Presenters: Antonio Regalado (Senior Editor for Biomed, MIT Tech Review), Othman Laraki (Co-Founder & CEO, Color Genomics), Katherine Chou (Head of Product, AI in Health, Google), Derek Zu (Strategy & Products, Baidu USA), Qirong Ho (Chief Technology Officer, Petuum), Nikjil Jain (Founder & CEO, ObEN) To read the full report click here for the digital edition. The American healthcare industry is paradoxical. Though approximately 20 percent of GDP is spent on healthcare, the quality of care has not improved despite spending increases and technological advancements. As healthcare spending doubles roughly every decade with individuals’ longevity remaining constant, Antonio Regalado, the panel’s moderator, asserted that the industry is experiencing Eroom’s law. Because of healthcare’s large impact both in economic and human wellness terms, artificial intelligence has the potential to overcome healthcare’s paradox to better the healthcare industry and mankind. The “AI in Healthcare” panel focused on ways to apply AI to improve the quality of healthcare while lowering its costs. Recognizing that healthcare has become more monopolistic, Othman Laraki keyed in on the structural challenges that the U.S. healthcare industry confronts. To address a growing need for effective and efficient care in the face of structural problems, the speakers underscored AI and its role in unlocking new medicinal innovations and treatment options. The panel assembled speakers involved with AI applications into healthcare. As engineers, the speakers took a tech approach to discussing the healthcare industry—each applying their expertise and company involvement to discuss how AI will supplement human intellect to overcome stagnation in quality of care.
“We have this paradox in healthcare that despite spending more, we’re not really getting that much bang for our buck.” -Antonio RegaladoKEY TAKEAWAYS Healthcare is an entrenched industry, but AI can disrupt it. Healthcare is ripe for disruption. As an industry that has not experienced significant growth, healthcare can benefit from information technologies and AI to capitalize on government spending and technological advancements. By identifying causes of entrenchment and anticipating ways that AI will disrupt the current system, healthcare can benefit from technology and improve the industry as a whole. Healthcare is entrenched because of structural challenges. Healthcare’s value chain has monopolistic layers. The small number of insurance companies with high market caps, Laraki argues, restrict market forces from letting the best healthcare product win in terms of beating out competitors and lowering product costs within the confines of supply and demand. Other structural challenges relate to medical records and accompanying ownership issues. Many Americans, according to Nikjil Jain, do not have access to their own medical records and do not know who does have ownership. These ownership and monopoly challenges have restricted the healthcare industry from improving quality of care and lowering costs. AI can cause regulatory and payment disruption. When people think of AI disruption, they associate it with job displacement. Derek Zu, however, argues that applying more artificial intelligence into healthcare will disrupt the industry’s regulatory and pay structures by transitioning into a value-based structure, rather than causing unemployment. Value-based care is more feasible today due to machine learning’s ability to take in large amounts of surrounding data on an individual basis. This technology then predicts an individual’s type of health risk, intervenes early on and prevents some of those risks for better patient outcomes and lower overall costs. Artificial intelligence needs to be trusted. People must trust artificial intelligence and its capabilities in order for AI to positively impact the healthcare industry. If the technology is distrusted, it cannot become operational and its abilities will remain unutilized. In order to foster trust, technology needs to perform well and meet the desired outcomes. Humans must also change their mindsets to recognize that AI is a tool rather than a replacement for humans and can improve healthcare and the overall human experience. Katherine Chou predicts that, through human-machine complementarity, AI will become more trusted if applied to preventive and incremental medicine under a value-based system. Change will take time. The term “disruption” implies that artificial intelligence will sharply alter the status quo. Qirong Ho refutes that notion, instead predicting that AI will gradually change the current healthcare system. Hospital regulations and existing processes prevent overnight change from occurring and require incremental change to alter healthcare’s current proceedings.
