++Survey sites recruit respondents with the promise of a reward, which may lead to bogus answers. That doesn’t mean the data is unusable. +
++Search around for ways to make a little extra money online, and you might find yourself at one of many sites that offer to pay you to take surveys. +
++There’s Swagbucks, SurveyJunkie, InboxDollars, and KashKick, for instance. On each of these sites, users are paid small amounts of money for completing surveys, playing games, or making purchases. +
++The surveys on these sites are “opt-in” surveys, meaning that participants are actively choosing to take them, rather than researchers pulling a random sample of a population to poll, as professional pollsters do. +
++Unsurprisingly, opt-in surveys can lead to some skewed results: earlier this week Pew Research Center wrote about their analysis of one such opt-in survey that found 20 percent of US adults under 30 believe that “The Holocaust is a myth.” Pew’s attempt to replicate this result via a random sampling of Americans found that just 3 percent of Americans under 30 agreed with an identically worded statement about the Holocaust — a percentage that was more or less the same across all age groups. +
++The analysis also included this incredible tidbit: +
++“In a February 2022 survey experiment, we asked opt-in respondents if they were licensed to operate a class SSGN (nuclear) submarine. In the opt-in survey, 12% of adults under 30 claimed this qualification, significantly higher than the share among older respondents. In reality, the share of Americans with this type of submarine license rounds to 0%.” +
++Oof, right? +
++The Google results for survey sites are filled with reviews from people who are mainly concerned with whether these sites are “legitimate” or scams. But the Pew analysis points to another question: just how good is the data collected for a survey when its participants are incentivized to speed through as many as possible in order to earn cash? +
++The problems with opt-in surveys, explained +
++I dug around and, surprise! It’s complicated. +
++“Errors are introduced (and remediated) in the survey process at every step,” noted David Rothschild, an economist at Microsoft Research. The fact that a survey was conducted online for a small reward isn’t necessarily enough information to analyze data quality in a meaningful way. +
++As Pew noted in its analysis, the Holocaust denial survey used an agree/disagree format that can lead to “acquiescence bias” — a tendency for respondents to give an affirmative reply. This means that while the survey collection method might have been part of the problem, the question itself may have also led to inaccurate results. +
++“There are many types of opt-in online audiences; some have strong vetting to ensure the respondents are who they say they are and produce high quality responses, while others just accept whomever without any pre-response quality control,” Rothschild added. +
++Here’s what you need to know. +
++How do online survey sites work? +
++Although there are a couple different models, the online survey sites we are talking about offer small rewards in exchange for survey participation. Most say they try to “match” users to relevant surveys based on the data they collect about their users, and generally speaking, you only get paid if you qualify to take the survey and complete each required question. +
++Typically, these sites pay users in points, which translate to small dollar amounts per survey, if they pass a set of screening questions and complete the entire survey. These points often do not translate to very much money: I created an account on Swagbucks and checked a list of available surveys. They included a 20-minute survey for 119 “Swagbucks,” which translates to … $1.19. +
++Longer surveys may offer more, while some surveys with a 10-minute time estimate offer less than a dollar. These are similar to the rates I saw on SurveyJunkie. On Amazon’s Mechanical Turk, a marketplace for work that includes survey taking, a survey might pay less than 10 cents. +
++Why would pollsters and researchers use sites like these to collect responses? +
++In some applications like election polls, as Pew noted, opt-in surveys can perform similarly to random probability-based surveys. Which is great, because they are generally much cheaper to conduct. +
++“Lower cost survey pools are great for exploration” and when you don’t need a very precise outcome, said Rothschild. The results are generally faster, cheaper, and more convenient. +
++“Especially for research that’s being done on a close-to-shoestring budget, opt-in online surveys are a natural choice for scholars trying to study diverse aspects of social behavior,” added Thomas Gift, an associate professor of political science at University College London. +
