We’re just over a year since it burst onto the scene and OpenAI’s ChatGPT program is somehow even more everywhere than it was in February. Our capability to regulate generative AI and mitigate its myriad real-world harms, on the other hand, continues to lag far behind the technology’s state-of-the-art. That makes 2024 a potentially pivotal year for generative AI in particular and machine learning in genera. ill AI continue to prove itself a fundamental revolution in human-computer communication, on par with the introduction of the mouse in 1963?, Or are we instead heading down yet another overhyped technological dead-end like 3D televisions? Let’s take a look at how OpenAI and its chatbot have impacted consumer electronics in 2023 and where they might lead the industry in the new year.

OpenAI had a great year, all things considered

“Meteoric” doesn’t do justice to OpenAI’s rise this year. The company released ChatGPT on November 30, 2022. Within five days, the program had passed 1 million users; by January, 100 million people a month were logging on to use it. It took Facebook four and a half years to reach those sorts of engagement numbers. ChatGPT outpaced the launches of both TikTok and Instagram to become the most quickly adopted program in the history of the internet in 2023. Heading into 2024, OpenAI (with billions in financial backing from Microsoft) stands at the forefront of the generative AI industry — whether the company can stay there, while billions more are being poured into its rivals’ R&D coffers, remains to be seen.

The company’s sudden success this year also launched its CEO Sam Altman into the media spotlight, with the 38-year-old former head of Y-Combinator basking in much of the praise formerly heaped upon Elon Musk. For a while, Altman was everywhere, repeatedly making appearances before Congressional committees and attending the Senate’s AI Safety Summits. He also conducted a 16-city world tour to Israel, India, Japan, Nigeria, South Korea, across Europe and to the UAE to help promote ChatGPT to developers and policy makers.

Even his termination at the hands of OpenAI’s board of directors in November ended up being a net positive. Fired on a Friday, Altman’s ouster set off 72 hours of panic in Silicon Valley with multiple OpenAI leaders resigning in solidarity, some 95 percent of rank and file staff threatening to walk without his reinstatement, the installation and removal of two interim CEOs in as many days and, ultimately, an indirect intervention by Microsoft. In the end, Altman is still CEO of OpenAI, now with a more compliant and agreeable board, and the tacit understanding throughout the industry that if you strike him down, Sam Altman will become more powerful than you can possibly imagine.

Keeping pace proved a challenge for OpenAI’s competition

A significant contributor to ChatGPT’s immediate and overwhelming success is that it was the first AI of its kind to market. Image generators like DALL-E and Midjourney were already popular diversions, and the public had long acclimated to more mundane machine learning tasks like language translation, but OpenAI was the first with a generative AI program that conversed naturally with its user. That novelty proved an invaluable advantage as even tech titans like Google and Amazon with their massive R&D budgets were caught unprepared for such demand and were slow to respond with competing products of their own.

Google was the most ignoble example of such imitators this year. Following ChatGPT’s debut, Google dedicated the vast majority of its I/O Developers Conference in March to a raft of brand new generative AI models and platforms, including the debut of the Google Bard chatbot. Bard was Google’s answer to ChatGPT, just not a particularly reliable one to start. Even before its public release, Bard made an embarrassing first impression when in February it confidently recited incorrect information about the James Webb Space Telescope in a Twitter ad.

Throughout the year, Google steadily added features, capabilities and access to Bard, eventually shunting the entire platform in December to its newly released foundational model, Gemini, which had been billed as Google’s “most capable and general model” built to date. Google was, of course, then immediately caught misrepresenting the system’s capabilities during a video demonstration. Even without once again getting caught in an easily disprovable lie, Gemini’s demo did little to quiet critics of Google’s stilted and frantic response to ChatGPT.

As a recent Bloomberg op-ed points out, yes, Gemini beat out ChatGPT in a majority of the industry’s standard performance benchmarks. However, Google used the as-yet unreleased Gemini Ultra model to earn its scores and the model only bested GPT-4 so by exceedingly narrow margins. GPT-4 came out nearly a year ago and Google’s best effort barely topped it in middle school-level algebra tasks. That’s not a great look from a corporation that boasts research budgets which rival the GDP of small nations.

Bing is doing just fine, thanks for asking. Microsoft dropped $10 billion on OpenAI in January as part of an ongoing multi-year partnership so now Bing — and literally everything else in the MS ecosystem — is being augmented with algorithmic intelligence. If there was one company that had a better 2023 than OpenAI, it’s Microsoft, which is reportedly set to receive 75 percent of all OpenAI’s profit until those invested billions are recouped.

Amazon placed its $4 billion generative AI bet on Anthropic’s Claude LLM, and made significant headway in leveraging the technology for use in its sprawling empire in 2023, from its Echo Frames smart glasses to Alexa with Generative AI to NFL Thursday Night Games. The company introduced its Bedrock foundational model platform (which will offer AI-generated text and images as a cloud service), launched a series of free AI Ready developer courses and an accelerator program to fund genAI startups, debuted generative tools for filling backgrounds and product listings and now offers a standalone image generator AI to rival DALL-E.

"Inside Amazon, every one of our teams is working on building generative AI applications that reinvent and enhance their customers' experience," CEO Andy Jassy said during the company’s Q2 earnings call in August. "But while we will build a number of these applications ourselves, most will be built by other companies, and we're optimistic that the largest number of these will be built on [Amazon Web Services]. Remember, the core of AI is data. People want to bring generative AI models to the data, not the other way around."

