Simulated Light Schedules for Sleep-Wake Adjustments: Exploring the Incorporation of Reinforcement Learning

For most of human history, our internal clocks were set by the sun. We woke with the dawn and slept with the dusk, a rhythm synchronized by the light hitting our eyes. Today, however, we live in a world of artificial light that never truly turns off. This constant glow can confuse the body's biological timing system, leading to sleep that feels shallow or a wakefulness that arrives at the wrong time. Scientists have long known that light is the primary signal telling our bodies when to rest and when to be active, more so than food, temperature, or noise. To fix a disrupted sleep schedule, researchers have traditionally relied on rigid, often uncomfortable plans that force people to stare into bright lights or sit in total darkness at specific times. These methods work in theory but are difficult to maintain in real life, where people have jobs, families, and social lives that do not fit into a perfect laboratory schedule. The challenge has been finding a way to use light to gently nudge the body back to a healthy rhythm without demanding that a person abandon their daily routine.

A team of researchers at Monash University Malaysia has taken a new approach to this problem by teaching a computer to learn how to design these light schedules. Instead of using a fixed set of rules or a pre-written manual, they built a system that uses a type of artificial intelligence known as reinforcement learning. Imagine a student learning to play a musical instrument; they try a note, hear if it sounds right, and adjust their finger placement for the next attempt. The researchers applied this same trial-and-error logic to light. They created a digital framework where a computer algorithm acts as a student, trying to modify a person's daily light exposure to achieve a specific sleep goal. The computer does not just guess; it learns from a database of real human behavior. It looks at how light patterns relate to when people actually fall asleep and wake up, then iteratively tweaks the light schedule to see if it moves the person closer to their desired sleep time.

To train this system, the researchers first needed real-world data. They recruited 19 young adults, mostly university students, and asked them to wear two small devices for six days. One device, worn like a pendant, measured the light hitting their eyes throughout the day, capturing not just brightness but also the specific colors of light that affect the body's internal clock. The other device, worn on the wrist, tracked their movement to determine exactly when they were asleep and when they were awake. The team also asked the participants to keep simple sleep diaries. From this data, the researchers grouped the participants into five distinct categories based on their average sleep times and how efficiently they slept. This grouping allowed them to treat "sleep types" as targets, much like a coach might group athletes by their natural running style.

The core of the study involved two main steps. First, the team built a classification model, a type of computer program that can look at a six-day light schedule and predict which of the five sleep categories a person belongs to. They tested this model on new data and found it could correctly identify the sleep category with high accuracy. This model became the "teacher" for the second part of the system. In the second phase, they used the reinforcement learning algorithm to generate new light schedules. The computer started with a person's original light routine and was given a target: "change this schedule so the person ends up in a different sleep category." The algorithm made small adjustments to the light levels, such as dimming the lights in the evening or brightening them in the morning, and then asked the classification model, "Did this change move us closer to the target?" If the answer was yes, the computer kept that change; if not, it tried something else. This process repeated thousands of times until the algorithm found a schedule that successfully shifted the predicted sleep pattern.

The results showed that the computer could indeed generate light schedules that moved people toward different sleep-wake patterns. For some of the sleep categories, the generated schedules looked very similar to the natural light patterns the researchers had observed in their participants. However, for other categories, the computer created schedules that were quite different from what people naturally experience, sometimes suggesting light levels that would be difficult to achieve with standard home lighting. The researchers simulated what would happen to the body's internal clock if someone followed these computer-generated schedules. In several cases, the simulations suggested the schedules would successfully shift the timing of melatonin release, a hormone that signals the body it is time to sleep. Yet, the study also highlighted a crucial limitation: the computer sometimes pushed the light levels to extremes that are not practical for everyday life. This suggests that while the technology works, it must be carefully balanced against what is actually possible for a human to do.

The study does not claim to have solved the problem of sleep disorders or to have found a perfect cure for jet lag. Instead, it demonstrates a new method for creating personalized light plans. The researchers emphasize that their work is a prototype, a starting point that proves the concept of using machine learning to design light diets. They note that their data came from a small group of young, healthy adults, so the results cannot yet be applied to everyone, such as older adults or people with specific health conditions. Furthermore, because the system relies on predicting sleep patterns based on light rather than measuring the body's internal chemistry directly, the findings are based on simulations and statistical patterns rather than direct physiological proof. The authors suggest that for this technology to become a practical tool, it will need to be tested on much larger groups of people and refined to ensure the light schedules it creates are not only effective but also realistic to follow.

Ultimately, this research opens a door to a more flexible future for sleep science. By moving away from rigid, one-size-fits-all rules and toward adaptive, learning systems, scientists may eventually be able to offer light recommendations that fit seamlessly into a person's actual life. The computer does not need to force a person to stare into a bright light for an hour; it can suggest subtle changes to the lighting in a home or office that accumulate over days to gently reset the body's clock. While the current version of the system is still in its early stages, the ability to generate these tailored schedules using reinforcement learning represents a significant step forward. It shifts the focus from what is theoretically possible in a lab to what might be achievable in the real world, offering a promising path toward better sleep for people navigating a world that never truly sleeps.