Briar Smith is CTO and co-founder of Attunement.
The evolution of AI in healthcare: adoption, hurdles, and potential industry changes
As co-founder and CTO of YC healthcare technology company Attunement, I have experienced first-hand the excitement and challenges of implementing AI within the healthcare system. AI is already having a major impact in automating tedious tasks, but its under-explored but huge potential for performing reliable diagnosis and treatment poses major challenges in implementation and model reliability. will face. While the jury is still out on the extent to which healthcare providers will be supported, or even replaced by AI in the future, more affordable and higher quality care will be made possible over time. is likely to come true.
Current state of generative AI in healthcare
Generative AI is emerging as a promising solution to America's broken healthcare system. Its potential uses include everything from assisting clinicians with diagnosis to creating comprehensive treatment plans and even implementing therapeutic interventions. Large-scale language models (LLMs) have demonstrated the ability to perform in parallel with human performance in these categories, providing precise instructions and data.
Under these circumstances, they can outperform even medical experts at things like diagnosing obsessive-compulsive disorder (OCD). Attunement's experiments with LLM for diagnosing fringe cases show that while even the best models and clinicians have difficulty, the number of cases that can be accurately diagnosed rapidly increases with each advancement. It became clear that there was.
Navigating the complex medical environment
However, the path to implementing these technologies is complex and requires overcoming many regulatory and logistical hurdles. Products must fall into the category of valid insurance reimbursement codes and must secure approval from insurance companies, as well as comply with strict FDA regulations, such as demonstrating sufficient clinical efficacy, and are subject to new technology. may take several years. Proving clinical efficacy is also difficult due to model illusions and inaccuracies, making them unreliable compared to health care providers with appropriate medical training.
The current healthcare system poses additional challenges due to misaligned incentive structures between healthcare providers and insurance companies. Providers are paid on a per-service basis (i.e., for more treatment rather than less). Therefore, better results are not encouraged.
Current application and implementation
Because of these constraints, the primary role of AI in healthcare has been focused on back-office automation and infrastructure improvements. This includes streamlining reception forms, writing meeting notes, and providing comprehensive summary services. Companies like DeepScribe have successfully brought AI to transcription services, and emerging platforms like Thoughtful AI are revolutionizing back-office automation by enhancing bill processing and claims management using AI agents. is bringing about. Agents have the ability to autonomously perform actions and perform tasks similar to humans, such as opening applications or copying and pasting.
Combining language models and agents creates artificial general intelligence (AGI) that can do almost anything a human can do. Many experts now predict that AGI could become a reality as early as 2028, much faster than previously predicted due to rapid advances in technology. AGI, trained on appropriate medical datasets, can exceed the abilities of clinicians in diagnosing, planning treatment, and even managing treatment.
Evolution of AI provider-assisted care…towards personal AI care for consumers?
Meanwhile, the healthcare industry is seeing an emerging trend towards “AI-assisted” care models. This approach includes AI working with healthcare providers to provide capabilities such as supporting the diagnostic process, enhancing treatment delivery, providing second opinions on diagnosis and treatment planning, and enhancing clinical decision-making. It will be.
This is likely to be a medium-term use case until the industry catches up and technology eliminates any remaining illusions. This list is likely to expand as capabilities improve, but the jury is still out on whether AI will become an assistive tool for healthcare providers in the long term or replace consumer-managed care. It's not out.
Consumer care theory appears to be becoming increasingly viable and desirable. This is care delivered through an AI system that is customized to an individual's characteristics, problems, and life circumstances. This method raises concerns ranging from privacy to ethics, but the need for a solution is so great that AI systems with a comprehensive understanding of the individual, or at least as much as AI can have. The realization of AI systems with some degree of understanding is probably inevitable.
Privacy and security challenges are likely to be solved through the use of on-device models, preventing sensitive data from being accessed and commercialized by businesses and potential malicious adversaries.
Current limitations and challenges
Several key challenges remain to the widespread adoption of AI in healthcare. Reliability gaps still exist as models struggle to achieve consistent accuracy in diagnosis and treatment. They remain fundamentally constrained by input data from providers and are often unable to capture the nuances of human behavior. Currently, these systems perform best when healthcare providers are trained on the symptoms input into the AI and when they rely on the AI to collect this information directly.
Another potential limitation is the critical role clinicians play in building trusting relationships with patients and collecting behavioral and emotional data that AI systems currently struggle to adequately assess. That's what it means. It's unclear if or when AI will be able to accomplish that.
In summary, the adoption of AI in healthcare faces multiple hurdles, including stringent regulatory requirements for new technologies, provider resistance to workflow disruption, complex reimbursement structures for AI-based solutions, and risk aversion in the healthcare sector. Masu.
Looking to the future
With our current health care system costing us an estimated $1.9 trillion annually, even incremental improvements could yield significant benefits. This potential for positive impact could help overcome the negative connotations often associated with AI in the medical field, particularly in the human interaction aspects of care and reliability.
The industry's progress toward value-based care (VBC) insurance models that prioritize quality over quantity of care represents a promising shift. This transition, coupled with continued improvements in AI reliability and capabilities, suggests a future in which AI plays an increasingly central role in healthcare delivery, despite regulatory and implementation hurdles. The path to success requires careful attention to the reliability and ethics of models, privacy protection, and the balance of technical and human elements in healthcare delivery.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs, and technology executives. Are you eligible?
