Month: December 2023

• What Gaganyaan tells us about chat AI, and vice versa

  Talk of chat AI* is everywhere, as I’m sure you know. Everyone would like to know where these apps are headed and what their long-term effects are likely to be. But it seems that it’s still too soon to tell what they will be, at least in sectors that have banked on human creativity. That’s why the topic was a centrepiece of the first day of the inaugural conference of the Science Journalists’ Association of India (SJAI) last month, but little came of it beyond using chat AI apps to automate tedious tasks like transcribing. One view, in the limited context of education, is that chat AI apps will be like the electronic calculator. According to Andrew Cohen, a professor of physics at the Hong Kong University of Science and Technology, as quoted (and rephrased) by Amrit BLS in an article for The Wire Science:

  When calculators first became available, he said, many were concerned that it would discourage students from performing arithmetic and mathematical functions. In the long run, calculators would negatively impact cognitive and problem-solving skills, it was believed. While this prediction has partially come true, Cohen says the benefits of calculators far outweigh the drawbacks. With menial calculations out of the way, students had the opportunity to engage with more complex mathematical concepts.

  Deutsche Welle had an article making a similar point in January 2023:

  Daniel Lametti, a Canadian psycholinguist at Acadia University in Nova Scotia, said ChatGPT would do for academic texts what the calculator did for mathematics. Calculators changed how mathematics were taught. Before calculators, often all that mattered was the end result: the solution. But, when calculators came, it became important to show how you had solved the problem—your method. Some experts have suggested that a similar thing could happen with academic essays, where they are no longer only evaluated on what they say but also on how students edit and improve a text generated by an AI—their method.

  This appeal to the supposedly higher virtue of the method, over arithmetic ability and the solutions to which it could or couldn’t lead, is reminiscent of a similar issue that played out earlier this year – and will likely raise its head again – vis-à-vis India’s human spaceflight programme. This programme, called ‘Gaganyaan’, is expected to have the Indian Space Research Organisation (ISRO) launch an astronaut onboard the first India-made rocket no earlier than 2025.

  The rocket will be a modified version of the LVM-3 (previously called the GSLV Mk III); the modifications, including human-rating the vehicle, and their tests are currently underway. In October 2023, ISRO chairman S. Somanath said in an interview to The Hindu that the crew module on the vehicle, which will host the astronauts during their flight, “is under development. It is being tested. There is no capability in India to manufacture it. We have to get it from outside. That work is currently going on. We wanted a lot of technology to come from outside, from Russia, Europe, and America. But many did not come. We only got some items. That is going to take time. So we have to develop systems such as environmental control and life support systems.”

  Somanath’s statement seemed to surprise many people who had believed that the human-rated LVM-3 would be indigenous in toto. This is like the Ship of Theseus problem: if you replace all the old planks of a wooden ship with new ones, is it still the same ship? Or: if you replace many or all the indigenous components of a rocket with ones of foreign provenance, is it still an India-made launch vehicle? The particular case of the UAE is also illustrative: the country neither has its own launch vehicle nor the means to build and launch one with components sourced from other countries. It lacks the same means for satellites as well. Can the UAE still be said to have its own space programme because of its ‘Hope’ probe to orbit and study Mars?

  Cohen’s argument about chat AI apps being like the electronic calculator helps cut through the confusion here: the method, i.e. the way in which ISRO pieces the vehicle together to fit its needs, within its budget, engineering capabilities, and launch parameters, matters the more. To quote from an earlier post, “‘Gaganyaan’ is not a mission to improve India’s manufacturing capabilities. It is a mission to send Indians to space using an Indian launch vehicle. This refers to the recipe, rather than the ingredient.” For the same reason, the UAE can’t be said to have its own space programme either.

  Focusing on the method, especially in a highly globalised world-economy, is a more sensible way to execute space programmes because the method – knowing how to execute it, i.e. – is the most valuable commodity. Its obtainment requires years of investment in education, skilling, and utilisation. I suspect this is also why there’s more value in selling launch-vehicle services rather than launch vehicles themselves. Similarly, the effects of the electronic calculator on science education speak to advantages that are virtually unknown-unknowns, and it seems reasonable to assume that chat AI will have similar consequences (with the caveat that the metaphor is imperfect: arithmetic isn’t comparable to language and large-language models can do what calculators can and more).

