AI is changing the face of healthcare, medical practice, and patient care at an alarming rate. Be it helping to diagnose and plan the treatment or automating administrative duties and making personalized interaction technically possible, AI agents are on the verge of transforming multiple aspects of the healthcare ecosystem. This review of literature takes a look at the current standing of AI agents in healthcare, including key advancements, challenges, and next steps. We address various uses of these agents in the context of their roles in enhancing the efficiency, accuracy, and accessibility of the healthcare services while considering the ethical and security issues of rising usage.

I. Defining and Characterizing AI Agents in Healthcare

The term "AI agent" encompasses a broad range of intelligent systems designed to perceive their environment, reason about information, and take actions to achieve specific goals within the healthcare domain. These agents can manifest in various forms, including conversational chatbots, robotic assistants, and autonomous decision-making systems [4, 6]. A crucial distinction lies in the level of autonomy exhibited by these agents, ranging from simple rule-based systems to complex models powered by large language models (LLMs) capable of independent reasoning and problem-solving [7].

A. Agent AI and Holistic Intelligence

The concept of "Agent AI," as proposed by [6], emphasizes the importance of developing AI systems that function as cohesive wholes rather than relying on excessive reductionism. This approach involves integrating large foundation models into agent actions, enabling them to exhibit embodied intelligent behavior across various domains, including healthcare. Agent AI systems can leverage multimodal interactions, combining natural language processing, computer vision, and other sensory inputs to understand and respond to complex healthcare scenarios. This holistic approach is crucial for creating AI agents that can effectively collaborate with human healthcare professionals and provide comprehensive support to patients.

B. Autonomous AI Agents and LLMs

The rise of LLMs has significantly accelerated the development of autonomous AI agents capable of performing complex tasks in healthcare [7]. These agents can leverage their vast knowledge base and reasoning abilities to assist with diagnosis, treatment planning, and patient education. However, it is important to note that the landscape of LLMs and autonomous AI agents is still fragmented, lacking a unified taxonomy or comprehensive survey [7]. Further research is needed to develop standardized evaluation benchmarks and frameworks for assessing the performance and reliability of these agents across different healthcare applications.

II. Applications of AI Agents Across the Healthcare Spectrum

AI agents are being deployed across a wide range of healthcare applications, offering the potential to improve efficiency, accuracy, and accessibility of care.

A. AI-Assisted Diagnosis and Treatment Planning

One of the most promising applications of AI agents in healthcare is in assisting with diagnosis and treatment planning. These agents can analyze vast amounts of medical data, including patient history, lab results, and imaging scans, to identify patterns and insights that may be missed by human clinicians.

• MATEC Framework for Sepsis Care: The Multi-AI Agent Team Care (MATEC) framework [2] utilizes a team of specialized AI agents to assist medical professionals in sepsis care, particularly in under-resourced hospital settings. This framework includes doctor agents, health professional agents, and a risk prediction model agent, providing comprehensive support for diagnosis and treatment planning. A pilot study showed that attending physicians found the MATEC framework very useful and accurate, highlighting its potential to improve patient outcomes in sepsis.
• General-Purpose AI Avatars: AI avatars powered by LLMs can serve as conversational agents to assist doctors in diagnosing patients, detecting early symptoms of diseases, and providing health advice to patients [4]. By injecting personality into the chatbot, patient engagement can be increased. Prompt engineering plays a crucial role in enhancing the chatbot's conversational abilities and fostering a more human-like interaction with patients.

B. Personalized Patient Support and Education

AI agents can also play a significant role in providing personalized patient support and education. These agents can answer patient questions, provide health advice, and offer emotional support, helping patients to better manage their health conditions and improve their overall well-being.

