#30 | Signs Of Water Emerge On Mars

Quote of the Week 💬

Evidence Of Water Uncovered Deep Underground On Mars

⌛ The Seven Second Summary: Scientists from UC Berkeley discovered evidence of a large underground reservoir of water on Mars.

🔬 How It Was Done: 

• NASA's InSight lander has been detecting and recording seismic activity on Mars since 2018, which includes marsquakes, tremors and other planetary movement.

• After many years examining the speed, frequency, and behavior of seismic waves as they traveled through the planet's interior, the team was able to reconstruct a representative model of the internal structure of Mars and assess what might be underground.

• This model includes the thickness and composition of the planet’s crust, mantle, and core, which is similar to the techniques researchers use to discover oil fields, underground caverns, fault lines, mineral deposits and more on Earth.

🧮 Key Results: The team realized their model indicates the planet currently has a reservoir large enough to fill oceans. Unfortunately, the water is between 11.5 - 20 kilometers (7 - 13 miles) below the surface, which is far enough underground to be a challenge to reach with state-of-the-art technologies on Earth.

💡Why This May Matter: An independent study from NASA's Perseverance rover shared more strong evidence of water activity on Mars based on the mineral composition of rock samples it retrieved from a recent mission. This bolsters the findings from NASA's InSight lander, and while the reservoir may not be accessible in the short-term, its mere existence makes it one the most promising places for humanity to search for extraterrestrial life.

📚 Learn More: UC Berkeley. PNAS.

Stat of the Week 📊

Solar-Powered Device Extracts Drinking Water From Air

3 liters

⌛ The Seven Second Summary: Researchers at King Abdullah University of Science and Technology developed a solar-powered device to extract water from thin air.

🔬 How It Was Done:

• The researchers drew design inspiration by emulating how plants move water from their roots to their leaves.

• Their invention uses a special fabric to absorb water in the atmosphere like a sponge, then a different part of its system extracts and transports the water from the fabric to a container by using heat generated from solar energy.

• They eventually tested their device on the rooftop of the university for several to assess how effective it was amidst changing humidity levels and other environmental conditions.

🧮 Key Results: The system did not require manual intervention and active maintenance to work, and it produced up to 3 liters of freshwater per square meter per day.

💡 Why This May Matter: According to the United Nations, over 1.4 billion people live in areas of high or extreme water vulnerability. Many of these places are arid and remote regions where it is expensive, time consuming and challenging to import freshwater to meet demand, so cheap and localized solutions are needed to address growing threats of water scarcity amidst rising global temperatures.

🔎 Elements To Consider: Humidity levels were as low as 40% on some days during the team’s tests. When the humidity was low, the device only captured 0.65 liters of freshwater per square meter per day, and this variability will make it difficult for communities to rely on this technology in its current state.

📚 Learn More: Independent. Nature.

AI x Science 🤖

More Cost Savings & Customization Of AI Models

It’s no secret the AI landscape is a highly competitive space, and the developments we saw since our last newsletter was no exception.

The declining cost curves of the industry continue its downward trajectory. Both Google and Anthropic announced their latest models will decrease latency by 50% and 85% respectively, as well as become 70% and 90% cheaper, respectively. These massive improvements are the results of similar cache techniques that allow developers to store and reuse frequently accessed data. This ultimately reduces the computational resources required for these models to process requests, and the result is yet another astounding achievement.

Our longtime readers are already well aware of the industry’s declining costs. We frequently highlight its ongoing development since it’s one of the most significant AI developments of the year in our opinion. That said, what we found particularly interesting about the Google and Anthropic’s news is that it also seamlessly enables more customization to improve their user experiences.

It is well known Large Language Models (LLMs) need to be adjusted and fine-tuned to extend domain specific knowledge and to tailor its communication style and tone to the use cases it will be deployed in. But another trend we have not mentioned as consistently, but is still vital to understand, is the attempts of the Big Tech research labs to make it their models easier to modify and optimize for their customers.

