Katie Bouman Created AI Algorithm That Renders The First True Image Of A Black Hole


Images of black holes weren't actually "black holes". As a matter of fact, they were illustrations based on mathematical equations, theories and imaginations.

This is changing, as a 29-year-old computer scientist has earned plaudits worldwide for helping develop an AI-powered algorithm that created the first-ever direct image of a black hole.

Katie Bouman led the development of the project, assisted by a team from MIT's Computer Science and Artificial Intelligence Laboratory, the Harvard-Smithsonian Center for Astrophysics and the MIT Haystack Observatory.

The photo below shows a halo of dust and gas 53 million light-years (501,418,715,046,782,440,000 kilometers) from Earth, captured by the Event Horizon Telescope (EHT) - a network of eight linked telescopes - rendered by Bouman's algorithm.

The black hole that stretches just under 40 billion kilometers is located at the center of the supergiant elliptical galaxy in the constellation Virgo, known as Messier 87, or simply M87.

The first direct image of a black hole
The first true and direct image of a black hole

Black holes by definition are supposed to be invisible when seen in the vast black dark space. This is because the region exhibits such strong gravitational effects that nothing - not even particles and electromagnetic radiation such as light - can escape.

But here, they can give off a shadow when they interact with the material around them. This theory was the foundation of Bouman's research.

Bouman started creating the algorithm since 2016, while she was still a graduate student at the Massachusetts Institute of Technology (MIT).

The algorithm, which Bouman named CHIRP (Continuous High-resolution Image Reconstruction using Patch priors) feed on the 5,000 terabytes of data collected by the Event Horizon Telescope.

The data was so large that it couldn't be transferred via the internet, and had to be stored on hundreds of hard drives. Weighing a combined half a ton, the hard drives needed to be physically delivered to the supercomputer at the MIT Haystack Observatory.

To give a sense of how hard the task was, the Milky Way’s black hole that is 250,614,750,218,665,392 kilometers away from Earth, and has a mass of 4.1 million Suns and a diameter of 60 million kilometers. That is an equivalent of travelling from London to New York 45 trillion times.

As noted by the EHT team, it is like being in New York and trying to count the dimples on a golf ball in Los Angeles, or imaging an orange on the moon.

To photograph something so impossibly far away, the team needed a telescope as big as the Earth itself.

But since such a gargantuan machine does not exist, the team connected together telescopes from around the planet, and combined their data to capture an accurate image of the black hole.

The telescopes needed to be stable, and their readings completely synchronized.

But since astronomical signals reach the radio telescopes at slightly different rates, Bouman and her team of more than 200 scientists had to figure out how to account for the variations so calculations would be accurate and visual information could be extracted.

This is where the team used atomic clocks that are so accurate that they lose just one second per hundred million years, turning the combined data into a cohesive image.

Bouman assisted the team by providing computational support to learn about general relativity in the strong-field regime.

She theorized that black holes leave a background shadow of hot gas, and this is where the machine learning algorithm fills in gaps in data produced by the telescopes.

The algorithm then reconstructed and refined the original images to prepare the final historical image of the black hole.

MIT explained that:

"Bouman adopted a clever algebraic solution to this problem: If the measurements from three telescopes are multiplied, the extra delays caused by atmospheric noise cancel each other out. This does mean that each new measurement requires data from three telescopes, not just two, but the increase in precision makes up for the loss of information."

In the hours after the photo's momentous release, Bouman became an international sensation, with her name trending on Twitter.

Bouman was also hailed by MIT and the Smithsonian on social media.

"3 years ago MIT grad student Katie Bouman led the creation of a new algorithm to produce the first-ever image of a black hole," wrote MIT's Computer Science & Artificial Intelligence Lab. "Today, that image was released."

The CHIRP algorithm is also said to be capable for use on any imaging system that uses radio interferometry.

One of the reasons why the team decided to capture the image of M87 instead of the Milky Way’s black hole which is a lot closer, was because the attempt was way too challenging to image accurately at the time due to rapid variability in light output.

"Watching in disbelief as the first image I ever made of a black hole was in the process of being reconstructed," she wrote in the caption to the Facebook post