Stanford AI100 Report: Five Key Scenarios of 'AI + Healthcare' and Human-AI Collaboration as the Foundation for Large-Scale Adoption

In 2014, Stanford University launched the “AI100” project, also known as the “One Hundred Year Study on Artificial Intelligence.” This initiative brought together leading researchers from various fields to study and predict how artificial intelligence will evolve, as well as its impact on humanity and society.

“Artificial Intelligence + Healthcare” has long been regarded as an emerging field with significant growth potential. In the coming years, AI-based applications are expected to improve the health status and quality of life for millions of people, while enhancing communication between healthcare providers and patients.

VCBeat (WeChat Official Account: vcbeat) has compiled the healthcare-related sections of the AI100 report. The main contents of this article include:

Clinical Setting: AI assistants help automate the consultation process;

Medical Analytics: Management of clinical records and patient data; automated image interpretation;

Medical Robots: Ergonomics + Intelligent Automation;

Digital Healthcare: Leveraging biometric technology to provide personalized recommendations;

Elderly Care: Multiple Innovative Technologies Facilitate Home Living.

The primary application areas of “AI + Healthcare” include clinical decision support, patient monitoring and guidance, surgical assistance, automated devices for patient care, and healthcare system management. Examples include leveraging social media to infer potential health risks, using machine learning to predict diseases, and employing robots to assist in surgeries.

However, how to gain the trust of doctors, nurses, and patients, and how to eliminate policy, regulatory, and commercial barriers are all issues that need to be addressed. As in other fields, data is a key driver. From personal monitoring devices combined with mobile applications, electronic health records (EHR) in clinical settings, to medical robots, researchers continue to innovate and have made significant progress in collecting useful medical data.

However, it has proven difficult for relevant stakeholders to leverage these data to deliver more precise diagnoses and treatments for individual patients and patient populations. Outdated regulations and incentive mechanisms have hindered the development and market launch of related products.

In the vast and complex healthcare system, imperfect human-computer interaction methods and the difficulties and risks associated with technology application pose challenges to the integration of artificial intelligence in the medical field. By reducing or eliminating these barriers, coupled with continuous innovation, the health outcomes of millions of people can be improved.

For decades, the concept of AI-driven clinical assistant has been repeatedly proposed. Although some "AI + Healthcare" pilot projects have achieved success, current healthcare systems remain structurally unprepared to accommodate this technology.

The incentives under the Affordable Care Act accelerated the adoption of electronic health records (EHRs) in clinical practice, but poor implementation has led clinicians to question their effectiveness. Issues include the domination of the EHR market by a small number of companies and the widespread perception that user interfaces fail to meet standards, exemplified by pop-up alerts that physicians often ignore.

Due to the aforementioned issues and regulatory requirements, the vision of leveraging artificial intelligence to analyze EHR data remains largely unrealized.

Over the next 15 years, if artificial intelligence advances rapidly, coupled with sufficient data and appropriate systems, it is expected to improve the work efficiency of clinicians. Currently, following a fixed workflow, patients first provide verbal descriptions of their symptoms, after which physicians correlate these symptoms with the clinical manifestations of known diseases.

If the above processes are automated, physicians can oversee the consultation process, leveraging their experience and intuition to guide data input and evaluate the machine’s intelligent output. Physicians’ “practical” experience will remain crucial. The greatest challenge lies in integrating humanized care with automated reasoning processes.

To achieve optimal outcomes, clinicians must be engaged from the outset to ensure the proper functioning of the system. Currently, a new generation of physicians, well-versed in these technologies, has begun utilizing specialized applications on mobile devices. Meanwhile, the workload of primary care physicians is set to increase substantially.

However, addressing regulatory, legal, and social issues can significantly enhance clinical analytics. This includes developing new learning methodologies, creating structured reasoning frameworks through automated analysis of scientific literature, and building cognitive assistants via free-form dialogue.

Artificial intelligence can analyze millions of patient clinical records, thereby enabling more accurate and personalized diagnosis and treatment. As whole-genome sequencing becomes a routine examination for patients, genotype-phenotype correlation analysis will also become feasible.