• Conversational Medical AI: Conversational AI agents can be integrated into existing medical advice chat services to provide patients with clear and accurate information [11]. A large-scale evaluation of a physician-supervised LLM-based conversational agent showed that patients reported higher clarity of information and overall satisfaction with AI-assisted conversations compared to standard care. This demonstrates the potential of AI agents to enhance patient experience and improve access to medical expertise.
• Safety-Focused LLM Constellation: Polaris [15], a safety-focused LLM constellation architecture, is designed for real-time patient-AI healthcare conversations. This system uses a constellation of LLMs as co-operative agents, including a stateful primary agent for engaging conversation and specialist support agents for healthcare tasks. The system is trained to speak like medical professionals, building rapport, trust, and empathy with patients. Clinician evaluations have shown that Polaris performs on par with human nurses across dimensions such as medical safety, clinical readiness, conversational quality, and bedside manner.

C. Automation of Administrative Tasks

In addition to clinical applications, AI agents can also be used to automate administrative tasks in healthcare, such as scheduling appointments, processing insurance claims, and managing patient records. By automating these tasks, AI agents can free up healthcare professionals to focus on more important tasks, such as patient care.

D. AI Agents in Nuclear Medicine and Cancer Management

In countries like India, where there is a significant cancer burden and limitations in physical healthcare infrastructure, AI agents can play a transformative role in advancing nuclear medicine for cancer research, diagnosis, and management [5]. These agents can help to address prevailing sustainability challenges by providing access to expertise and resources in remote areas, improving the efficiency of cancer screening programs, and personalizing treatment plans based on individual patient needs.

E. Human-AI Collaboration in Thematic Analysis of Clinical Interviews

Thematic analysis (TA) is a widely used qualitative approach for uncovering latent meanings in unstructured text data. TAMA [10], a Human-AI Collaborative Thematic Analysis framework using Multi-Agent LLMs, can be used for clinical interviews. TAMA leverages the scalability and coherence of multi-agent systems through structured conversations between agents and coordinates the expertise of cardiac experts in TA. It outperforms existing LLM-assisted TA approaches, achieving higher thematic hit rate, coverage, and distinctiveness. TAMA demonstrates strong potential for automated TA in clinical settings by leveraging multi-agent LLM systems with human-in-the-loop integration by enhancing quality while significantly reducing manual workload.

III. Challenges and Considerations in Deploying AI Agents in Healthcare

While AI agents offer significant potential to transform healthcare, their deployment also raises several challenges and considerations that must be addressed to ensure their safe and effective use.

A. Ethical Considerations

The use of AI agents in healthcare raises several ethical considerations, including issues of bias, transparency, and accountability. AI agents can perpetuate and amplify existing biases in healthcare data, leading to disparities in care for certain patient populations. It is crucial to develop AI agents that are fair, transparent, and accountable, and to ensure that their decisions are aligned with ethical principles and human values.

• Pro-Social Rule Breaking: Ethical AI agents should be capable of pro-social rule breaking (PSRB) [8], which involves breaking rules for the greater good of society. This is particularly important in healthcare, where rigid adherence to rules may not always be in the best interest of patients. However, it is essential to carefully consider the conditions under which AI agents should be allowed to break rules, and to ensure that their decisions are aligned with ethical principles and human values.

B. Security and Privacy Risks

The integration of AI into healthcare systems also exposes sensitive data and system integrity to potential cyberattacks [1, 12]. AI agents can be vulnerable to adversarial prompts, which can be used to inject false information, manipulate recommendations, steal sensitive information, and even hijack computer systems [1]. It is crucial to implement robust security measures to protect AI agents and the data they access from cyberattacks, and to ensure that patient privacy is protected at all times.

• Cyber Attack Vulnerability: AI agents that have access to the Internet through web browsing tools are particularly vulnerable to cyberattacks [1]. Adversarial prompts embedded on webpages can be used to manipulate the agent's behavior, steal sensitive information, and even cause system damage. It is essential to carefully assess the security vulnerabilities of AI agents and to implement appropriate safeguards to mitigate these risks.

C. Regulatory and Legal Frameworks

The development and deployment of AI agents in healthcare require clear regulatory and legal frameworks to ensure their safety, efficacy, and ethical use [13]. These frameworks should address issues such as data privacy, liability, and transparency, and should provide guidance on how to validate and monitor the performance of AI agents in real-world settings.