We’re still not ready for the age of AI

Even when it's not being used for obviously nefarious purposes like defrauding the elderly and amplifying political misinformation, generative AI technology has proven immensely disruptive to numerous industries and institutions from logistics and manufacturing to education and healthcare. It has been touted as a replacement for humans in professions ranging from medical imaging, computer programming and accounting to journalism and digital visual arts — in many cases, layoffs have been quick to follow.

This year also saw labor strikes by the Writers Guild of America and the Screen Actors Guild, in part, to prevent their works and likenesses from being used to train future AI models. Independent artists, whose intellectual property has been shamelessly scraped by disreputable firms for model training (looking at you, Stability AI), have had far less success in protecting their works — leading some creators to take drastic and damaging countermeasures.

Data privacy has proven a sticking point for AI companies in 2023. A ChatGPT bug found in March had apparently been sharing chat history titles (and potentially payment data). A trio of Samsung employees inadvertently divulged company secrets when they used ChatGPT to summarize the events of a business meeting in April. Microsoft AI researchers accidentally uploaded 38TB of company data to an open access Azure web folder in September, right around the time it was discovered that Google had been unknowingly leaking users’ Bard conversations into its general search results. As recently as November security researchers were finding that even “silly” attacks like telling ChatGPT to repeat the word “poem” ad infinitum would trick the system into revealing personally identifiable information.

The institutional response to these growing issues was tepid to start the year, mostly school districts, government agencies and Fortune 500 companies restricting use of chatbot AIs by their employees (and students). These initial efforts proved largely ineffective, due to the difficulty in actually enforcing them. The federal government's regulatory efforts are expected to have far more teeth.

The Biden White House has made AI regulation a centerpiece of its administration, developing a “blueprint” for its AI Bill of Rights last October, investing millions into new AI R&D centers for the National Science Foundation, wringing development guardrail concessions from leading AI companies and launching an AI Cyber Challenge, among other efforts. The administration’s most ambitious action came in October when the President issued a sweeping executive order establishing broad protections and best practices regarding user privacy, government transparency and public safety in future AI development by federal contractors. The US Senate and House have both been busy as well this year, holding congressional hearings on federal oversight rules for the AI industry, hosting a pair of AI Safety Summits and drafting legislation (which has yet to receive a vote).

Looking ahead to OpenAI’s 2024 and beyond

It’s OpenAI’s lead to lose heading into the new year. CEO Sam Altman holds firmer control over the company than ever, all dissenting voices on the board calling for caution have been silenced and the company is poised to further expand its operations in 2024 as the technology continues its global advance. I expect to see OpenAI’s competitors make a better showing in the new year with Google, Meta and Amazon spending freely on AI research in order to catch up and surpass the GPT platform.

And even though the entire ChatGPT craze got started with individual users, Paul Silverglate, vice chair of Deloitte LLP, sees the largest gains in 2024 coming from enterprise applications. “Expect to see generative AI integrated into enterprise software, giving more knowledge workers the tools they need to work with greater efficiency and make better decisions,” he wrote in a recent release.

A recent study by McKinsey & Company estimates that the current generation of conversational AI systems “have the potential to automate work activities that absorb 60 to 70 percent of employees’ time” thanks to rapid advancements in natural language processing technology with “half of today’s work activities" potentially being automated away from human hands "between 2030 and 2060." That’s a decade sooner than previously estimated.

Sand, salt, iron, copper, oil and lithium — these foundational materials are literally what the modern world is built on. Without sand for glass, say goodbye to our fiber optic internet. No copper means no conductive wiring. And a world without lithium is a world without rechargeable batteries. 

For the final installment of Hitting the Books for 2023, we're bringing you an excerpt from the fantastic Material World: The Six Raw Materials That Shape Modern Civilization by Ed Conway. A finalist for the Financial Times and Schroders Business Book of the Year award, Material World walks readers through the seismic impacts these six substances have had on human civilization throughout history, using a masterful mix of narrative storytelling and clear-eyed technical explanation. In the excerpt below, Conway discusses how the lithium ion battery technology that is currently powering the EV revolution came into existence.  

Thanks very much for reading Hitting the Books this year, we'll be back with more of the best excerpts from new and upcoming technology titles in post-CES January, 2024!  

Penguin Random House

Excerpted from Material World: The Six Raw Materials That Shape Modern Civilization by Ed Conway. Published by Knopf. Copyright © 2023 by Ed Conway. All rights reserved.

A Better Battery

The first engineer to use lithium in a battery was none other than Thomas Edison. Having mastered the manufacture of concrete by focusing religiously on improving the recipe and systematising its production, he sought to do much the same thing with batteries. The use of these devices to store energy was not especially new even when he began working on them at the dawn of the twentieth century. Indeed, the very earliest days of the electrical era were powered almost exclusively by batteries. Back before the invention of the dynamos and generators that produce most of our electricity today, the telegraphs and earliest electric lights ran on primitive batteries.