++Gift and another researcher studied the potential of fraudulent responses in online opt-in studies after using an opt-in study themselves to study a separate question. “It was only during the fielding of the experiment that large cohorts of respondents seemed to be giving suspicious answers about their backgrounds,” he said. “So we investigated further.” +
++Why, and when, are online surveys prone to bogus respondents? +
++Researchers can use a lot of tools, including screening questions, to weed out bad responses and end up with a set of usable data. But there are some instances, such as obscure beliefs or surveys where you need really precise data, where opt-in online surveys are going to be a problem. +
++Pew noted a few considerations here: based on their research over the years, online opt-in polls have a tendency to overestimate fringe beliefs (they gave the example of belief in conspiracy theories). That overrepresentation is more severe among younger respondents, and among Hispanic adults, they noted. +
++Gift and his research partner hired a “nationally-recognized” marketing firm — which they left unnamed in their paper for liability reasons — to conduct a survey for them that collected respondents with experience in the Army. This firm, they said, distributed the survey to a number of sub-vendors that provided financial incentives for responses (these sub-vendors were also left anonymous). +
++In order to detect whether respondents really did have experience in the Army or not, Gift used screening questions embedded in the survey. Respondents were asked about saluting protocol, and for specific information on their military background. +
++Based on their analysis of those screeners, nearly 82 percent of respondents may have pretended to be associated with the Army in order to take the survey and get paid for it. About 36 percent of those respondents passed the knowledge screening test, but were identified as probably misrepresenting themselves based on their answers to the survey questions themselves. +
++And there was also evidence in the survey results that some respondents were taking the survey a bunch of times, giving nearly identical answers and tweaking their demographic data enough to pass as different people, presumably to get paid multiple times for the same survey. +
++How can researchers minimize bogus responses and end up with useful data from an online survey? +
++Essentially, by testing the respondent. Online surveys use attention checks, IP tracking, anti-bot software, and monitoring the time it takes for someone to complete a survey in order to try to mitigate fraud. Asking respondents questions like the one Pew flagged about having a license to drive a submarine is a pretty good way to tell whether someone is just cruising through and answering questions as quickly as possible, or if they’re actually reading the questions. +
++Nothing is going to catch every single bogus response, and, as Rothschild noted, some low-quality responses will slip through attention checks. +
++There are also other models for collecting data online, Gift noted. Opt-in volunteer surveys “aren’t without their limitations,” but they create a different set of incentives for participants that don’t rely on a financial reward. Gift highlighted the work of the Harvard Digital Lab for Social Sciences, an online platform that allows people to volunteer to participate in social science research. +
++While researchers might not be able to catch every single bad response, they can be transparent about how they collected their data, Rothschild noted. And it’s worth looking for that information the next time you see a shocking headline about a shocking belief held by The Youth. +
++The battle proves that time is political, any way you cut it. +
++Scientists dealt a resounding blow this week in a long-running fight over one big question: Have humans messed up the Earth so badly that we’re now living in a new climate epoch? +
++For 15 years, an intrepid band of geologists has been trying to argue exactly that. They claimed that humanity has ushered in the Anthropocene, a new chapter in the Earth’s history borne of our impact on the planet. And they hunted all around the globe for proof. +
++But it’s not easy to make the case that we’re in a new epoch. That’s a technical term that describes a chunk of time typically lasting a few million years (sounds like a lot, but it’s nothing compared to a geological “period” like the 54-million-year Jurassic or “era” like the 186-million-year Mesozoic). Scientists have to vote on whether the term “epoch” applies. +