“AI can augment what the clinicians can do in that space, since they actually no longer have to be necessarily focused just on the specifics of the images—now they can be thinking of the patient holistically.” - Katherine ChouAI can improve clinical practices. Artificial intelligence increases efficiency and solves problems in a faster, more effective manner. This enables humans to leave data processing, detailed-oriented and habitual work to technology in order to specialize in more nuanced, interaction-based and holistic work that humans do best. Additionally, technological advancements can expand medical capabilities to sophisticate procedures, improving the lives of both physicians and patients. Artificial intelligence can streamline back office hospital processes. Hospitals are consumed by processes rather rather than the provision of care. Regalado notes that hospital staffs have grown, but the amount of doctor jobs have not—rather, jobs go to administrators, clerks and managers of hospital processes instead of towards people delivering the actual medical care. Ho linked this to hospitals’ regulatory report requirements and their backlogging since the current electronic databases are inadequate. Additional slow hospital procedures relate to electronic record systems and hospital interactions with insurance companies. Artificial intelligence can step in to structure unstructured data and interface with electronic record systems and insurance companies to streamline hospital processes. Not only will this improve the system, but it will provide staff with more patient interaction time. Technology can advance and improve medical procedures. Artificial intelligence is being used to make medical procedures less invasive and more effective. AI has helped realize diabetic retinopathy screening and liquid biopsy procedures, allowing doctors to do more groundbreaking surgeries while providing more options to patients. Zu and Chou highlight how AI can classify images, specifically using algorithms to identify disease areas and tumor regions in pathology slides. This gives doctors more information and greater proficiency to assess and treat patients. Artificial intelligence advantages and its use of genomic data, deep learning applications and next generation sequencing play a big role in preventative care and personalized medicine. AI use enables doctors to spend more time with patients. Artificial intelligence intends for doctors to spend less time with systems and more time with people by decreasing interface time with electronic record systems and insurance companies. Ultimately, artificial intelligence is about improving doctors’ work lives, whether by improving procedural capabilities, supplementing their own knowledge with AI capabilities or transitioning their time away from systems and towards patient interaction. Healthcare needs to become more available. In order for healthcare to become better, it has to become more available to individuals. As AI intends to simultaneously lower healthcare costs while improving its quality, patient wellness and health serve as primary metrics in determining progress. By using technologies to lengthen individuals’ life spans and make doctor care less expensive, healthcare will benefit more people. Availability implies more accessibility and affordability. Availability does not simply equate to accessibility. If healthcare is accessible but overly expensive, hospitals will deter individuals from receiving medical care. Thus, affordability is necessary as well. Recognizing a tradeoff between quality, cost and access within the healthcare system, Chou argues that AI can overcome this dilemma. Technologies should not only drive down costs, but also should additionally improve the quality of care and its availability since AI streamlines procedures and allows doctors to make an efficient use of their time. When medical staff is unavailable, AI should be used. Many locations in the developing world do not have access to the specialized medical staff found in the United States. However, in remote locations, artificial intelligence can serve as a potential substitute to certain types of doctors such as radiologists, pathologists, dermatologists and ophthalmologists since their line of work has an aspect of imaging classification and recognition—things that AI does well. Google, for instance, applied AI in India (where there is a lack of care access) when doing a diabetic retinopathy screening. AI’s ultimate goal is to make technology and care more available. Technology has a trickle-down effect. As groundbreaking technology is used initially in larger enterprises, overtime, its technology reaches households. The goal is to apply AI into healthcare so that its technologies pervade homes; providing individuals with availability to AI and more practical medical care. This is most important in cancer care as early prevention is key and cancer patients’ frequent hospitals to receive needed care. AI will change doctor roles in the future. The future of medicine and its interaction with artificial intelligence will alter traditional physician roles. As technology has improved, it can process abundant raw and unstructured data and use this information to detect risks and ailments early on, while moving forward to prevent such risks. That is not to say, however, that artificial intelligence can and will replace human doctors. Rather, technologies and humans will specialize in their separate