For instance, treatment plans can be determined through cohort-like analysis, which involves identifying “similar patients.” Patient stratification is based on data from social platforms as well as traditional and non-traditional medical sources. Each group is managed by a dedicated system comprising healthcare providers along with automated recommendation and monitoring mechanisms. Applying this technology to the clinical records of hundreds of millions of individuals could fundamentally improve healthcare delivery.

Furthermore, artificial intelligence technologies can also deliver personalized medical services—for instance, by automatically capturing personal environmental data through wearable devices to generate tailored analyses and recommendations. Currently, companies such as ShareCare are applying this technology in healthcare settings.

However, achieving rapid innovation still requires overcoming numerous challenges. The FDA has been slow in approving innovative diagnostic software; the Health Insurance Portability and Accountability Act (HIPAA) mandates the protection of patient privacy, thereby creating legal barriers to the use of patient data through artificial intelligence technologies. Approved drugs or products may have unexpected negative impacts; for instance, mobile applications designed to analyze drug interactions may be prohibited from extracting necessary information from patient records.

Overall, the lack of universal privacy protection methods and standards has hindered artificial intelligence research and innovation in the healthcare sector. The FDA’s delayed approval of innovative software is partly due to the difficulty in weighing the costs and benefits of these systems. If regulatory agencies, primarily the FDA, recognize that post-market surveillance can effectively mitigate certain safety risks, they may accelerate the approval of new therapies and interventions.

For decades, automated image interpretation has been a field with immense development potential. Advances in this area have garnered significant attention, such as the interpretation of large volumes of weakly labeled images (e.g., massive photo datasets scraped from the web). Prior to this, medical image interpretation had not achieved comparable progress. This is because most medical imaging modalities (CT, MRI, ultrasound) are inherently digital, with images systematically archived, and large, technologically mature companies (such as Siemens, Philips, and General Electric) specializing in imaging research.

However, obstacles remain that constrain the development of this field. Most hospitals have only digitized their image archives within the past decade. More importantly, addressing medical challenges relies not merely on identifying objects in images, but on making accurate clinical judgments. Such high-stakes decisions are subject to stringent regulatory oversight.

Even with the most advanced technology, radiologists may still need to review images, rendering their diagnostic conclusions less conclusive. Furthermore, healthcare regulations prohibit cross-institutional data sharing. Consequently, only large integrated healthcare systems, such as Kaiser Permanente, are positioned to address these challenges.

Nevertheless, the field of automated/augmented image interpretation continues to evolve rapidly. Over the next 15 years, while fully automated radiology is unlikely to emerge, initial applications for image “triage” or secondary review are expected to enhance the speed and cost-effectiveness of medical imaging.

Integrated with electronic health record (EHR) systems, machine learning technologies can be applied to medical imaging data on a large scale. For instance, several major healthcare systems maintain records for millions of patients, each associated with corresponding radiological data. Furthermore, relevant literature indicates that deep neural networks can be trained to analyze radiological data with high reliability.

Fifteen years ago, medical robots existed only in science fiction. A company named Robodoc (a subsidiary of IBM) developed robotic systems for orthopedic procedures such as hip and knee replacements. However, the company encountered commercial difficulties, ultimately went bankrupt, and its technology was acquired. Recently, however, research and practical applications of medical robots have experienced explosive growth.

In 2000, Intuitive Surgical launched the da Vinci system, a new technology for minimally invasive coronary artery bypass grafting that was later adapted for the treatment of prostate cancer. In 2003, Intuitive Surgical merged with its competitor, Computer Motion.

Currently, the fourth-generation da Vinci system offers 3D visualization (as opposed to 2D laparoscopy) on an ergonomic platform. It is considered the standard tool for laparoscopic surgery, with nearly 750,000 procedures performed annually, providing a new data platform for researching surgical processes.

The system will conduct in-depth learning on how healthcare professionals engage in the care process, providing insights for innovation across various fields, including new instruments, image fusion, and novel biomarkers. Furthermore, the success of this ergonomic platform has spurred development in the field of robotic surgery, notably Verb Surgical, which received investment from Verily (formerly Google Life Sciences) and Ethicon (a medical device company under Johnson & Johnson).

Another area related to robotics is intelligent automation. Approximately 20 years ago, HelpMate developed a robot capable of assisting hospitals in transporting items such as meals and medical records.

