My vote for Time’s Person of the Year is Artificial Intelligence (AI). I think AI is the most talked about and hyped (overhyped?) development of 2024, already transforming operations across numerous sectors, from manufacturing to financial services. In the health sector, AI has ushered in groundbreaking advancements in several areas, including psychotherapy, substituting for therapists and also posing ominous portents for physicians. AI systems that learn independently and autonomously – as opposed to iteratively – are the ones to keep an eye on.
Iterative learning and autonomous learning differ in terms of process and decision-making scope. Iterative learning involves a step-by-step process where an AI model is trained through repeated cycles or iterations. Each cycle refines the model based on errors or feedback from the previous iteration. This type of learning often involves human supervision, with periodic interventions to adjust hyperparameters, refine datasets, or evaluate outcomes. In a health care setting, iterative AI might be used in diagnostic tools that analyze imaging data, where radiologists provide feedback on the AI’s initial assessments, allowing the system to learn and improve its diagnostic accuracy.
In contrast, autonomous learning refers to an AI system’s ability to independently acquire knowledge or adapt its behavior in real-time without explicit instructions or frequent human input. These systems are self-guided, seeking and utilizing data or experiences on their own to enhance performance. They are adaptable to changing environments and can learn new tasks or optimize their performance in open-ended scenarios. Autonomous AI in health care could potentially manage routine tasks such as patient monitoring or medication management, making decisions based on clinical signs and symptoms. Robotic surgery systems can make real-time adjustments during procedures, utilizing AI to enhance precision and efficiency.
Both approaches are valuable and are often combined in practice. For instance, iterative learning might pre-train a model that subsequently engages in autonomous learning during deployment, fine-tuning its abilities based on real-world data. This combination allows for both structured development and dynamic adaptability.
A compelling example where both iterative and autonomous AI approaches are combined in health care is in the development and deployment of personalized medicine platforms, particularly in oncology, where iterative AI is initially used to train models on large datasets comprising genetic information, treatment outcomes, and patient histories, and autonomous AI analyzes new patient data, recommending personalized treatment plans based on the insights derived from its extensive pre-training.
If you watch a lot of science fiction, like I do, then perhaps the fear of autonomous AI systems “taking over” and eliminating human functions – or humans themselves – feels both familiar and unsettling. It is a topic fueled not only by science fiction and fantasy but also by philosophical debate. Former Google chairman and CEO Eric Schmidt’s new book Genesis: Artificial Intelligence, Hope, and the Human Spirit has been described as “[a] profound exploration of how we can protect human dignity and values in an era of autonomous machines.” I’m worried about protecting our species – let alone our “spirit.”
Theoretically, several factors currently prevent doomsday scenarios. These can be divided into technical limitations, ethical safeguards, social structures, and systemic dependencies.
Technical limitations
Autonomous AI systems are highly specialized and lack general intelligence. While they excel in narrow tasks, they do not possess the creative, emotional, or abstract thinking capabilities required for broad, human-like cognition. Current AI systems operate within strict parameters, and their decision-making is bound by the data and algorithms they are trained on. Even advanced systems that can adapt or learn in real-time are limited in scope and do not have the capacity for complex, independent planning or motivation—essential components for “taking over.”
Ethical safeguards
AI development is guided by ethical principles, regulations, and oversight designed to prevent harm. Developers and governments are implementing frameworks such as AI ethics guidelines, explainability requirements, and safety measures to ensure AI systems act in accordance with human values. Examples include the European Union’s AI Act and AI ethical principles recommended by the U.S. Department of Defense and organizations like OpenAI (there are 200 or more guidelines and recommendations for AI governance worldwide). These guardrails aim to prevent misuse or unintended consequences.
Social structures
AI systems are tools created, owned, and operated by humans or organizations. They lack autonomy in the sense of independence from these structures. Governments, institutions, and corporations establish rules and maintain oversight over how AI is deployed, ensuring that it serves specific purposes and remains under human control. Social and political systems also resist relinquishing significant power to autonomous systems due to economic, ethical, and existential concerns.
Systemic dependencies
Autonomous AI systems depend on infrastructure, energy, and maintenance, all of which remain under human control. They cannot sustain themselves without these resources. Furthermore, AI systems often require human input or oversight for ongoing relevance and adaptation, particularly in unpredictable environments.
Preventing harm
The idea of AI systems intentionally “eliminating” humans assumes a level of sentience, malice, and motive that current AI lacks. AI systems do not have desires, self-preservation instincts, or moral reasoning. Any harm caused by AI arises from flawed design, inadequate safeguards, or malicious use by humans – not from the systems themselves. Efforts to mitigate such risks focus on robust design, testing, and mandating accountability in AI deployment.
Future considerations
As AI evolves, ensuring its alignment with human values and control becomes increasingly critical. This includes developing general AI, also known as Artificial General Intelligence (AGI), a type of AI that possesses the ability to understand, learn, and apply knowledge across a wide range of tasks at a level comparable to human intelligence. The development of AGI is a major goal in the field of AI research, but it remains largely theoretical at this point, as current AI systems are specialized and lack the generalization capabilities of human cognition.
Public discourse, interdisciplinary collaboration, and regulatory oversight will play pivotal roles in preventing scenarios where AI could displace humans in destructive ways. While theoretical risks exist, the current state of AI lacks the capacity or motive for such dramatic outcomes. Vigilance in research, ethical frameworks, and societal control will continue to guarantee that AI systems augment human capabilities rather than threaten them.
To boldly go
If you are not convinced of that future reality, I suggest you watch the original Star Trek episode “The Ultimate Computer.” An advanced artificially intelligent control system, the M-5 Multitronic unit, malfunctions and engages in real war rather than simulated war, putting the Enterprise and a skeleton crew at risk. Kirk disables M-5, but he must gamble on the humanity of an opposing starship captain to not retaliate against the Enterprise. The Enterprise is spared. Kirk tells Mr. Spock that he knew the captain personally: “I knew he would not fire. An advantage of man versus machine.”
God help us should we lose that advantage.
Arthur Lazarus is a former Doximity Fellow, a member of the editorial board of the American Association for Physician Leadership, and an adjunct professor of psychiatry at the Lewis Katz School of Medicine at Temple University in Philadelphia, PA. He is the author of several books on narrative medicine, including Medicine on Fire: A Narrative Travelogue and Story Treasures: Medical Essays and Insights in the Narrative Tradition.