We see numerous examples of this. From Meta allowing its developer community to generate synthetic data and distill models with their Llama 3.1 release, to Minstral’s plan to release an Agent platform, to Anthropic’s product updates of Claude Projects and Artifacts, and countless others. This trend is notable because the industry’s converging paths of hyper fast, cheap and personalized models is what Apple ultimately wants for Siri: a model that is small enough, fast enough and smart to act as a virtual assistant embedded within each device. This will allow Siri, Gemini and any other on-device models to be attuned to all the knowledge and context of a device in order to be as personalized and helpful as possible.

Of course, phones are not the only devices where embedding an AI model within the device may be useful, it is simply the device where people currently spend the most time across the widest variety of tasks. In this sense, Google and Apple remain in pull position on the consumer side, whereas the business landscape remains more of a jump ball. Time will tell if OpenAI’s hardware device ambitions materialize in a meaningful way, and if their voice assistant release disrupts the status quo to their benefit, or simply pushes consumers even further into their existing phone ecosystems.

Our Full AI Index

• Algorithm Predicts Diseases By Analyzing Tongues: Researchers from Middle Technical University and The University of South Australia developed an algorithm to predict various diseases by analyzing the color of the human tongue. The system was trained on a dataset of 5260 tongue images with people who have diabetes, anemia, asthma, liver and gallbladder issues, as well as other diseases. When the system was tested on 60 never-before-seen images of patients with these different conditions in real-time, it diagnosed these images with 96% accuracy. It is always interesting to see how models develop a taxonomy as it starts to understand the patterns of a problem. In this example, tongue color, contusions, moisture levels and other characteristics are indicative of a variety of ailments. Yet another example of how AI systems may be used in medical settings in the future. University of South Australia. MDPI.

• An Automated Robotic Dental Procedure: The company Perceptive claims its robotic dentistry system successfully completed the world's first fully automated dental procedure on a human. Their system integrates 3D imaging software with a robotic arm to perform tasks such as crown placements in just 15 minutes. If this type of system can safely replicate its performance at scale, then advanced robotics technologies can help dentists significantly reduce the time to perform various procedures. Thus far, Perceptive reports its system can detect caries with at least 90% accuracy, and time will tell how much further the technology can improve upon today’s baseline performance. Businesswire.

• Brain-Computer Interface Helps Man With ALS ‘Speak’: Researchers from UC Davis developed a brain-computer interface (BCI) to help a man with ALS to communicate when he attempted to speak. Their system was able to decode his cortical neural activity into text, display his desired speech on a screen and vocalize the man’s speech through a text-to-speech software. The system’s performance is impressive because it sustained a 97.5% accuracy rate for the patient after 8 months of active use. Furthermore, the BCI device required less than 2 total hours of training to achieve these results, which is incredibly promising for how quickly these devices can be calibrated and personalized for optimal long-term performance. UC Davis Health. NEJM.

Other Observations 📰

FDA Approves Nasal Spray to Treat Serious Allergic Reactions

The United States’ Food and Drug Administration (FDA) recently approved Neffy, a nasal spray to treat severe allergic reactions for both adults and children over 30kg (~60lbs). This approval is a milestone achievement because it is the first epinephrine product that does not require a needle for treatment, such as EpiPens.

The approval was based on the results of four Phase 1 clinical trials. The trials gave a single dose of Neffy to 175 healthy adults and compared the results to those of existing epinephrine injectors. The studies found Neffy was as effective as existing injectors at sending the right amount of epinephrine into the bloodstream, and the nasal spray caused similar increases in blood pressure and heart rate as treatments with needles.

These performance similarities were critical for the FDA’s decision because the studies did not test the nasal spray’s effectiveness while patients were actively suffering from an allergic reaction; it would be highly unethical to purposely induce a life-threatening medical emergency in trial participants. However, since the nasal spray is the same epinephrine medication as needle alternatives, its ability to produce comparable physiological responses in patients is indicative of its effectiveness.