Recently, Aethon introduced the TUG robot for material transport. However, few hospitals have invested in this technology to date. Nevertheless, robotics has proven practical and cost-effective in other service industries, such as hospitality and warehousing, exemplified by the innovative applications of Amazon Robotics (formerly Kiva).

In the future, various medical tasks will become simpler due to robotics, but full automation will not be achieved. For example, robots can deliver items to hospital wards, but human staff are still required to place them in their final locations. With the assistance of walking aids, patients can move more easily along corridors (though it remains challenging for post-surgical rehabilitation patients or elderly individuals to navigate hallways crowded with equipment and other patients). These applications demonstrate that many systems and technologies will involve close interaction between humans and machines.

The advancement of automation will lead to a new understanding of medical processes. Based on past applications, robotics has not been a data-driven or data-oriented discipline. However, this is changing as (semi-)automation is increasingly applied in the healthcare sector.

With the launch of patient care platforms, quantitative and predictive analytics are built upon the data these platforms provide. This data is used to assess quality, identify errors, and improve performance. In short, these platforms link processes to outcomes, making a true healthcare “closed loop” a reality.

To date, evidence-based medical analytics have relied on traditional healthcare data, primarily electronic health records (EHRs). In clinical settings, artificial intelligence can offer novel solutions. For instance, TeleLanguage enables clinicians to communicate with and provide treatment to multiple patients simultaneously with the assistance of an AI companion. Lifegraph extracts behavioral patterns from data collected via patients’ smartphones and issues alerts; psychiatrists in Israel have already leveraged such products to detect early symptoms in patients.

With the advancement of mobile computing, platforms and applications related to “biometrics” will continue to emerge. Currently, thousands of mobile applications are available that provide information, correct behaviors, or identify “similar patients.” These applications, combined with specialized activity-tracking devices (such as Fitbit) and their connectivity with home environment and health monitoring devices, will drive innovation in the healthcare sector.

By integrating social and medical data, some applications can analyze, learn from, and make predictions based on the captured information. Although their predictive capabilities are relatively basic, this convergence of data and functionality may spur the development of innovative products. For instance, a certain fitness app not only proposes workout plans but also recommends optimal exercise times and provides guidance to help users adhere to their regimens.

Over the next 15 years, the elderly population in the United States will grow by more than 50%. The U.S. Bureau of Labor Statistics predicts that the number of home health aides will increase by 38% over the next decade. Innovative applications in the field of elderly care include interactive and communication devices, home health monitoring equipment, and mobility aids (such as walkers). However, progress in this sector has been relatively slow over the past 15 years.

With the emergence of various innovative applications, older adults’ acceptance of technology is also changing. Currently, 70-year-olds may have experienced personalized information technology for the first time only in middle age or later, whereas 50-year-olds demonstrate greater acceptance of new technologies. Therefore, artificial intelligence holds significant market potential for improving the physical, emotional, social, and spiritual well-being of older adults.

Quality of Life and Independence

• Automated vehicles help older adults live more independently and broaden their social horizons.

• Information sharing will facilitate communication among family members, and predictive analytics may be leveraged to promote positive family behaviors, such as reminding individuals to “call home.”

• Smart home devices will assist with daily activities, such as cooking, when needed. If robots’ operational capabilities improve, they can also help with dressing and personal hygiene.

Health

• Mobile apps that monitor activity, combined with social platforms, will provide recommendations for maintaining physical and mental well-being.

• Detect changes in mood or behavior and alert caregivers through home health monitoring and the provision of health information.

• Personalized Health Management

Treatment Methods and Equipment

• Hearing aids and visual assistive devices will mitigate the negative impacts of hearing and vision loss, provide a safer environment for older adults, and enhance their social connections.

• Personalized rehabilitation and home-based treatment will reduce the need for hospitalization or care facilities.

• Assistive devices (such as smart walkers and wheelchairs) will expand the mobility range of frail individuals.

Researchers anticipate that low-cost sensing technologies will advance rapidly, providing convenience for elderly individuals living at home. In addition to sensing technology, the entire intelligent system will encompass multiple domains, such as natural language processing, reasoning, learning, perception, and robotics.

[References]

https://ai100.stanford.edu/sites/default/files/ai_100_report_0831fnl.pdf