To the Stars with Data: December 12 2021
[Beautifully] visualizing anthropogenic mass, using stem cells to cure diabetes, a demogorgon-esque rocket, and more
Hello Datanauts!
This week, I have some exciting articles to send your way. First, let’s take a journey through humanity’s anthropogenic mass. Next, let’s dive into some awesome news on the diabetes front, a language AI that outperforms AIs 25 times its size, and a demogorgon-esque rocket. Oh, and I also included some companies doing futuristic things, of course.
Let’s go to the stars with data!
Thought-Provoking Data Viz
🏗 Visualizing the Accumulation of Human-Made Anthropogenic Mass
Quick hits. This beautifully-illustrated visualization explores anthropogenic mass, or the mass of materials embedded in inanimate solid objects that are made by humans. In 2020, we recently passed a tipping point; the total anthropogenic mass weighed more than biomass, or the weight of all living things on Earth (here’s another gorgeous viz about that biomass). Notably, the researchers of the study where this data originated predict that humans will triple the amount of anthropogenic mass by 2040. 😱
Digging deeper. Humans represent only 0.01% of global biomass, yet we’ve outproduced the weight of global biomass and may triple it within the next 30 years. As noted in the article, the pace of our anthropogenic mass generation is more than 1 person generating their weight every week.
Earth News
💉 Researchers use stem cells to restore insulin production in trial participant
Quick hits. By producing islet cells (the cells that produce insulin) from stem cells and implanting them into a patient with Type 1 diabetes, researchers were able to restore insulin production. Amazingly, the patient was able to fully produce the insulin his body needed after the procedure. The current iteration of the therapy requires the patient to take immunosuppressant drugs to ensure his body does not attack the islet cells (as they are perceived by the body to be foreign), but the researchers are confident they'll be able to iterate a version that doesn't require these drugs. Though these results are promising, this is only the result from the first trial participant and the results are not peer-reviewed. The trial will last until ~2028 and involve 16 other participants.
Digging deeper. Insulin was not discovered until 1922, so people with diabetes before then (and until insulin was widely available) had to cope with their diabetes with only fasting and calorie-restricted diets. Living with diabetes is not easy (see this episode of Ologies), but technology has thankfully progressed to a point where people with diabetes can remain informed about their blood sugar levels and react accordingly. Seeing results like those from this trial is promising for the ~5% of the population that lives with diabetes, though there will unfortunately be a long time before treatments such as these can become available for the general population (and especially those in low- and middle-income countries).
🤖 DeepMind releases RETRO, a language AI that boasts it can beat AI 25 times its size
Quick hits. DeepMind, an AI company in the UK, recently announced RETRO, a language AI that can beat AI 25 times its size (measured in number of parameters). The key to RETRO’s success is its structure, the look-up structure, which provides an accompanying database that it can reference while operating. Because RETRO can access a database of material, it requires far fewer parameters to train: only 7 billion, compared to the hundreds of billions required by its competitors. RETRO’s approach also provides a number of other benefits, including the ability to add to its database material as updates are needed, the ability to remove harmful, biased, or innacurate content from its database, and the ability to understand more about RETRO’s decisions compared to traditional AI models because the database provides its reference material.
Digging deeper. To be clear, this is not the first time an AI with a look-up structure has been developed, though this is the first time its been used on a large-scale language model. Though their approach does not fix the problem of large-scale language AIs creating harmful ‘-ist’ content (see more here), the look-up approach provides us a chance to address issues, unlike its competitors. From an environmental perspective, I’m pleased with the look-up structure. Because RETRO can be trained with only 7 billion parameters (instead of hundreds of billions), the model requires significantly less power.
Space News
🚀 Rocket Lab unveils their Neutron rocket
Quick hits. Rocket Lab recently introduced the most recent iteration of their Neutron vehicle. (By most recent iteration, I mean that it will very likely change before they officially “launch” (pun intended) Neutron in a number of years.) Rocket Lab refers to Neutron as a “rocket of 2050” because of a few innovations that truly stand out (though I think we’ll be a lot further along by 2050, personally). Before digging deeper, I recommend you check out the announcement video. Rocket Lab obviously went all out and made it as epic as possible.
Digging Deeper. Here are a few of Neutron’s specs:
It stands at 40m tall and its base diameter is 7m
It uses 7 Archimedes engines (engines designed by Rocket Lab and yet to be tested)
It can launch 8,000kg (max 15,000kg) to low-earth orbit (LEO) and 1,500 kg to Mars or Venus
And here are some of the innovations that make it pretty cool:
The rocket is made from the lightweight yet strong carbon composite, which is atypical for comparative rockets because it's complex to manufacture. However, 3D-printing technology allows Rocket Lab to manufacture it quickly and efficiently.
Neutron will release its payload through a demogorgon-esque robotic flower. (You need to see the pic below to believe it.) As a result, the fairings (the “petals” of the flower) stay attached to the rocket so the rocket remains essentially whole the entire trip. (Props to the folx at The Orbital Index for the demogorgon comparison - it’s very apt!)
Though the rocket is reusable, the second stage (what propels the payload out of Neutron) is not reusable. Rather, it hangs inside the rocket instead of needing to support its own weight (and face the huge force during launch). Because the second stage can weigh significantly less, the payload can be heavier.
The first stage of Neutron (the rocket minus the second stage) will return directly to the launchpad instead of landing on costly barges in oceans.
Future-Thinking Companies
🥤 Carbios. They use enzymes to break down plastics. It’s that simple.
Note: Does this mean we should continue manufacturing plastic and not feel bad about its environmental impact? Definitely not. But is something like Carbios part of the solution to climate change by addressing the plastic already created? Definitely.
🍯 MeliBio. They make honey without bees! Their mission may seem counter to “saving the bees”, but it fits in just fine; commercial beekeeping/honey harvesting places immense pressure for wild and native bee populations.