  ---

  * I remain wary of the label ‘AI’ applied to “chat AI apps” because their intelligence – if there is one beyond sophisticated word-counting – is aesthetic, not epistemological, yet it’s also becoming harder to maintain the distinction in casual conversation. This is after setting aside the question of whether the term ‘AI’ itself makes sense.

  2023.12.29

• A survey of El Salvador’s bitcoin adoption

  On December 22, a group of researchers from the US had a paper published in Science in which they reported the results of a survey of 1,800 households in El Salvador over its members’ adoption, or not, of bitcoin as currency.

  In September 2021, the government of El Salvador president Nayib Bukele passed a ‘Bitcoin Law’ through which it made the cryptocurrency legal tender. El Salvador is a country of 6.3 million people, many poor and without access to bank accounts, and Bukele pushed bitcoins as a way to circumvent these issues by allowing anyone with a phone with an internet connection to access a central-bank-backed cryptocurrency wallet and trading the virtual coins. Yet even at the time, adoption was muted by concerns over bitcoins’ extreme volatility.

  In the new study, the researchers’ survey spotlighted the following issues, particularly that the only demographic that seemed eager to adopt the use of bitcoins as currency was “young, educated men with bank accounts”:

  Privacy and transparency concerns appear to be key barriers to adoption; unexpectedly, these are the two concerns that decentralized currencies such as crypto aim to address. … we document that this payment technology involves a large initial adoption cost, has benefits that significantly increase as more people use it …, and faces resistance from firms in terms of its adoption. … Moreover, our survey work using a representative sample sheds light on how it is the already wealthy and banked who use crypto, which stands in stark contrast with recurrent hypotheses claiming that the use of crypto may help the poor and unbanked the most.

  Bitcoin isn’t private. Its supporters claimed it was because the bitcoin system could evade surveillance by banks, but law enforcement authorities simply switched to other checks-and-balances governments have in place to track, monitor, and – if required – apprehend bitcoin users, with help from network scientists and forensic accountants.

  The last line is also reminiscent of several claims advanced by bitcoin supporters – rather than well-thought-out “hypotheses” advanced by scholars – in the late 2010s about the benefits the use of cryptocurrencies could bring to the Global South. The favour the cryptocurrency enjoyed among these people was almost sans exception rooted in its technological ‘merits’ (such as they are). There wasn’t, and still isn’t in many cases, any acknowledgment of the social institutions and rituals that influence public trust in a currency – and the story of El Salvador’s policy is a good example of that. The paper’s authors continue:

  There is substantial heterogeneity across demographic groups in the likelihood of adopting and using bitcoin as a means of payment. The reasons that young, educated men are more likely to use bitcoin for transactions remain an open question. One hypothesis is that this group has higher financial literacy. We found that, even conditional on access to financial services and education, young men were still more likely to use bitcoin. However, financial literacy encompasses several other areas of knowledge that are not captured by these controls. An alternative hypothesis is that young, educated men have a higher propensity to adopt new technologies in general. The literature on payment methods has documented that young individuals have a greater propensity to adopt means of payment beyond cash, such as cards (87). Nevertheless, further research is necessary to causally identify the factors contributing to the observed heterogeneity across demographic groups.

  India and El Salvador are very different except, by virtue of being part of the Global South, they’re both good teachers. El Salvador is teaching us that something simply being easier to use won’t guarantee its adoption if people also don’t trust it. India has taught me that awareness of one’s own financial illiteracy is as important as financial literacy, among other things. I’ve met many people who won’t invest in something not because they understand it – they might – but because they don’t know enough about how they can be defrauded of their investment. And if they don’t, they simply assume they will lose their money at some point. It’s the way things have been, especially among the erstwhile middle class, for many decades.

  This is probably one of several barriers. Another is complementarity (e.g. “benefits that significantly increase as more people use it”), which implies the financial instrument must be convenient in a variety of sectors and settings, which implies it needs to be better than cash, which is difficult.