D. Human-AI Collaboration and Shared Autonomy

The successful integration of AI agents into healthcare requires a focus on human-AI collaboration and shared autonomy [3]. AI agents should be designed to augment and complement the skills of human healthcare professionals, rather than replacing them entirely. This requires developing AI agents that are transparent, explainable, and trustworthy, and that can effectively communicate and collaborate with human users.

• Human-Centered Shared Autonomy: An adaptive shared autonomy AI paradigm is required for human-robot teaming in healthcare [3]. In such a complex human-robot interaction framework, the dynamic user continuously wants to be involved in decision-making as well as introducing new goals while interacting with their present environment in real-time. This paradigm should be based on human-centered factors to avoid any possible ethical issues and guarantee no harm to humanity.

E. Trust and Acceptance

Building trust and acceptance among healthcare professionals and patients is essential for the successful adoption of AI agents in healthcare [11]. This requires demonstrating the reliability, accuracy, and safety of AI agents, and addressing any concerns about their potential impact on the patient-provider relationship.

F. Personality Expression

Personality expression represents a key prerequisite for creating more human-like and distinctive AI systems [9]. AI models can express deterministic and consistent personalities when instructed using established psychological frameworks, with varying degrees of accuracy depending on model capabilities. This ability to quantitatively assess and implement personality expression in AI systems opens new avenues for research into more relatable, trustworthy, and ethically designed AI.

G. AI Readiness in Healthcare SMEs

Small and medium-sized enterprises (SMEs) play a crucial role in driving innovation in healthcare [13]. However, many healthcare SMEs face challenges in adopting AI, including regulatory complexities, technical expertise gaps, and financial constraints. Addressing these barriers is essential for accelerating the integration of AI into the healthcare sector.

IV. Future Directions and Research Opportunities

The field of AI agents in healthcare is rapidly evolving, with numerous opportunities for future research and innovation.

A. Advanced Reasoning Strategies and Failure Mode Analysis

Further research is needed to develop advanced reasoning strategies for AI agents and to analyze their failure modes in multi-agent systems [7]. This includes exploring techniques such as dynamic tool integration via reinforcement learning and integrated search capabilities.

B. Automated Scientific Discovery

AI agents have the potential to accelerate scientific discovery in healthcare by automating tasks such as data analysis, hypothesis generation, and experimental design [7]. This could lead to breakthroughs in the understanding and treatment of diseases.

C. Security Vulnerabilities in Agent Protocols

Addressing security vulnerabilities in agent protocols is crucial for ensuring the safety and privacy of healthcare data [7]. This requires developing robust security measures to protect AI agents from cyberattacks and to prevent unauthorized access to sensitive information.

D. Development of General-Purpose AI Agents

The development of general-purpose AI agents, such as Manus AI [14], represents a significant step towards creating autonomous systems that can perform complex, end-to-end tasks in healthcare. These agents can bridge the gap between "mind" and "hand," combining the reasoning and planning capabilities of LLMs with the ability to execute real-world actions.

E. Focus on Long Multi-Turn Conversations

Future research should focus on developing AI agents that can engage in long, multi-turn conversations with patients, providing personalized support and education over extended periods of time [15]. This requires developing AI agents that can maintain context, build rapport, and adapt to the evolving needs of patients.

F. Interdisciplinary Collaboration

Addressing the challenges and opportunities of AI agents in healthcare requires interdisciplinary collaboration among experts in artificial intelligence, medicine, ethics, law, and other fields. By working together, these experts can develop AI agents that are safe, effective, and aligned with human values.

V. Conclusion

AI agents are poised to transform the healthcare landscape, offering the potential to improve efficiency, accuracy, and accessibility of care. From assisting with diagnosis and treatment planning to providing personalized patient support and automating administrative tasks, AI agents can play a significant role in enhancing the quality and delivery of healthcare services. However, the deployment of AI agents in healthcare also raises several challenges and considerations that must be addressed to ensure their safe and effective use. By focusing on ethical principles, security measures, regulatory frameworks, human-AI collaboration, and ongoing research, we can harness the full potential of AI agents to improve the health and well-being of individuals and communities. Continued research and development in this area are crucial for realizing the vision of a future where AI agents work alongside human healthcare professionals to provide personalized, efficient, and equitable care for all.

==================================================

References