Their chemistry went back to Alessandro Volta, an Italian who, at the turn of the nineteenth century, had discovered that by stacking layers of zinc and copper discs separated by cardboard soaked in brine, he could generate an electric current, flowing from one electrode (in this case the metallic discs) to the other. His pile of electrodes was the world’s first battery — a voltaic cell — or as it’s still sometimes called, a pile (since a pile is precisely what it was). That brings us to the prickly question of what to call these things. Purists would argue that a single one of these units, whether it was Volta’s first effort or the thing you find in your smartphone, should be called a cell. A battery, they say, is a word only to be used about an array of multiple cells. But these days most people (including this author) use the words interchangeably.

Half a century later the French physicist Gaston Planté came up with the first rechargeable battery using a spiral of lead electrodes bathed in acid, housed in a glass container. Lead-acid batteries, versions of which are still used to help start car engines today, could provide quick bursts of power, but their relatively low energy density meant they were not especially good at storing power.

In an effort to improve on the chemistry, Edison began to experiment his way through the periodic table. Out went lead and sulphuric acid and in came a host of other ingredients: copper, cobalt and cadmium to name just a few of the Cs. There were many false starts and one major patent battle along the way but eventually, after a decade of experimentation, Edison landed upon a complex mixture of nickel and iron, bathed in a potassium hydroxide solution and packed into the best Swedish steel. 

“The only Storage Battery that has iron and steel in its construction and elements,” read the advertising.

Edison’s experiments underlined at least one thing. While battery chemistry was difficult, it was certainly possible to improve on Planté’s lead–acid formula. After all, as Edison once said, “If Nature had intended to use lead in batteries for powering vehicles she would not have made it so heavy.” And if lead was a heavy metal then there was no doubt about the lightest metal of all — the optimal element to go into batteries. It was there at the opposite end of the periodic table, all the way across from lead, just beneath hydrogen and helium: lithium. Edison added a sprinkling of lithium hydroxide to the electrolyte solution in his battery, the so-called A cell, and, alongside the potassium in the liquid and the nickel and iron electrodes, it had encouraging results. The lithium lifted the battery’s capacity by 10 per cent — though no one could pin down the chemistry going on beneath the surface.

In the following years, scientists followed in Edison’s footsteps and developed other battery chemistries, including nickel–cadmium and nickel–metal hydride, which are the basis for most consumer rechargeable batteries such as the AA ones you might have at home. However, they struggled to incorporate the most promising element of all. Decade after decade, scientific paper after paper pointed out that the ultimate battery would be based on a lithium chemistry. But up until the 1970s no one was able to tame this volatile substance enough to put it to use in a battery. Batteries are a form of fuel — albeit electrochemical rather than fossil. What occurs inside a battery is a controlled chemical reaction, an effort to channel the explosive energy contained in these materials and turn that into an electric current. And no ingredient was more explosive than lithium.

The first breakthrough came in the 1970s at, of all places, Exxon-Mobil, or as it was then known, Esso. In the face of the oil price shock, for a period the oil giant had one of the best-funded battery units anywhere, staffed by some of the world’s most talented chemists trying to map out the company’s future in a world without hydrocarbons. Among them was a softly spoken Englishman called Stan Whittingham. Soon enough Whittingham had one of those Eureka moments that changed the battery world forever.

Up until then, one of the main problems facing battery makers was that every time they charged or discharged their batteries it could change the chemical structure of their electrodes irreversibly. Edison had spent years attempting to surmount this phenomenon, whose practical consequence was that batteries simply didn’t last all that long. Whittingham worked out how to overcome this, shuttling lithium atoms from one electrode to the other without causing much damage.

At the risk of causing any battery chemists reading this to wince, here is one helpful way of visualising this. Think of batteries as containing a set of two skyscrapers, one of which is an office block and the other is an apartment block. These towers represent the anode and cathode — the negative and positive electrodes. When a rechargeable smartphone or electric car battery is empty, what that means in electrochemical terms is that there are a lot of lithium atoms sitting in the cathode — in the apartment block — doing very little.

But when that battery gets charged, those atoms (or, as they’re technically called, since they hold a charge, ions) shuttle across to the other skyscraper — the anode or, in this analogy, the office block. They go to work. And a fully charged battery is one where the anode’s structure is chock-full of these charged lithium ions. When that battery is being used, the ions are shuttling back home to the apartment block, generating a current along the way.

Understand this shuttling to and fro between cathode and anode and you understand broadly how rechargeable batteries work. This concept — the notion that ions could travel across from the crystalline structure of one electrode to nest in the crystalline structure of another — was Whittingham’s brainwave. He called it intercalation, and it’s still the basis of how batteries work today. Whittingham put the theory to work and created the world’s first rechargeable lithium battery. It was only a small thing — a coin-sized battery designed for use in watches — but it was a start. Per kilogram of weight (or rather, given its size, per gram), his battery could hold as much as 15 times the electrical charge of a lead–acid battery. But every time Whittingham tried to make a battery any bigger than a small coin cell, it would burst into flames. In an effort to tame the inherent reactivity of lithium, he had alloyed it with aluminium, but this wasn’t enough to subdue it altogether. So Whittingham’s battery remained something of a curio until the following decade, when researchers working in the UK and Japan finally cracked the code.