  2023.12.28

• Unexpected: Magnetic regions in metal blow past speed limit

  You’re familiar with magnetism, but do you know what it looks like at the smallest scale? Take a block of iron, for example. It’s ferromagnetic, which means if you place it near a permanent magnet – like a refrigerator magnet – the block will also become magnetic to a large extent, larger than materials that aren’t ferromagnetic.

  If you zoom in to the iron atoms, you’ll see a difference between areas that are magnetised and areas that aren’t. Every subatomic particle has four quantum numbers, sort of like its Aadhaar or social security ID. No two electrons in the same system can have the same ID, i.e. one, some or all of these numbers differ from one electron to the next. One of these numbers is the spin quantum number, and it can have one of two values, or states, at any given time. Physicists refer to these states as ‘up’ and ‘down’. In the magnetised portions, in the iron block, you’ll see that electrons in the iron atoms will either all be pointing up or all down. This is a defining feature of magnetism.

  Scientists have used it to make hard-disk drives that are used in computers. Each drive stores information by encoding it in electrons’ spins using a magnetic field, where, say, ‘1’ is up and ‘0’ is down, so a series of 1s and 0s become a series of ups and downs.

  In the iron block, the parts that are magnetised are called domains. They demarcate regions of uniform electron spin in three dimensions in the block’s bulk. For a long time, scientists believed that the ‘walls’ of a domain – i.e. the imaginary surface between areas of uniform spin and areas of dis-uniform spin – could move at up to around 0.5 km/s. If they moved faster, they could destabilise and collapse, allowing a kind of magnetic chaos to spread within the material. They arrived at this speed limit from their theoretical calculations.

  The limit matters because it says how fast the iron block’s magnetism can be manipulated, to store or modify data for example, without losing that data. It also matters for any other application that takes advantage of the properties of ferromagnetic materials.

  In 2020, a group of researchers from the Czech Republic, Germany, and Sweden found that if you stacked up a layer of ferromagnets, the domain walls could move much faster – as much as 14 km/s – without collapsing. Things can move fast in the subatomic realm, yet 14 km/s was still astonishing for ferromagnetic materials. So scientists set about testing it.

  A group from Italy, Sweden, and the US reported in a paper published in Physical Review Letters on December 19 (preprint here) that they were able to detect domain walls moving in a composite material at a stunning 66 km/s – greater than the predicted speed. Importantly, however, existing theories that explain a material’s magnetism at the subatomic scale don’t predict such a high speed, so now physicists know their theories are missing something.

  In their study, the group erected a tiny stack of the following elements, in this order: tantalum, copper, a cobalt-iron compound, nickel, the cobalt-iron compound, copper, and tantalum. Advanced microscopy techniques revealed that the ferromagnetic nickel layer (just a nanometre wide) had developed domains of two shapes: some were like stripes and some formed a labyrinth with curved walls.

  The researchers then tested the domain walls using the well-known pump-probe technique: a blast of energy first energises a system, then something probes it to understand how it’s changed. The pump here was an extremely short pulse of infrared radiation and the probe was a similarly short pulse of ultraviolet (UV) radiation.

  The key is the delay between the pump and probe pulses: the smaller the delay, the greater the detail that comes to light. (Three people won the physics Nobel Prize this year for finding ways to make this delay as small as possible.) In the study it was 50 femtoseconds, or 500 trillionths of a second.

  The UV pulse was diffracted by the electrons in nickel. A detector picked up the diffraction patterns and the scientists ‘read’ them together with computer simulations of the domains to understand how they changed.

  How did the domains change? The striped walls were practically unmoved but the curved walls of the labyrinthine pattern did move, by about 17-23 nanometres. The group made multiple measurements. When they finally calculated an average speed (which is equal to distance divided by time), they found it to be 66 km/s, give or take 20 km/s.