++Now, a top body of Earth’s professional timekeepers has voted — against canonizing the Anthropocene. +
++And while this might just seem like a smackdown over semantics, the fight over “Anthropocene” is much more. It’s a deeply political fight over how to make meaning of what we humans are doing to the planet. +
++The case for the Anthropocene +
++Earth has gone through distinct geological epochs, chunks of time defined by changes in rock layers. To prove that the Anthropocene represents a new chunk, a group of geologists had to find a “golden spike” — a physical site where the rock, sediment, or ice clearly records the change from a previous chapter in time to a new one. +
++In 2009, they started scouring the planet and found a range of strong candidates, from a peat bog in Poland to a coral reef in Australia to the ice of Antarctica. +
+ ++But the Anthropocene Working Group (AWG), as the group was called, wanted to pick a site where the rock record indisputably shows that we’ve left behind the Holocene epoch, which started 11,700 years ago when the last ice age ended. +
++In 2023, the geologists said they’d found their holy grail: little Crawford Lake in Ontario, Canada. +
++There, the waters are so deep that whatever sinks to the floor usually remains without mixing with the upper layers of water, so it stays preserved, offering an amazingly good record of geological change. +
++Since the middle of the 20th century, the sediment there has been inundated by the byproducts of human activity: plutonium isotopes from the nuclear bombs we’ve detonated, ash from the fossil fuels we’ve burned, and nitrogen from the fertilizer we’ve used. +
++That was also when we started to see major changes in phenomena like global warming, sea-level rise, ocean acidification, and the explosive growth of domestic animal populations. So the AWG said the Anthropocene began around 1950. +
++That would prove controversial. +
++A golden spike, but not a silver bullet +
++Some scientists argued that it doesn’t make sense to recognize our current interval as its own epoch, since it’s incredibly brief in geological time. +
++If the previous epoch, the Holocene, lasted 11,700 years, does it make sense to give the same designation to an interval that hasn’t yet spanned 75 years? +
++But even among those who agreed that human activity had ushered in a new epoch, there was disagreement over when the epoch started. +
+/cdn.vox-cdn.com/uploads/chorus_asset/file/25321983/GettyImages_1935927165__1_.jpg)
Bonnie Jo Mount/The Washington Post via Getty Im
++Paul Crutzen, the atmospheric chemist who originally coined the term “Anthropocene,” said it started in the late 18th century thanks to the greenhouse gas emissions that took off with the Industrial Revolution. Others looked further back to the colonial powers that ravaged the so-called New World. Still others said humans have been transforming the planet since the dawn of agriculture, so trying to pinpoint any later starting point would be arbitrary. +
++Erle Ellis, an ecologist who’d been part of the AWG for 14 years, objected so strongly to its idea of drawing a bright line between pre- and post-1950 that he ultimately resigned. +
++Carving up time that way “does real damage by denying the deeper history and the ultimate causes of Earth’s unfolding social-environmental crisis,” Ellis wrote in his resignation letter. +
++“Are the planetary changes wrought by industrial and colonial nations before 1950 not significant enough to transform the planet? The political ramifications of such a misleading and scientifically inaccurate portrayal are clearly profound and regressive.” +
++Time is political, any way you cut it +
++The thing is, carving up time is inherently political, because scientists are not the only ones who use geological labels. The public uses them too. They feature in our school textbooks, our museum exhibits, and even our music. +
++The term “Anthropocene” is already widely used and understood — in 2020, the musician Grimes even released an album dubbed Miss Anthropocene. The term has become a way to get people to take climate change more seriously. +
++While some scientists were uncomfortable with the idea of using the “Anthropocene” label to make a political statement about what humanity is doing to the planet, other scholars embraced that. +
++The geologist Emlyn Koster, for example, told the New York Times in 2022 that geologists shouldn’t think of defining the Anthropocene as solely the AWG’s business. “I always saw it not as an internal geological undertaking,” he said, “but rather one that could be greatly beneficial to the world at large.” +