comparative advantages to improve the healthcare system. AI should cover responsibilities that technology is better at than doctors. Physician responsibilities need to be redefined in the artificial intelligence era. This “peeling off” and “decomposition” of doctor tasks will allow technology to assume roles that are inherent to doctors and that technology can better perform. AI is best at imaging classifications related to medical imaging and processing granular and continuous data. Doctors, meanwhile, specialize in physician-patient interactions, interpreting models and applying processed data to analyze patients holistically. AI augments human intellect. AI can solve problems that the human mind cannot. However, it is not a competition between physicians and technology: both are needed in concert to lower healthcare costs and improve quality of care. Rather than being a rival, AI is an asset for humans to use in overcoming human shortcomings and pushing past barriers mainly related to processing continuous and unstructured data that previously obstructed human advancement. Human-technology interactions have different stages. Similar to how there are five levels to autonomous vehicles, Jain argues there are different levels of AI-care interaction. The care levels differ from AV levels since AI is less involved in the higher levels—the opposite of AV levels. The first level of care is a call center, in which artificial intelligence can completely commandeer humans in that role. As human interaction is not vital in that space, AI can help overcome bias, with Google’s “Duplex” serving as an example. The second stage relates to triage nurses that assign patients to the next level of care. Artificial intelligence can also play a direct role in this level, but with less interaction than the previous stage. Upon reaching the third level of senior nurses, care is more clinical and more analysis is required. At the third stage, there is a transition and less AI is applied. As healthcare continues to determine the level of AI involvement in hospitals, recognizing these levels of care will enable doctors and hospital staff to benefit most from AI involvement. Existing technologies should be repurposed into healthcare applications. AI is not confined to one industry or platform. As companies like Google have applied their technologies into healthcare, quality of care has improved without needing to reinvent the wheel and drive costs in developing new technology. Google has applied its capabilities to medical procedures. Specifically, when seeking to apply AI to improve fundus imaging, Google used the same convolutional neural network technology as a Google image search for ophthalmological procedures. Additionally, Google Translate technology and its accompanying sequencing models are used to predict a patient’s future health risks. By repurposing Google technology into healthcare, the company has shown that AI can effectively and practically be implemented into the industry. ObEN.AI repurposed its technologies to assist with mental illness. ObEN, a company that makes virtual copies of humans that “look like you, talk with you, act like you” transitioned into the healthcare industry by applying its “Avatar PAI” (Personal Artificial Intelligence) technology in providing care work. Recognizing that the older generation struggles to take their pills on time (or at all) without pressure from their grandchildren or primary doctors, the company set out to create avatars of busy primary care doctors and grandchildren to help overcome mental health challenges that seniors face. As there is a strong correlation to seniors not taking medication and increases in healthcare costs over time, the company’s application of AI focuses on lowering costs while improving care. AI needs guardrails to overcome liability issues. In the event that AI inaccurately diagnoses a patient, it is important that measures already exist that recognize the technology will not always be perfect. As similar liability dilemmas confront automated vehicles, we must consider how it will overcome such challenges, constructing guardrails to ensure that technology can endure setbacks. Liability guardrails begin at the product level. When transitioning to greater AI involvement in healthcare, doctors need to play a greater role in reviewing and signing off on AI results. Also, it is important that the technology has a built-in sorter that can identify false negatives so that doctors do not have to review all of the machine’s findings, defeating the point of using technology in the first place. Thus, guardrails should ensure that technology does not derail and patients are not worse off in the event that AI makes a mistake. Guardrails follow a concentric circle path from the model itself to society at large. The first guardrail, Chou notes, focuses on the model itself in ensuring that its technology is accurate and has been signed off by experts. The next level relates to the individual physician and making sure that they are qualified in interpreting the data provided. Moving outwards, the third circle corresponds to the clinical environment and whether the hospital has properly defined how it intends to use AI’s technology. The final guard rail adopts a historical and societal approach to assess the effect of AI application to society as a whole.
The views presented in this article are the author’s own and do not necessarily represent the views of any other organization.