  

  The observation of extreme wall speed under far-from-equilibrium conditions is the … most significant result of this study,” they wrote in their paper. This is true: even though the researchers found that the domain-wall speed limit in a multilayer ferromagnetic material is much higher than 0.5 km/s – as the 2020 group predicted – they also found it to be a lot higher than the expected 14 km/s. Of course, it’s also stunning because the curved domain walls moved at more than 10-times the speed of sound in that material – and the more curved a portion was, the faster it seemed to move.

  The researchers concluded that “additional mechanisms are required to fully understand these effects” – as well as that they could be “important” to explain “ultrafast phenomena in other systems such as emerging quantum materials”.

  This is my second recent post about scientists finding something they didn’t expect to, but in settings more innocuous than in the vast universe or at particle smashers. Read the first one, about the way paint dries, here.

  2023.12.21

• You can do worse than watching paint dry – ask physics

  I live in Chennai, a city whose multifaceted identity includes its unrelenting humidity. Its summers are seldom hotter than those in Delhi but they are more unbearable because it leaves people sweaty, dehydrated, and irritated. Delhi’s heat doesn’t have the same effect because when people sweat there, the droplets evaporate into the air, whose low relative humidity allows it to ‘accommodate’ moisture. But in Chennai, the air is almost always humid, more so during the summer, and the sweat on people’s skin doesn’t evaporate. Yet their bodies continue to sweat because it’s one of the few responses they have to the heat.

  Paint, fortunately, has a different story to tell. Fresh paint on a wall doesn’t dry faster or slower depending on how humid the air is. This is because pain is made of water plus some polymers whose molecules are much larger than those of water. At first, water does begin to escape the paint and evaporate from the surface. This pulls the polymer molecules to the surface in a process called advection. On the surface, the polymer molecules form a dense layer that prevents the water below from interacting directing with the air, or its humidity. So the rate of evaporation slows until it reaches a constant low value. This is why, even in dry weather, paint takes its time to dry.

  Scientists have verified that this is the case in a new study, in which they also reported that their findings can be used to understand the behaviour of little respiratory droplets in which viruses travel through the air. (Some studies – like this and this – have suggested that a virus’s viability may depend on the relative humidity and how quickly the droplet dries, among other factors. Since the relative humidity varies by season, a link could explain why some viral outbreaks are more seasonal.)

  Generally, the human skin – as the largest outer-organ of the human body – is responsible for making sure the body doesn’t lose too much water through evaporation. Scientists think that it can adjust how much sweat is released on the skin by modifying the mix of lipids (fatty substances) in its outermost layer. If it did, it would be an example of an active process – a dynamic response to environmental and biological conditions. Paint drying, on the other hand, is a non-active process: the rate of evaporation is limited by the polymer molecules at the surface and their properties.

  In 2017, a chemical engineer at the University of Bordeaux named Jean-Baptiste Salmon predicted that an active process may not be needed at all to explain humidity-independent evaporation because it arises naturally in solutions like that of paint. The new study tested the prediction of Salmon et al. using a non-active polymer solution, i.e. one that’s incapable of developing an active response to changes in humidity.

  They filled a plastic container with polyvinyl alcohol, then drilled a small hole near the bottom and fit a glass tube there with an open end. The liquid could flow through the tube and evaporate from the end. To prevent the liquid from evaporating from its surface, they coated it with an oily substance called 1-octadecene. They placed this container on a sensitive weighing scale and the whole apparatus inside a sealed box with adjustable humidity. The researchers adjusted the humidity from 25% to 90% and each time studied the evaporation rate for more than 16 hours.

  They found that Salmon et al. were right: the evaporation rate was higher for around three hours before dropping to a lower value. This was because polymer molecules had accumulated at the layer where the liquid met the air. But in these three hours, the rate of evaporation didn’t drop even when the humidity was increased. In other words, humidity-independent evaporation begins earlier than Salmon et al. predicted.

  The researchers also reported another divergence: the evaporation rate wasn’t affected by a relative humidity of up to 80% – but beyond that, the rate fell if the humidity increased further. So what Salmon et al. said was at play but it wasn’t the full picture; some other forces were also affecting the evaporation.

