Today in Science: Can you decode this message from space?

August 3, 2023: Mysterious message from space, what's behind changes in GPT-4's behavior and body image Barbie. Read more below.
Andrea Gawrylewski, Chief Newsletter Editor
TOP STORIES

E.T. Calling

On May 24 the ExoMars Trace Gas Orbiter beamed a message from Mars toward Earth. An artist-led team created the message as a test to see if we Earthlings could decode an alien transmission. The communiqué was a 40-gigabyte string of numbers describing the waveform of the telemetry data, interwoven with the alien message. A week's effort of filtering the data segment eventually led to an 8.2-kilobyte bitmap image of five speckled clusters set against a blank background. So far no one has figured out what it means. 

Why this matters: Communication with beings from another planet would be tricky, given how difficult it already is for humans from different cultures to understand one another (or from the same family, sometimes!). The creators aimed to craft a message as alien as possible–steering clear of language and even math. 

What the experts say: "If we ever receive a message from an extraterrestrial civilization, I can imagine that there will never be an agreement over the cultural interpretation," says Daniela de Paulis, a co-creator of the message (and one of only three people on Earth to know the solution). "I think there would necessarily be some miscommunication."
This "extraterrestrial message" was created by artists and scientists for a project called A Sign in Space, which challenges people around the world to interpret an imagined alien signal. Credit: A Sign in Space

AI Drift

Three computer scientists testing the AI GPT-4 noticed a decline in its ability between March and June–it successfully identified prime numbers only 2.4 percent of the time in June, whereas in March its success rate was 97.6 percent. The AI also gave less explanation for its answers and developed other behavioral quirks. The researchers wondered: Is GPT-4 getting dumber

What's going on: It's difficult to determine what might be causing the changes in GPT-4's performance. OpenAI, the company that runs GPT, declined to discuss how it develops and trains its large language models, not to mention the inscrutable "black box" nature of AI algorithms. But making changes to the code or adding training data with one outcome in mind carries the potential for ripple effects elsewhere.

What the experts say: Studies of other models have also shown this sort of behavioral shift, or "model drift," over time. That alone could be a big problem for developers and researchers who've come to rely on this AI in their own work."You'll begin to trust a certain kind of behavior, and then the behavior changes without you knowing," says Vishal Misra, a computer science professor at Columbia University. From there, "your whole application that you built on top starts misbehaving."
TODAY'S NEWS
• Scientists recently discovered a jellyfish fossil that is more than 500 million years old, dating to the Cambrian explosion. Here's a primer on that period on Earth. | 12 min read
• The European Space Agency's ambitious Euclid space telescope is up and running and has snapped its first images (scroll down to see one). | 7 min read
• A clinical health psychologist explains the science of how playing with Barbie dolls has affected some women and girls. | 5 min read
• Some scientists imagine that a gigantic shield could be anchored to an asteroid in space to block the sun's rays and lower temperatures on Earth. | 5 min read
More News
EXPERT PERSPECTIVES
• It's a good movie, yes, but Oppenheimer falls a bit flat when it comes to the accuracy of the underlying science, writes Charles Seife, the director of the Arthur L. Carter Institute of Journalism at New York University. "It's unbelievably difficult to tell a cracking story, one with all the tension and drama that the audience demands, while simultaneously staying true to the underlying science, characters and history," he writes. | 6 min read
More Opinion
IMAGE OF THE DAY
Credit: ESA/Euclid/Euclid Consortium/NASA (CC BY-SA 3.0 IGO)
The European Space Agency's Euclid telescope launched on July 1 and traveled about a million miles before turning on this week. Its mission is to collect data about dark energy in the universe. On Monday, the scope sent its first images back to earth. The instrument collected light for 100 seconds to enable its Near-Infrared Spectrometer and Photometer to create this image. During nominal operation, it is expected to collect light for roughly five times longer, unveiling many more distant galaxies. 
The clustered dots in the partially deciphered message above remind me of a dandelion seed head, or clusters of galaxies in a map of what astronomers call our "local group." What do you see? 
Email me and let me know, or send any other feedback to: newsletters@sciam.com. Same time tomorrow!
—Andrea Gawrylewski, Chief Newsletter Editor
Subscribe to this and all of our newsletters here.

Scientific American
One New York Plaza, New York, NY, 10004
Support our mission, subscribe to Scientific American here

Comments

Popular Posts