When preparing their annual breakthrough issue, the editors and reporters of MIT Technology Review try to identify the 10 technologies “poised to have a transformative effect on the way we live and work.”
Given the current penetration of intelligent machines and networks in our world, narrowing the list sounds daunting. But even though innovations—new ideas, methods, and devices—appear on the winds daily, breakthroughs happen, and 80% of this year’s picks are in critical areas of bioscience and environmental engineering. Specifically, four are medical and bioscience solutions that support drug development and disease tracking, and four address the crisis in global warming. Only two are directly computer-related issues—computer security involving passwords and an emerging tool for AI called synthetic data.
An important key to understanding diseases by mapping proteins was created by the DeepMind Technologies AI lab in London, United Kingdom. Designated one of the most difficult problems in science, protein folding involves the prediction of the three-dimensional structure of a protein from its primary structure (its amino acid sequence). In 2018, an early version of DeepMind’s AlphaFold successfully predicted the most accurate structure for 24 out of 43 proteins, and the program won the 13th Critical Assessment of Protein Structure Prediction (CASP), a worldwide challenge held every two years. In the 2020 CASP, AlphaFold proved to be as accurate as lab techniques, which take incomparably long time to do the same analysis.
Today, the AlphaFold 2 software can predict the structure of proteins to the nearest atom, and researchers working on COVID-19, cancer, and antibiotic resistance have started using the technique. This is important, TR10 reports, because “Nearly everything your body does, it does with proteins…. And what a protein does is determined by its three-dimensional shape.” DeepMind has a public database of protein structures that continues to grow as it maps new ones. It numbers 800,000 entries, and the company says it will add 100 million more in the next year. That’s almost all the known proteins. DeepMind also promises to collaborate with biotech and pharmaceutical companies.
Paxlovid—A Pill for COVID
As you might expect, a new pharmaceutical to address COVID is on this year’s list, but it isn’t one of the vaccines. Pfizer’s Paxlovid is a medication given to people a few days into contracting the virus, and it can reduce the chance of hospitalization by 89%. It works by binding with the protein protease and blocks the virus’s ability to copy itself. Because similar protease enzymes occur in other kinds of coronavirus, the drug might be important in future pandemics. The editors note, “Never before has an entirely new molecule to defeat a disease gone so quickly from a chemist’s bench into the mouths of volunteers and gained approval from the U.S. Food and Drug Administration.”
In October 2021, the World Health Organization finally approved GlaxoSmithKline’s RTS,S or Mosquirix, the world’s first vaccine to prevent malaria. It requires three doses in children between 5 and 17 months, and then a fourth dose at 12 to 15 months. The efficacy is about 50% against severe malaria, a disease that kills more than 600,000 people each year. This is the first vaccine ever approved for a parasitic disease, and other vaccines for other parasitic diseases are now in development.
COVID Variant Tracking
In order to detect new variants of COVID-19, about two in every 100 positive test samples have been subjected to gene sequencing in order to update the map of the virus’s genome. The result has been that the SARS-CoV-2 virus is the most sequenced organism on Earth. Open databases like GISAID and Nextstrain now offer more than seven million genetic maps of the disease.
Carbon Extraction Factories
In September 2021, the Swiss company Climeworks’ carbon capture plant, called Orca, went online at the Hellisheiði geothermal power station outside Reykjavík, Iceland. Orca is the largest direct air carbon capture facility in the world. Powered by 100% renewable energy, the machine draws in ambient air and filters out carbon dioxide (CO2) before returning the clean air through exhaust ventilation. The captured CO2 is heated and mixed with water and then pumped underground where it binds with basalt rock. Orca can extract 4,000 metric tons of carbon dioxide in a year.
Canada-based Carbon Engineering is working on larger carbon extraction plants in Scotland and Norway, and it will start construction this year for a plant in the southwestern United States that will be able to remove one million tons of CO2 a year.
Iron Grid Batteries
Renewable energy sources are subject to fluctuating flows, as when the sun goes down and the winds subside, and to maintain availability, cheap, long-lasting storage batteries are needed. New types of iron-based batteries from companies like ESS and Form Energy offer a solution. The ESS batteries have four- to 12-hour capacities and are, today, installed in grid-scale settings. Form Energy has batteries capable of 100 hours of storage, and they’re slated for installation in a one-megawatt test plant in Minnesota in 2023. There are still some structural problems with the technology, but others like lithium-ion batteries have both operational challenges and issues with the availability of materials.