The key figure here is an extraordinary man called John B. Goodenough, an American physicist who, as it happens, was born in Jena, the German city where Otto Schott and Carl Zeiss first perfected technical glassmaking. After studying at Yale, Chicago and the Massachusetts Institute of Technology, Goodenough eventually found himself in charge of the inorganic chemistry lab at the University of Oxford in the late 1970s and early 1980s, where he played the pivotal role in the battery breakthrough. Among his team’s achievements — commemorated today in a blue plaque on the outside of the lab — was the discovery of the optimal recipe for the cathode (that apartment skyscraper) in a lithium-ion battery. The material in question was lithium cobalt oxide, a compound that improved the safety and the capacity of these batteries, providing them with a stable cathode matrix in which the lithium ions could nest. It wasn’t that battery explosions could be ruled out, but at least they were no longer inevitable.

The final intellectual leaps occurred a few years later in Japan, where a researcher called Akira Yoshino perfected the other ingredients. He paired Goodenough’s lithium cobalt oxide cathode with an anode made from a particular type of graphite — that very variety they still make from the needle coke produced at the Humber Refinery — and the combination worked brilliantly. Lithium ions shuttled safely and smoothly from one side to another as he charged and discharged the battery. He also worked out the best way to fit these two electrodes together: by pasting the materials on to paper-thin sheets and coiling them together in a metal canister, separated by a thin membrane. This final masterstroke — which meant that if the battery began to overheat the separator would melt, helping to prevent any explosion — also evoked those first cells created in France by Gaston Planté. The rechargeable battery began life as a spiral of metal compressed into a canister; after more than a century of experimentation and a complete transformation of materials, it came of age in more or less the same form.

But it would take another few years for these batteries to find their way into consumers’ hands, and it would happen a long way from either Esso’s laboratories or Oxford’s chemistry labs. Japanese electronics firm Sony had been on the lookout for a better battery to power its camcorders, and came across the blueprints drawn up by Goodenough and adjusted by Yoshino. Adapting these plans and adding its own flourishes, in 1992 it created the first production lithium-ion battery: an optional power pack for some of their Handycam models. These packs were a third smaller and lighter than the standard nickel–metal hydride batteries, yet they carried even more capacity. In the following years, lithium-ion batteries gradually proliferated into all sorts of devices, but it wasn’t until the advent of the smartphone that they found their first true calling. These devices, with their circuitry, their semiconductors, their modem chips and bright displays, are incredibly power hungry, demanding the most powerful of all batteries. Today, almost all smartphones run on batteries derived from the discoveries of Whittingham, Goodenough and Yoshino. The trio was awarded the Nobel Prize in Chemistry in 2019.

That this invention — first prototyped in America and then mostly developed in England — only came to be mass produced in Japan is one of those topics that still causes frustration in the Anglophone world. Why, when so many of the intellectual advances in battery design happened in Europe and the Americas, was production always dominated by Asia? The short answer was that Japan had a burgeoning market for the manufacture of the very electronic goods — initially video cameras and Walkmans — that needed higher-density batteries.

As the 1990s gave way to the 2000s, lithium-ion batteries became an essential component of the electronic world, in laptops, smartphones and, eventually, electric cars. Smartphones could not have happened without the extraordinary silicon chips inside, powering the circuitry, housing the processing units and bestowing memory storage, not to mention providing optical sensors for the camera. But none of these appliances would have been practical without light, powerful batteries of far greater energy density than their predecessors.

All of which is why demand for lithium has begun to outstrip our ability to extract it from the earth. And unlike copper or iron, which we have many centuries’ experience producing, the lithium industry remains in its infancy. Up until recently there were few mines and the pools in the Salar de Atacama were still relatively small. Today they are big enough to be easily visible from space, a gigantic pastel paint palette smack bang in the middle of the desert.

The year had started so well for robotaxis. Cruise and Waymo came into 2023 riding high on fresh investments from General Motors and Google, respectively, as well as rapidly growing interest from the general public and a downright rabid rate of adoption by city governments. Things were looking up, very up, for the burgeoning self-driving vehicle industry! Then a driverless Crusie taxi accidentally dragged a hit-and-run victim down a San Francisco street for a few dozen feet and everything just sort of went to shit from there. So fragile, these Next Big Things. Let’s take a look back through the year that was to see how autonomous taxi tech might recover from this catastrophe.

Cruise (Out of) Control

Cruise came into this year looking like a nigh-on unstoppable force of transportational change as the core of GM's self-driving efforts. The company received a $1.5 billion investment from the automaker in March 2022 after GM spent $2.1 buying equity ownership for the startup from Softbank Vision Fund. In February the company announced that its test fleet of driverless taxis had traveled a million miles of San Francisco’s streets without a human behind the wheel. The program had only started the previous November.

"When you consider our safety record, the gravity of our team’s achievement comes into sharper focus," Mo Elshenawy, Cruise's EVP of engineering, said in February. "To date, riders have taken tens of thousands of rides in Cruise AVs. In the coming years, millions of people will experience this fully driverless future for themselves."

Cruise CEO Kyle Vogt had been installed at his position in December 2021 after GM CEO Mary Barra ousted Dan Ammann from the spot. Vogt spent the following year laying out a grand vision of “zero crashes, zero traffic, and zero emissions,” though, according to a November report from the New York Times, the company “put a priority on the speed of the program over safety” during his tenure, cutting corners on safety in order to get more vehicles on the road. And expand Cruise did, into Houston and Los Angeles most notably, despite a growing number of traffic incidents and accidents left behind by its vehicles.