++Now that the Subcommission on Quaternary Stratigraphy, the body in charge of recognizing geological time units, has rejected “Anthropocene” as a new epoch, some scientists are at pains to emphasize that humans are still screwing up badly. +
++“We are in the Anthropocene, irrespective of a line on the time scale,” said Francine McCarthy, an earth scientist at Canada’s Brock University who participated in the AWG. “And behaving accordingly is our only path forward.” +
++This story appeared originally in Today, Explained, Vox’s flagship daily newsletter. Sign up here for future editions. +
++
++The company you might not have heard of is now worth $2 trillion — more than Google or Amazon. +
++Only four companies in the world are worth over $2 trillion. Apple, Microsoft, the oil company Saudi Aramco — and, as of 2024, Nvidia. It’s understandable if the name doesn’t ring a bell. The company doesn’t exactly make a shiny product attached to your hand all day, every day, as Apple does. Nvidia designs a chip hidden deep inside the complicated innards of a computer, a seemingly niche product more are relying on every day. +
++Rewind the clock back to 2019, and Nvidia’s market value was hovering around $100 billion. Its incredible speedrun to 20 times that already enviable size was really enabled by one thing — the AI craze. Nvidia is arguably the biggest winner in the AI industry. ChatGPT-maker OpenAI, which catapulted this obsession into the mainstream, is currently worth around $80 billion, and according to market research firm Grand View Research, the entire global AI market was worth a bit under $200 billion in 2023. Both are just a paltry fraction of Nvidia’s value. With all eyes on the company’s jaw-dropping evolution, the real question now is whether Nvidia can hold on to its lofty perch — but here’s how the company got to this level. +
++From games to crypto mining to AI +
++In 1993, long before uncanny AI-generated art and amusing AI chatbot convos took over our social media feeds, three Silicon Valley electrical engineers launched a startup that would focus on an exciting, fast-growing segment of personal computing: video games. +
++Nvidia was founded to design a specific kind of chip called a graphics card — also commonly called a GPU (graphics processing unit) — that enables the output of fancy 3D visuals on the computer screen. The better the graphics card, the more quickly high-quality visuals can be rendered, which is important for things like playing games and video editing. In the prospectus filed ahead of its initial public offering in 1999, Nvidia noted that its future success would depend on the continued growth of computer applications relying on 3D graphics. For most of Nvidia’s existence, game graphics were Nvidia’s raison d’etre. +
++Ben Bajarin, CEO and principal analyst at the tech industry research firm Creative Strategies, acknowledged that Nvidia had been “relatively isolated to a niche part of computing in the market” until recently. +
++Nvidia became a powerhouse selling cards for video games — now an entertainment industry juggernaut making over $180 billion in revenue last year — but it realized it would be smart to branch out from just making graphics cards for games. Not all its experiments panned out. Over a decade ago, Nvidia made a failed gambit to become a major player in the mobile chip market, but today Android phones use a range of non-Nvidia chips, while iPhones use Apple-designed ones. +
++Another play, though, not only paid off, it became the reason we’re all talking about Nvidia today. In 2006, the company released a programming language called CUDA that, in short, unleashed the power of its graphics cards for more general computing processes. Its chips could now do a lot of heavy lifting for tasks unrelated to pumping out pretty game graphics, and it turned out that graphics cards could multitask even better than the CPU (central processing unit), what’s often called the central “brain” of a computer. This made Nvidia’s GPUs great for calculation-heavy tasks like machine learning (and, crypto mining). 2006 was the same year Amazon launched its cloud computing business; Nvidia’s push into general computing was coming at a time when massive data centers were popping up around the world. +
++That Nvidia is a powerhouse today is especially notable because for most of Silicon Valley’s history, there already was a chip-making goliath: Intel. Intel makes both CPUs and GPUs, as well as other products, and it manufactures its own semiconductors — but after a series of missteps, including not investing into the development of AI chips soon enough, the rival chipmaker’s preeminence has somewhat faded. In 2019, when Nvidia’s market value was just over the $100 billion mark, Intel’s value was double that; now Nvidia has joined the ranks of tech titans designated the “Magnificent Seven”, a cabal of tech stocks with a combined value that exceeds the entire stock market of many rich G20 countries. +