Practical Fusion Reactors
Research into fusion reactors began in the 1940s, and fusion power is the only item on the breakthrough list that isn’t available now. Availability is predicted in “about 10 years.” Fusion has, however, attracted billions in investment, and companies and national labs around the world continue to refine models. The editors explain the current urgency of development: “Fusion promises to generate cheap, carbon-free, always-on energy, with no meltdowns and little radioactive waste and nearly limitless fuel sources.” It’s a form of fusion where two lighter atomic nuclei combine to form a heavier nucleus while releasing energy. Cosmic models that offer demonstrable proof of the efficiency of the process can be observed in the sun and other stars.
Recent developments of commercial projects include a critical element that Commonwealth Fusion Systems (CTS) announced late last year. The Cambridge, Mass.-based company demonstrated a high-temperature superconducting magnet that’s the most powerful version of this component of its future fusion reactor. CTS is now building a factory to manufacture the magnets, and it hopes to deliver fusion energy by the early 2030s.
Also, the privately owned research project Tokamak Energy in Oxford, U.K. announced that its ST40 spherical fusion reactor has reached the threshold temperature for commercial fusion energy (100 million °C/180 million °F).
Proof of Stake
The amount of energy required to process (mine) cryptocurrency transactions is alarming. The editors offer an example: “In 2021, the Bitcoin network consumed upwards of 100 terawatt-hours, more than the typical annual energy budget of Finland.” Seen from the other end of the mining transaction, environmental author David Wallace-Wells warns, “Bitcoin now produces as much CO2 each year as a million transatlantic flights.”
Now, Ethereum, the number two cryptocurrency, plans to shift to a different method for processing cryptocurrency called proof of stake, which it says will reduce energy use by 99.5%. The current blockchains used by miners work to solve cryptographic puzzles and to provide rewards in cryptocurrency. With proof of stake, the validators don’t consume massive computing power competing for the right to verify a new block. The editors explain, “Instead, their cache, or stake, of cryptocurrency allows them to enter a lottery. Those who are chosen, gain the authority to verify a set of transactions (and so earn more cryptocurrency).” Ethereum intends to be the largest network to use proof of stake, and if successful, “Ethereum’s proof-of-stake blockchain could set the stage for wider adoption of the energy-saving technology.”
The End of Passwords
For about 50 years, passwords have been the authentication method preferred by almost everyone, despite their ability to be guessed, brute-force cracked, or simply stolen from sticky notes or hacked customer databases. And even though “1-2-3-4” and the word “password” continue to show up in the lists of top-ten most-used passwords, the vulnerable identifiers persist today. But now, there’s a shift away from the alphanumeric to biometrics, fingerprint and facial recognition, and physical security keys like YubiKey that provide strong two-factor and multifactor sign-in.
Physical keys are available that will serve as a fingerprint reader for anything you plug it into or connect via Bluetooth. Given that a recent Verizon data breach investigation found “81% of hacking-related breaches leveraged either stolen and/or weak passwords,” a better way is long overdue, and two-factor authentication is a good start.
Synthetic Data for AI
To effectively train AI, you need data, often big data sets. When these aren’t available for a new project, or the data at hand is too sensitive to use, synthetic data can often fill the need. Synthetic data sets are computer-generated samples with the same statistical characteristics of genuine sets. These have been in use for a while. An example the editors point to are the autonomous cars that are trained on virtual streets before they’re ready for macadam and curbs.
As the applications of AI have proliferated, companies like Datagen and Synthesis AI, along with a number of universities, now offer synthetic data for facial recognition systems, finance, and insurance. Some of the rapid growth comes from a particular branch of AI. The editors explain, “This boom in synthetic data sets is driven by generative adversarial networks (GANs), a type of AI that is adept at generating realistic but fake examples, whether of images or medical records.”
The time between the flash of the innovative idea and the noise generated by a genuine breakthrough can very long, as in the case of fusion energy, or very short, in the record time it took to get Paxlovid out of the lab and into pharmacies. The time between isn’t a measure of the importance of the breakthrough, but more likely of the difficulty of building the final version. Development times, though, seem to be shortening as technologies continue to converge and accelerate.