In April, the company was given permission to operate its driverless vehicles throughout San Francisco, 24/7 as well as pick up paying passengers during daylight hours. Previously, only Cruise employees were allowed to ride in the robotaxis and they could only operate when the sun was out. In August, the California Public Utilities Commission (CPUC) voted 3-to-1 in favor of allowing Cruise (and Waymo as well) to to pick up paying passengers at all hours.

Not everybody was fully on board with the robotaxi takeover, mind you. In January 2023, San Francisco officials requested the CPUC slow or even halt the expansion of self-driving vehicle services in the city, arguing that the free-for-all growth OK’d by state regulators was becoming an “unreasonable” burden. In fact, barely a week after the CPUC voted in favor of expansion, the California DMV opened an investigation into an altercation between a Cruise taxi and a fire truck. In response, the DMV had Cruise cut its operating fleet in half — down to 50 vehicles during daylight hours and 150 at night — until it had completed its investigation. Then there was the whole “using robotaxis as love hotels” issue in August.

Those mishaps were bad. The events of October 3 and Cruise’s response to the resulting investigation proved unforgivable. As the company initially explained in the above thread, a human-driven vehicle struck a pedestrian, pushing her into the path of the Cruise taxi in the lane to her right. The taxi ran the woman over, despite aggressively braking, and ended up dragging her 20 feet until coming to a stop. EMS crews were able to extract the pedestrian from underneath the taxi using the jaws of life, and rushed her to medical treatment with critical injuries.Though she has not been identified, the pedestrian was reportedly in serious condition as late as October 25.

If that weren’t bad enough, Cruise then allegedly misled regulators about when the taxi engaged its brakes — telling them that the taxi had stopped immediately, not eventually, after slowly traveling another 20 feet down the block. The company then repeatedly delayed in releasing video of the incident to investigators until October 19.

The company’s cover-up efforts puts Cruise in financial jeopardy with the CPUC, which is currently considering fining it as much as $1.5 million for its obfuscating actions. The Commission's decision will be made in early February at an upcoming evidentiary hearing.

More immediately, the accident itself set off a whole slew of investigations, regulatory and internal alike. The Exponent consulting firm was brought in as an independent investigator and promptly dredged up some rather unflattering data regarding the robotaxis’ difficulties with spotting and reacting to the presence of small children. That revelation wasn’t so bad, at least compared to the company’s decision to keep the vehicles on the road even after being informed of the potentially deadly defect.

The California DMV was not amused and, two weeks after the accident occurred, the department suspended Cruise’s license to operate within the state, effectively shuttering its robotaxi operations. That’s a huge blow to GM, which has sunk billions into the startup and was anticipating the robotaxi service to generate as much as $5 billion annually when operations were to begin in 2025. In mid-November, the company recalled all 950 of its autonomous taxis in operation, and even paused robotaxi rides with human safety drivers behind the wheel a week later, as part of a “full safety review.”

Then things got even worse. On November 18, CEO Kyle Vogt announced his resignation from his position a week after GM installed EVP of Legal and Policy Craig Glidde (who was already a Cruise board member) as Chief Administrative Officer. The following day, company co-founder and Chief Product Officer Daniel Kan also announced his departure.

In response to Vogt's departure, GM promoted Mo Elshenawy from EVP of Engineering to the dual role of President and CTO, leaving the CEO position currently vacant. GM CEO Mary Barra told reporters recently that the company has “a lot of confidence with what the two co-presidents will do,” but will be “leaning in to make sure that it meets our strict requirements from a safety perspective.”

GM suddenly found itself holding the multibillion dollar bag, so it cut off funding near immediately, slashing budgets to the tune of “hundreds of millions” of dollars. As a result, Cruise has since suspended its equity program and begun laying off employees, starting with those in autonomous vehicle operations.

"The most important thing for us right now is to take steps to rebuild public trust," Cruise said in a statement. "Part of this involves taking a hard look inwards and at how we do work at Cruise, even if it means doing things that are uncomfortable or difficult."

But Cruise isn’t entirely dead yet, as Elshenawy explained in a recent email to staff. The company plans to scale back its self-driving ambitions and relaunch with a renewed focus on the current Chevy Bolt AV robotaxi platform, rather than its custom-built Origin vehicle. As such the company is pausing production on the Origin at least through 2024 but does plan to continue the program at some point in the future.

Waymo Money, Waymo Problems

Waymo entered 2023 in much the same way as Cruise did: riding high on the hype and promise of self-driving vehicle technology. However it is ending the year in a very different place from its biggest competitor.

Google-backed Waymo had received glowing praise from Swiss RE, a leading global reinsurer, regarding the safety of its vehicles versus human drivers the previous September, and had just launched its second Waymo One taxi service area that December, this time in Phoenix, Arizona, running a route between downtown and the Phoenix Sky Harbor International Airport.

Los Angeles joined Waymo’s stable of cities in February. Much as it was rolled out in San Francisco, Waymo’s self-driving vehicles were initially made available only to riders who were part of the Waymo Research Trusted Tester program in a limited area (in this case, Santa Monica), always outside of rush hour and only in limited numbers.

The following month the company launched a similar effort in Austin, Texas, a town where it had conducted some of its earliest self-driving tests back in 2015. Austin is a hot town to test self-driving vehicles in, on account of a 2017 state law that prevents cities from locally regulating the technology’s use and deployment on their streets.

