individuals. He revealed they opted against using it frivolously but acknowledged that personal tests had yielded encouraging results, despite its limitations for those under 50. Mak finally tried it himself, receiving a ten-year age reduction, leaving him intrigued yet cautious.

After receiving my photographic evaluation, FaceAge suggested I appeared much younger than my actual age—with one image indicating as little as 32 years old, a full two decades younger. While initially gratifying, the stark difference left me feeling skeptical, considering my family’s history of health issues. Although flattering, this assessment felt exaggerated, similar to a disingenuous compliment from a waiter. I find myself questioning the accuracy of FaceAge.