This is promising because many children and adults also have significant anxiety about needle injections. Their kicks, screams and tremors about these fears can make it exceedingly difficult for people to treat them during an allergic reaction, which worsens the state of a life-threatening situation where every minute counts. Hopefully, this alternative product will make it easier for people to receive the treatment they need at a moment’s notice and reduce the amount of emergency room visits and needless deaths due to anaphylaxis around the U.S., and eventually, the world. FDA.

Our Full Science Index

• Model Identifies Neural Patterns Of A Brain As It Learns: Researchers at Cedars-Sinai used an AI model to analyze the brain activity in 17 patients with epilepsy to understand how the brain learns and represents knowledge. The model identified geometric patterns of neurons firing in the hippocampus to represent how people were learning to adapt to changing rules in order to make informed decisions. For example, the participants would learn to associate certain images with either a 'left' or 'right' response in a picture game, and then have to switch to the opposite association when the rules changed and they needed to adapt and infer information from the new context they were placed in. Cedars Sinai. Nature.

• China Remains On Track For A Decline In Annual Emissions: A new Carbon Brief analysis found the country’s CO² emissions in Q2 dropped by 1% year-over-year. This means the nation is still on track to decrease its full year CO² emissions in spite of its ever increasing energy demands. If this trend continues, then 2023 would represent China’s emissions all-time high, and the country would start to display similar ‘structural declines’ to what is already transpiring in Europe. For reference, these developments were projected by the International Energy Agency last year. China’s path to peak emissions has seismic implications for understanding the planet’s trajectory of temperature rise and pollution. Carbon Brief. Reuters.

• Guinea Eliminates Maternal and Neonatal Tetanus: Researchers and healthcare workers in Guinea announced the nation has completed a successful vaccination campaign against maternal and neonatal tetanus, and the disease is now eliminated from the country. Guinea became the 49^th country to reach this milestone out of an initial list of 59 total priority countries assessed by the United Nations (UN). Just 10 countries remain on the UN’s list, which means the world is one meaningful step closer to eliminating the disease from society entirely. UNICEF USA.

Media of the Week 📸

Playing Table Tennis With A Robot

The Olympics may be over but the games are ongoing!

Researchers from Google Deepmind developed a robot that can play competitive table tennis at an amateur human-level. The robot won 45% of matches against 29 human players of varying skill levels, with a breakdown of 100% wins against beginners, 55% against intermediate players, and 0% against advanced players. This demonstration is impressive because it showcases the robot's ability to learn and adapt to complex, high-speed physical games in real-time.

When do you think robots will be competitive with table tennis Olympians? Google. arXiv.

Visualizing Global Carbon Dioxide Emissions

Researchers at NASA's Goddard Space Flight Center used a weather model called GEOS to simulate the movement and density of CO² in the global atmosphere, and created a map of CO² concentrations from January 2020 to March 2020. The map is incredibly detailed and showcases how interconnectedness our planet’s relationship with CO² is. The model uses billions of data points from ground observations and satellite instruments to create a resolution 100x a typical weather model. The simulation shows CO² emissions rising from power plants, fires, cities, and other sources, and you can see the emissions spread across continents and oceans. It’s the type of dataset that can inform climate policy and predict future changes in global emissions, and we hope to find more simulations like this throughout the Internet. NASA.

A Colony Of Anemones Emitting A Fluorescent Glow

Researchers at the Scripps Institution of Oceanography studied the biofluorescence of anemones and discovered a unique protein called AnthoYFP. This protein controls the animal’s fluorescent light as well as provides antioxidant properties that are helping the anemones survive in a changing ocean. The team’s study analyzed nearly 6,000 anemones and found the AnthoYFP protein may help invertebrates reduce oxidative stress induced by temperature extremes, ocean acidity and climate-induced stressors. This discovery may have implications for the future of coral reefs and other marine ecosystems, because scientists can use similar knowledge to genetically engineer corals and other species to become more resilient to climate change. Smithsonian Magazine. PNAS.