Things were going so well for Waymo come summer that the company announced it would shift gears, pushing back plans for its self-driving truck idea to instead focus fully on its expanding robotaxi service.

“Given the tremendous momentum and substantial commercial opportunity we’re seeing on the ride-hailing front, we’ve made the decision to focus our efforts and investment on ride-hailing,” Waymo co-CEOs Tekedra Mawakana and Dmitri Dolgov wrote in a July blog post. "We’re iterating more quickly than ever on our technology by pushing forward state of the art AI/ML, and seeing significant business growth and rider demand in San Francisco, Phoenix, and Los Angeles.”

By August, Waymo announced that Austin would be joining those towns as the fourth city to host its autonomous taxi service, with the program rolling out through the Fall. That same month, Waymo received its driverless deployment permit from the California Public Utilities Commission (CPUC), enabling the company to begin charging passengers for its robotaxi rides as well as expanding the service to additional customers. Previously, the company could only charge for rides if a human safety driver was behind the wheel. The company acknowledged at the time that demand was “incredibly high” (signups had already reportedly passed 100,000 users) but that it was working to make its fully autonomous trips "available to everyone over time."

“Things are growing… The ridership is increasing in both Phoenix and SF,” he continued, noting that the company provides more than 10,000 trips per city each week. Overall, it would have been a pretty great year for Waymo — especially after chief rival, Cruise, effectively imploded over the course of Q4 — had the company’s workforce not been subject to not one, not two, but three rounds of layoffs impacting over 300 employees.

The Road Ahead for Robotaxis

As we head into the new year, Waymo is effectively the only game in town, now that Cruise isn’t a viable commercial entity for the foreseeable future.

Midway through the year, analysts predicted the robotaxi market, valued at just over $1.1 billion in 2022, could rise to anywhere from $45.7 billion in 2030 to $118 billion in 2031 citing, “increasing demand for shared transportation, advancements in vehicle technology, growing interest in fuel-efficient public transportation, and improved infrastructure.”

Those outlooks have been tempered in recent months, at least for short term estimates, with Cruise temporarily out of the picture. Forrester Analytics, for example, now expects drone delivery services to become the dominant self-driving vehicle segment in 2024 as pushback from regulators slows development of robotaxi transit tech.

“Expect a booming year for self-driving forklifts, curbside delivery robots, and drone delivery, driven by the increasing popularity of e-commerce, the need for last-mile delivery solutions, and more sophisticated autonomous technologies,” wrote Craig Le Clair, Vice President and Principal Analyst at Forrester.

We are, of course, still waiting on those million robotaxis Elon Musk promised us back in 2019.

Company Towns — wherein a single firm provides most or all necessary services, from housing and employment to commerce and amenities to a given community — have dotted America since before the Civil War. As we near the end of the first quarter of the 21st century, they're making a comeback with a new generation of ultra-wealthy elites gobbling up land and looking to build towns in their own image. 

And why should only terrestrial workers be exploited? Elon Musk has long talked of his plans to colonize Mars through his company SpaceX and those plans don't happen without a sizeable — and in this case, notably captive — workforce on hand. The same Elon Musk who spent $44 billion to run a ubiquitous social media site into the ground, whose brain computer interface company can't stop killing monkeys and whose automotive company can't stop killing pedestrians, wants to construct entire settlements wholly reliant on his company's largesse and logistics train. Are we really going to trust the mercurial CEO with people's literal air supplies?

In this week's Hitting the Books, Rice University biologist and podcaster Kelly Weinersmith and her husband Zach (of Saturday Morning Breakfast Cereal fame) examine what it will actually take to put people on the red planet and what unforeseen costs we might have to pay to accomplish such a goal in their new book A City on Mars: Can we settle space, should we settle space, and have we really thought this through?

Penguin Random House

Excerpted from A City on Mars: Can we settle space, should we settle space, and have we really thought this through? by Kelly and Zach Weinersmith. Published by Penguin. Copyright © 2023 by Kelly and Zach Weinersmith. All rights reserved.

On the Care and Feeding of Space Employees

One of the first things to know about company towns is that companies don’t appear to want to be in charge of housing. In our experience, people often think housing was an actively pursued control tactic, but if you look at the available data and the oral histories, companies often seem downright reluctant to supply housing at all. In Dr. Price Fishback’s economic analysis of coal towns in early-twentieth-century Appalachia, Soft Coal, Hard Choices, he found that companies able to have a third party supply housing typically did. This is hard to square with the idea that housing was built specifically with sinister intentions.

There are also good theoretical reasons to explain why companies build housing and rent it out to workers. Suppose Elon Musk is building the space city Muskow. Having wisely consulted the nearest available Weinersmith, he decides he shouldn’t own employee housing due to something or other about the risks of power imbalance. He looks to hire builders, but immediately runs into a problem: very few companies are available for construction on Mars. Let’s consider the simple case where only one company is willing to do it.

Well, guess what. That company now has monopoly power. They can raise home prices or lower home quality, making Muskow less attractive to potential workers. Musk can now only improve the situation by paying workers more, costing him money while lining the pockets of the housing provider.

