In no world would I have ever expected to call myself a tech fanatic. Over the past few years, I’ve cloistered myself in an anti-tech bubble, consumed by the fear of a dystopia with emotionless robots taking over humankind. Through some investigation, however, I did the one thing I had convinced myself not to do: I changed my mind, and began to support the endeavors of one specific AI enterprise: TinyML.
Despite my stubborn protest, the presence of artificial intelligence is only growing. Around me, friends and family are adapting to a world of dependence on software and machine learning, a branch of AI utilizing data to draw from patterns and inferences with increasingly powerful and humanistic capabilities, all within smaller and smaller devices. According to ABI Research, Tiny Machine Learning devices, requiring significantly less power than traditional machine learning models, are expected to grow from 15.2 million shipments in 2020 to 2.5 billion in 2030.
No matter how much one attempts to avoid AI, its wave of influence will surely affect each and every one of us. If we hope to leave a positive impact on our world, we must understand the applications of these new technologies.
TinyML, a subfield of machine learning and artificial intelligence, holds great promise in advancing our society. While this new field must be approached with caution, we should welcome its ability to increase privacy, reduce energy consumption, and advance our scientific and technological capabilities.
To learn more about TinyML and its applications, I spoke with Dr. Matthew Stewart, a postdoctoral researcher working with Professor Vijay Janapa Reddi, Associate Professor in the John A. Paulson School of Engineering and Applied Sciences and leader of research within the TinyML field.
“We generally define Tiny Machine Learning as a device that runs on about one milliwatt, so it’s very, very resource constrained,” said Stewart. “It has maybe one megabyte of flash memory, a very small amount of RAM, and it’s running on really low power.”
Despite being grouped with “edge computing,” any gadget in which client data is processed as close to the source as possible, TinyML is a smaller and “super edge device…like a thermometer…running on very, very low power,” Steward claimed.
With the use of TinyML, devices such as Amazon Alexa and Google Nest Audio can recognize trends from historical data. Overtime, the devices thus get “smarter,” predicting future outcomes with increased accuracy as they accrue input data.
Yet, as the use of artificial intelligence enters the realm of natural language processing, research points to concerns surrounding devices’ abilities to listen and collect unconsented audio transcriptions and forward said data to advertisement-seeking companies. Interviewed by consumer technology website Lifewire, Erik Haig from Harbor Research, a strategy consultant and developing firm, believes that “devices like [the Amazon Echo] and their counterparts … are not only always in your home, constantly listening to everything you say or do, but they—through years of data collection from their users—have perfected natural language processing.”
Through a simple solution, however, Stewart showcased how users can benefit from TinyML without compromising their privacy. Rather than transferring data to the cloud, TinyML keeps our data within our devices.
Currently, Stewart and Reddi are hoping to utilize TinyML to further address privacy concerns. “We came up with this idea of … building sensors, which basically protects user privacy by doing the processing of the raw data on the device,” said Stewart. In other words, since all of these models run locally, no data can be sent or stored within servers.
I asked Stewart if the benefits of TinyML really outweigh its drawbacks.
“I would say yes,” he responded. “But with a caveat of, you know, you could say that about any technology, right? You can do many terrible things with the internet. But you can also do very cool things with the internet that really help people’s lives, save people’s lives, enrich people’s lives.”
With an open mind, TinyML’s wide range of applications holds a significant and beneficial impact on our day-to-day lives. Machine-related issues can be easily fixed through TinyML’s detection functions. An Australian start-up adopted this strategy through their wind turbines, attaching TinyML to the turbine’s exterior in an effort to detect malfunctions beforehand and notify authorities.
TinyML has gone so far as to take root in the agricultural sector. Through PlantMD, an app utilizing TensorFlow, Google’s machine learning program, farmers can now take photos of sick plants and identify their ailments even without any internet connection.
TinyML reaches the medical world too. Devices like the Solar Scare Mosquito now curb the spread of mosquito-transmitted diseases like Dengue Fever and Malaria. At the third annual TinyML EMEA Innovation Forum, TinyML was shown to improve the monitoring of vital signs such as respiratory rate, heart rate, and blood pressure through the embedding of TinyML into wearable devices.
TinyML also advances a more sustainable world. “The main application for [TinyML] is… I can monitor the emissions of forests, or I can monitor pollution in cities, or I can monitor traffic better,” explained Stewart. “You put sensors in buildings, and you could use it to … improve the sustainability of that building. That’s a very basic thing that could make a huge difference.”
In our local community, the positive impacts of TinyML are already being recognized. Just this past year, Harvard aided the funding of Mather as a Living Lab, a Mather House project aimed at utilizing miniature TinyML sensors to measure the use of energy, waste, and consumption throughout the seasons.
Should we view this new technology as a friend or a foe? For some, a new age of artificial intelligence and machine learning seems daunting—apocalyptic, perhaps. Beginning my research with these assumptions, I was, at first, intent on not only proving Stewart wrong, but also justifying the dangers brought on by machine learning, adamant to embrace the inevitable wave of artificial intelligence. Yet, I had failed to look at the bigger picture and see what Stewart and many others identified: the power to change lives.
Kaia Patterson ’27 (kpatterson@college.harvard.edu) has yet to learn how to use a smart watch.