++“Their competitors were asleep at the wheel,” says Gil Luria, a senior analyst at the financial firm D.A. Davidson Companies. “Nvidia has long talked about the fact that GPUs are a superior technology for handling accelerated computing.” +
++Today, Nvidia’s four main markets are gaming, professional visualization (like 3D design), data centers, and the automotive industry, as it provides chips that train self-driving technology. A few years ago, its gaming market was still the biggest chunk of revenue at about $5.5 billion, compared to its data center segment, which raked in about $2.9 billion. Then the pandemic broke out. People were spending a lot more time at home, and demand for computer parts, including GPUs, shot up — gaming revenue for the company in fiscal year 2021 jumped a whopping 41 percent. But there were already signs of the coming AI wave, too, as Nvidia’s data center revenue soared by an even more impressive 124 percent. In 2023, its revenue was 400 percent higher than the year before. In a clear display of how quickly the AI race ramped up, data centers have overtaken games, even in a gaming boom. +
++When it went public in 1999, Nvidia had 250 employees. Now it has over 27,000. Jensen Huang, Nvidia’s CEO and one of its founders, has a personal net worth that currently hovers around $70 billion, an over 1,700 percent increase since 2019. +
++It’s likely you’ve already brushed up against Nvidia’s products, even if you don’t know it. Older gaming consoles like the PlayStation 3 and the original Xbox had Nvidia chips, and the current Nintendo Switch uses an Nvidia mobile chip. Many mid- to high-range laptops come packed up with an Nvidia graphics card as well. +
++But with the AI bull rush, the company promises to become more central to the tech people use every day. Tesla cars’ self-driving feature utilizes Nvidia chips, as do practically all major tech companies’ cloud computing services. These services serve as a backbone for so much of our daily internet routines, whether it’s streaming content on Netflix or using office and productivity apps. To train ChatGPT, OpenAI harnessed tens of thousands of Nvidia’s AI chips together. People underestimate how much they use AI on a daily basis, because we don’t realize that some of the automated tasks we rely on have been boosted by AI. Popular apps and social media platforms are adding new AI features seemingly every day: TikTok, Instagram, X (formerly Twitter), even Pinterest all boast some kind of AI functionality to toy with. Slack, a messaging platform that many workplaces use, recently rolled out the ability to use AI to generate thread summaries and recaps of Slack channels. +
++Nvidia’s chips continue to sell out — for now +
++For Nvidia’s customers, the problem with sizzling demand is that the company can charge eye-wateringly high prices. The chips used for AI data centers cost tens of thousands of dollars, with the top-of-the-line product sometimes selling for over $40,000 on sites like Amazon and eBay. Last year, some clients clamoring for Nvidia’s AI chips were waiting as much as 11 months. +
++Just think of Nvidia as the Birkin bag of AI chips. A comparable offering from another chipmaker, AMD, is reportedly being sold to customers like Microsoft for about $10,000 to $15,000, just a fraction of what Nvidia charges. It’s not just the AI chips, either. Nvidia’s gaming business continues to boom, and the price gap between its high-end gaming card and a similarly performing one from AMD has been growing wider. In its last financial quarter, Nvidia reported a gross margin of 76 percent. As in, it cost them just 24 cents to make a dollar in sales. AMD’s most recent gross margin was only 47 percent. +
++Nvidia’s fans argue that its yawning lead was earned by making an early bet that AI would take over the world — its chips are worth the price because of its superior software, and because so much of AI infrastructure has already been built around Nvidia’s products. But Erik Peinert, a research manager and editor at the American Economic Liberties Project who helped put together a recent report on competition within the chip industry, notes that Nvidia has gotten a price boost because TSMC, the biggest semiconductor maker in the world, has struggled for years to keep up with demand. A recent Wall Street Journal report also suggested that the company may be throwing its weight around to maintain dominance; the CEO of an AI chip startup called Groq claimed that customers were scared Nvidia would punish them with order delays if it got wind they were meeting with other chip makers. +