If he wants to avoid this, Musk’s ideal option is to attract more building companies, so they can compete with each other. If that’s not possible, as was often the case in remote company towns, then the only alternative is to build the housing himself. This works, but the tradeoff is that he’s now managing housing in addition to focusing on his core business. He’s also acquired a lot of control over his employees. None of this setup requires Musk to be a power-hungry bastard — all it requires is that he needs to attract workers to a place where there’s zero competition for housing construction.

Historically, where things get more worrisome is in rental agreements, which often tied housing to employment. Even these can partially be explained as rational choices a non- evil bastard might non- evilly make. Workers in mines were often temporary. Mines were temporary, too, existing only until the resources were no longer profitable. This made homeownership a less compelling prospect for a worker. Why? Two reasons. First, if a town may suddenly fold in fifteen years because a copper mine stops being profitable, buying a house is a bad investment. Second, if you own a home, it’s hard for you to leave. This is a problem because threatening to leave is a classic way to enhance your bargaining position as a worker.

Once you have people whose housing is tied to their job, the potential for abuse is enormous — especially during strikes. Rental agreements were often tied to employment, and so striking or even having an injury could mean the loss of your home. When your boss is also your landlord, their ability to threaten you and your family is tremendous, and indeed narrative accounts refer to eviction of families with children by force. If employees either owned their homes or had more secure rental agreements, power would have run the other way. They could have struck for better wages or conditions and occupied those homes to make it harder for their employer to bring in replacements.

It may be tempting to see this as a purely capitalist problem, but very similar results occurred in Soviet monotown housing. Employees tended to get reasonably nice company-town housing; if they lost their jobs, they had to go to the local Soviet, which provided far worse accommodations. As one author put it, “Thus, housing became the method of controlling workers par excellence.” This suggests that there’s a deep structural dynamic here — when your employer owns your housing, they’re apt to use it against you at some point.

In space, you can’t kick people out of their houses unless you’re prepared to kill them or pay for a pricey trip home. On Mars, orbital mechanics may preclude the trip even if you’re able to afford it. In arguing with space-settlement geeks, housing concerns are often set up as binaries — “Look, they’re not going to kill the employees, so they’ll have to treat them well.” In fact, there’s a spectrum of bastardry available. A company-town boss on Mars could provide lower-quality food, reduce floor space, restrict the flow of beet wine, deny you access to the pregnodrome. They could also tune your atmosphere. We found one account by a British submariner, in which he claimed to adjust the balance of oxygen to carbon dioxide depending on whether he wanted people more lethargic or more active. Whether it’ll be worth the risk of pissing off employees who cost, at least, millions to deliver to the settlement is harder to say.

This overall logic — companies must supply amenities, therefore companies acquire power — repeats across contexts in company towns. To attract skilled employees who may have families, the company must supply housing, yes, but they also must supply other regular town stuff — shopping, entertainment, festivals, sanitation, roads, bridges, municipal planning, schools, temples, churches. When one company controls shopping, they set the prices and they know what you buy. When they control entertainment and worship, they have power over employee speech and behavior. When they control schools, they have power over what is taught. When they control the hospitals, they control who gets health care, and how much.

Even if the company does a decent job on all these fronts, there may still be resistance, basically because people don’t love having so much of their lives controlled by one entity. Fishback argued that company towns, for all their issues, were not as bad as their reputation. In theorizing why, he suggested one problem you might call the omni-antagonist effect. Think about what groups you’re most likely to be angry at during any given moment of adult life. Landlord? Home-repair company? Local stores? Utility companies? Your homeowners association? Local governance? Health-care service? Chances are you’re mad at someone on this list even as you read this book. Now, imagine all are merged into a single entity that is also your boss.

In space, as usual, things are worse: the infrastructure and utility people aren’t just keeping the toilet and electricity running; they’re deciding how much CO2 is in your air and controlling transportation in and out of town. Even if the company is not evil, it’s going to be hard to keep good relations, even at the best of times.

And it will not always be the best of times.

When Company Towns Go Bad

Unionization attempts on September 3, 1921, reporting on the then ongoing miners strike in West Virginia, the Associated Press released the following bulletin:

Sub district President Blizzard of the United Mine Workers . . . says five airplanes sent up from Logan county dropped bombs manufactured of gaspipe and high explosives over the miners’ land, but that no one was injured. One of the bombs, he reports, fell between two women who were standing in a yard, but it failed to explode.

“Failed to explode” is better than the alternative, but well, it’s the thought that counts.

Most strikes were not accompanied by attempted war crimes, but that particular strike, which was part of early-twentieth-century America’s aptly named Coal Wars, happened during a situation associated with increased danger — unionization attempts.

Looked at in strictly economic terms, this isn’t so surprising. From the company’s perspective, beyond unionization lies a huge unknown. Formerly direct decisions will have to run through a new and potentially antagonistic committee. The company will have less flexibility about wages and layoffs in case of an economic downturn. They may become less competitive with a nonunion entity. They may have to renegotiate every single employee contract.

Whether or not a union would be good per se in a space settlement, given how costly and hazardous any kind of strife would be, you may want to begin your space settlement with some sort of collective bargaining entity purely to avoid a dangerous transition. A union would also reduce some of the power imbalance by giving workers the ability to act collectively in their own interest. However, this may not happen in reality if the major space capitalists of today are the space company-town bosses of the future—both Elon Musk and Jeff Bezos kept their companies ununionized while CEOs.

