Once confined to the realms of science fiction or relatively crude internet death calculators, AI-driven predictions about longevity are inching closer to reality. Questions about the accuracy and value of these forecasts remain.
In recent years, researchers and companies around the globe have been pursuing answers to the ultimate question: How long have we got left? These models leverage cutting-edge tools, such as artificial intelligence and machine learning, drawing on a variety of parameters to deliver statistically grounded insights.
Conceptually, they function like established diagnostic tools such as QRisk for heart disease or CHA₂DS₂-VASc Score for stroke risk. Yet, the debate remains: are these predictions meaningful advancements or little more than modern-day fortune-telling?
The tech behind lifespan prediction
With the advent of artificial intelligence, increasing numbers of tools are emerging on the market. At the same time, biotech companies, such as Altos Labs and BioAge Labs, among others, are engaging such technologies to develop state-of-the-art therapeutics.
AI isn’t a one-stop fix for all challenges. The reality is that it’s a technological tool, albeit with immense potential, that utilizes sophisticated algorithms to get results, much like any other modern technology.
The same approach applies to lifespan prediction technologies built on AI. They engage a suite of technologies, such as:
- Neural networks. These mimic the brain’s architecture, similarly to neurons making connections, and help uncover complex patterns in health and lifestyle data.
- Machine learning algorithms. These analyze high-dimensional datasets, including genomic sequences, wearable device outputs and lifestyle choices, to establish relationships that could impact aging.
- Random forests and decision trees. These can help to identify the critical biomarkers or lifestyle factors that influence the human lifespan.
The power of big data
Perhaps one of the major advantages of AI over traditional solutions hinges on its capabilities in analyzing extensive data sets. These can include:
- Lifestyle metrics. By tabulating diet, exercise, sleep patterns, and other defined factors, it becomes possible to establish their roles in both healthspan and lifespan.
- Medical history. This data can offer insights into chronic illnesses, prior interventions, and potential risk factors for the future.
- Genomic data. Identifying hereditary risk factors and aging-related genes can establish possible risk levels.
- Real-time biometrics. Data gathered from wearable technology, such as heart rate and oxygen levels, can provide insights to overall health longitudinally.
By combining these together, a bigger picture can be established that analyzes the probability of a patient developing a disease or condition or experiencing a negative health outcome within a specific period.
The relationship of risk and lifespan
Predicting the risks of specific diseases is not the same as predicting lifespan, and there may be ethical, moral, and legal concerns. To predict lifespan itself, companies working in this sector often choose to focus on specific metrics: biomarkers of aging.
- Epigenetic clocks. These are used to evaluate DNA methylation patterns in order to estimate biological age. Tools such as GrimAge and DeepAge are already using this technology.
- Blood and wearable biomarkers. These can be used to detect changes in inflammation or metabolism. Tthey offer real-time insights into health trajectories and risk factors.
- Lifestyle biomarkers. These integrate diet, stress, and physical activity in order to allow an AI to suggest actionable interventions that could potentially improve lifespan.
Companies exploring lifespan prediction
Despite the challenges, scientific and human curiosity drive companies to seek answers to those all-important questions. Currently, multiple companies are focused on lifespan and on measuring specific risk factors.
Life2vec: This company offers a transformer-based AI model that analyzes life trajectories, predicting events such as death and health outcomes. It draws upon comprehensive datasets from six million individuals to make its predictions. These include socioeconomic, health, and behavioral data for granular predictions. According to the company’s stats, its accuracy rate is between 70% to 90%; however, it remains unclear how this is calculated.
AI-ECG Risk Estimation (Aire): Aire draws upon electrocardiograms (ECGs) to predict mortality risk. It does so by identifying subtle changes in the heart’s function and potential abnormalities. Estimates from the NHS show its accuracy at 78%.
Impacts to accuracy
As this is a new technology, questions arise relating to its accuracy. For example, Life2vec is a transformer-based system that integrates vast datasets to predict mortality risk with a level of granularity. However, challenges remain:
- Fundamental limitations. Pinpointing death remains an almost unattainable goal. It’s almost impossible to approximate perfect accuracy even with a personalized risk assessment.
- Data bias. AI models are often trained on data sets that lack diversity, which limits their accuracy and can make them biased in certain populations. It’s likely that such models require the same sorts of adjustments as BMI calculators.
- Complexity of aging. This is a developing field, and models will struggle to account for all factors and assign the correct weights to them. In addition, evolving factors, such as emerging illnesses, pandemics, and accidents, will always play a role in lifespan.
Ethics
Like the practicalities of proving accuracy and efficacy, utilizing predictive AI technology for human health has a vast range of legal and ethical implications.
- Data privacy. Health data is confidential and subject to a variety of laws, depending on jurisdiction. Misuse or incorrect use could lead to situations that result in updates to healthcare laws and data protection legislation.
- Ownership and consent. Failure to get informed consent could lead to issues with the ownership of data behind AI predictions. This issue has already arisen with companies such as 23andMe, which have faced criticism for sharing genetic data with third parties.
- Bias and inequality. AI is designed by human programmers that might miss biased data in datasets, which could lead to inaccuracies among some populations and possible legal implications.
- Psychological impact. Just as knowing a risk factor could have health implications, so too could knowing one’s predicted lifespan. This may cause additional unwanted health outcomes, such as anxiety, depression, and orthorexia.
World Health Organization (WHO) has been vocal about such issues and calls for transparent algorithms and ethical frameworks to govern the use of AI tools in health.
The future of lifespan prediction
Advancements are on the horizon. Moving forward, the next generation of tools for enhanced biomarker analysis could seek to integrate more complex and accurate data from epigenetic clocks, wearable devices, and molecular studies. This would allow them to deliver highly personalized lifespan predictions, even if accuracy remains a point of contention.
In addition to this, wearable technology, such as watches or rings, could enable real-time updates, dynamically adjusting predictions based on daily health behaviors. This could foster a nudge-style approach to health management.
Healthspan prediction has the potential to seamlessly integrate into everyday routines, especially for consumer technology and interactions with medical practitioners. This could enable doctors to tailor healthcare treatments to include preventive care and interventions based on an individual’s projected lifespan and biomarkers.
Of course, there are much broader implications to implementing these technologies in day-to-day life, and they extend beyond healthcare to such things as societal issues, personal finance, and the relationship of work to life, which are also affected by enhanced lifespans.
Do we need to know the future?
AI’s potential to analyze data and predict outcomes is something never seen before. As the world, including healthcare, learns how to adapt to this, this knowledge should always be taken with a pinch of salt. Validating predictive tools to the level where they can be reliably used necessitates rigorous testing for accuracy and consideration of how they should be used.