++It’s undeniable that Nvidia put in the investment into courting the AI industry well before others started paying attention, but its grip on the market isn’t unshakable. An army of competitors are on the march, ranging from smaller startups to deep-pocketed opponents, including Amazon, Meta, Microsoft, and Google, all of which currently use Nvidia chips. “The biggest challenge for Nvidia is that their customers want to compete with them,” says Luria. +
++It’s not just that their customers want to make some of the money that Nvidia has been raking in — it’s that they can’t afford to keep paying so much. Microsoft “went from spending less than 10 percent of their capital expenditure on Nvidia to spending nearly 40 percent,” Luria says. “That’s not sustainable.” +
++The fact that over 70 percent of AI chips are bought from Nvidia is also cause for concern for antitrust regulators around the world — the EU recently started looking into the industry for potential antitrust abuses. When Nvidia announced in late 2020 that it wanted to spend an eye-popping $40 billion to buy Arm Limited, a company that designs a chip architecture that most modern smartphones and newer Apple computers use, the FTC blocked the deal. “That acquisition was pretty clearly intended to get control over a software architecture that most of the industry relied on,” says Peinert. “The fact that they have so much pricing power, and that they’re not facing any effective competition, is a real concern.” +
++Will the AI love affair cool off? +
++Whether Nvidia will sustain itself as a $2 trillion company — or rise to even greater heights — depends, fundamentally, on whether both consumer and investor attention on AI can be sustained. Silicon Valley is awash with newly founded AI companies, but what percentage of them will take off, and how long will funders keep pouring money into them? +
++Widespread AI awareness came about because ChatGPT was an easy-to-use — or at least easy-to-show-off-on-social-media — novelty for the general public to get excited about. But a lot of AI work is still focusing on AI training rather than what’s called AI inferencing, which involves using trained AI models to solve a task, like the way that ChatGPT answers a user’s query or facial recognition tech identifies people. Though the AI inference market is growing (and maybe growing faster than expected), much of the sector is still going to be spending a lot more time — and money — on training. For training, Nvidia’s first-class chips will likely remain the most coveted, at least for a while. But once AI inferencing explodes, there will be less of a need for such high-performance chips, and Nvidia’s primacy could slip. +
++Some financial analysts and industry experts have expressed wariness over Nvidia’s stratospheric valuation, suspecting that AI enthusiasm will slow down and that there may already be too much money going toward making AI chips. Traffic to ChatGPT has dropped off since last May and some investors are slowing down the money hose. +
++“Every big technology goes through an adoption cycle,” says Luria. “As it comes into consciousness, you build this huge hype. Then at some point, the hype gets too big, and then you get past it and get into the trough of disillusionment.” He expects to see that soon with AI — though that doesn’t mean it’s a bubble. +
++Nvidia’s revenue last year was about $60 billion, which was a 126 percent increase from the prior year. Its high valuation and stock price is based not just on that revenue, though, but for its predicted continued growth — for comparison, Amazon currently has a lower market value than Nvidia yet made almost $575 billion in sales last year. The path to Nvidia booking large enough profits to justify the $2 trillion valuation looks steep to some experts, especially knowing that the competition is kicking into high gear. +
++There’s also the possibility that Nvidia could be stymied by how fast microchip technology can advance. It has moved at a blistering pace in the last several decades, but there are signs that the pace at which more transistors can be fitted onto a microchip — making them smaller and more powerful — is slowing down. Whether Nvidia can keep offering meaningful hardware and software improvements that convince its customers to buy its latest AI chips could be a challenge, says Bajarin. +
++Yet, for all these possible obstacles, if one were to bet whether Nvidia will soon become as familiar a tech company as Apple and Google, the safe answer is yes. AI fever is why Nvidia is in the rarefied club of trillion-dollar companies — but it may be just as true to say that AI is so big because of Nvidia. +
++