  The researchers ended their paper with an idea. They took a closer look at the open end of the tube, where the polyvinyl alcohol evaporated, with a microscope. They found that the polymer layer was overlaid with a stiffer semisolid, or gel-like, layer. Such layers are known to form when there is a compressive stress, and further block evaporation. The researchers found that their equations to predict the evaporation rate roughly matched the observed value when they were modified to account for this stress. They also found that a sufficiently thick gel layer could form within one second – a short time span considering the many hours over which the rate of evaporation evolves.

  “These discrepancies motivate the search for extra physics beyond Salmon et al., which may again relate to a gelled polymer skin at the air-solution interface,” they concluded in their paper, published in the journal Physical Review Letters on December 15.

  2023.12.17

• A new path to explaining the absence of antimatter

  Our universe was believed to have been created with equal quantities of matter and antimatter, only for antimatter to completely disappear over time. We know that matter and antimatter can annihilate each other but we don’t know how matter came to gain an upper hand and survive to this day, creating, stars, planets, and – of course – us.

  In the theories that physicists have to explain the universe, they believe that the matter-antimatter asymmetry is the result of two natural symmetries being violated. These are the charge and parity symmetries. The charge (C) symmetry is that the universe would work the same way if we replaced all the positive charges with negative charges and vice versa. The parity (P) symmetry refers to the handedness of a particle. For example, based on which way an electron is spinning, it’s said to be right- or left-handed. All the fundamental forces that act between particles preserve their handedness except the weak nuclear force.

  According to most particle physicists, matter won the war against antimatter through some process that violated both C and P symmetries. Proof of CP symmetry violation is one of modern physics’s most important unsolved problems.

  In 1964, physicists discovered that the weak nuclear force is capable of violating C and P symmetries together when it acts on a particle called a K meson. In the 2000s, a different group of physicists found more evidence of CP symmetry violation in particles called B mesons. These discoveries proved that CP symmetry violation is actually possible, but they didn’t bring us much closer to understanding why matter dominated antimatter. This is because of particles called quarks.

  Quarks are the smallest known constituent of the universe’s matter particles. They combine to form different types of bigger particles. For example, all mesons have two quarks each. All the matter that we’re familiar with are instead made of atoms, which are in turn made of protons, neutrons, and electrons. Protons and neutrons have three quarks each – they’re baryons. Electrons are not made of quarks; instead, they belong to a group called leptons.

  To explain the matter-antimatter asymmetry in the universe, physicists need to find evidence of CP symmetry violation in baryons, and this hasn’t happened so far.

  On December 7, a group of researchers from China published a paper in the journal Physical Review D in which they proposed one place where physicists could look to find the answer: the decay of a particle called a lambda-b baryon to a D-meson and a neutron.

  Quarks come in six types, or flavours. They are up, down, charm, strange, top, and bottom. A lambda-b baryon is the name for a bundle containing one up quark, one down quark, and one strange quark. A D-meson is any meson that contains a charm quark. In the process the researchers have proposed, the D-meson exists in a superposition of two states: a charm quark + an up anti-quark (D⁰ meson) and a charm anti-quark and an up quark (D⁰ anti-meson).

  The researchers have proposed that the probability of a lambda-b baryon decaying to a D⁰ meson versus a D⁰ anti-meson could be significantly different as a result of CP symmetry violation.

  The proposal is notable because the researchers have tailored their prediction to an existing experiment that, once it’s upgraded in future, will collect data that can be used to look for just such a discrepancy. This experiment is called the LHCb – ‘LHC’ for Large Hadron Collider and ‘b’ for beauty.

  The LHCb is a detector on the LHC, the famous particle-smasher in Europe that slams energetic beams of protons together to pry them open. The detectors then study the particles in the detritus and their properties. LHCb in particular tracks the signatures of different types of quarks. Physicists at CERN are planning to upgrade LHCb to a second avatar that’s expected to begin operating in the mid-2030s. Among other features, it will have a 7.5-times higher peak luminosity – a measure of the number of particles the detector can detect.

  If the lambda-b baryon’s decay discrepancy exists in the new LHCb’s observed data, the decay proposed in the new study will be one way to explain it, and pave the way for proof of CP symmetry violation in baryons.

  2023.12.10