Economic Chaos

Another basic problem here is that company towns, being generally oriented around a single good, are extremely vulnerable to economic randomness. Several scholars have noted that company towns tend to be less prone to strife when they have fatter margins. It’s no coincidence that the pipe-bomb incident above came about during a serious drop in the price of coal early in the twentieth century. Price drops and general bad economic conditions can mean renegotiations of contracts in an environment where the company fears for its survival. Things can get nasty.

If Muskow makes its money on tourism, it might lose out when Apple opens a slightly cooler Mars resort two lava tubes over. Or there could be another Great Depression on Earth, limiting the desire for costly space vacations. So what’s a space CEO to do? In terrestrial company towns, if a Great Depression shows up, one option is for the town to just fold. It’s not a fun option, but at least there’s a train out of town or a chance to hitchhike. Mars has a once-every-two-years launch window.* Even a trip to Earth from the Moon requires a 380,000-kilometer shot in a rocket, which will likely never be cheap.

The biggest rockets on the drawing board today could perhaps transport a hundred people at a time. Even for a settlement of only ten thousand people, that’s a lot of transport infrastructure in case the town needs to be evacuated. Throw in that, at least right now, we don’t even know if people born and raised on the Moon or Mars can physiologically handle coming “back” to Earth, and, well, things get interesting.

The result is that there is a huge ethical onus on whoever’s setting this thing up. Not just to have a huge reserve of funding and supplies and transportation, so that people can be saved or evacuated if need be, but also to do the science in advance to determine if it’s even possible to bring home people born in partial Earth gravity.

There is some precedent for governments being willing to prop up company towns. Many old Soviet monotowns now receive economic aid from the Russian government. We should note, however, that keeping a small Russian village on life support will be a lot cheaper than maintaining an armada of megarockets for supplies and transportation.

The Washington Post reports that after a marathon 72-hour debate European Union legislators Friday have reached a historic deal on a broad-ranging AI safety development bill, the most expansive and far-reaching of its kind to date. Details of the deal itself were not immediately available. 

The proposed regulations would dictate the ways in which future machine learning models can be developed and distributed within the trade bloc, impacting its use in applications ranging from education to employment to healthcare. AI development would be split among four categories, depending on how much societal risk each potentially poses — minimal, limited, high, and banned. 

Banned uses would include anything that circumvents the user's will, targets protected groups or provides real-time biometric tracking (like facial recognition). High risk uses include anything "intended to be used as a safety component of a product,” or are to be used in defined applications like critical infrastructure, education, legal/judicial matters and employee hiring.

“The European Commission once again has stepped out in a bold fashion to address emerging technology, just like they had done with data privacy through the GDPR,” Dr. Brandie Nonnecke, Director of the CITRIS Policy Lab at UC Berkeley, told Engadget in 2021. “The proposed regulation is quite interesting in that it is attacking the problem from a risk-based approach,” similar what's been suggested in Canada’s proposed AI regulatory framework.

The EC had previously addressed the growing challenges of managing emerging AI technologies through an variety of efforts, releasing both the first European Strategy on AI and Coordinated Plan on AI in 2018, followed by the Guidelines for Trustworthy AI in 2019. The following year, the Commission released a White Paper on AI and Report on the safety and liability implications of Artificial Intelligence, the Internet of Things and robotics

”Artificial intelligence should not be an end in itself, but a tool that has to serve people with the ultimate aim of increasing human well-being," the European Commission wrote in its draft AI regulations. "Rules for artificial intelligence available in the Union market or otherwise affecting Union citizens should thus put people at the centre (be human-centric), so that they can trust that the technology is used in a way that is safe and compliant with the law, including the respect of fundamental rights.” 

“At the same time, such rules for artificial intelligence should be balanced, proportionate and not unnecessarily constrain or hinder technological development," it continued. "This is of particular importance because, although artificial intelligence is already present in many aspects of people’s daily lives, it is not possible to anticipate all possible uses or applications thereof that may happen in the future.”

 Developing...

If riding velociraptors and giant mechas wasn't enough for its discerning players, Fortnite is apparently incorporating a semi-standalone racing game into its free-for-all universe. On Thursday at the 2023 Game Awards, Epic revealed first gameplay footage for the new mode — days after releasing the cinematic trailer and mere hours before it's set to go live across the globe! 

Developed by Psyonix, the folks who built Rocket League (itself still an e-sport staple), the Rocket Racing mode operates within the larger Fortnite game and set at the Festive Falls track where racers go head-to-head to compete for the top spot. 

In the trailer shown at the Awards, players raced through the track using directional boosters to launch their cars over obstacles, drift them through turns, and fly through hidden shortcuts. Players will also apparently be able to call items from their inventories to use mid-race, blinking them into existence in their opponents' paths. Rocket Racing is set to officially launch on Friday December 8.

Not content to rest on its honkin' laurels after the success of 2019's indie smash hit Untitled Goose Game, House House unveiled the first teaser footage for its new adventure title at the 2023 Game Awards Thursday night. 

Big Walk looks to be a significant departure from its predecessor in both tone and scope. The team describes it as "a cooperative online walker-talker" that encourages players to "hang out and get lost with close friends in a big world." 

The online coop game is set on an expansive bruch-covered island that's overflowing with mysteries to uncover and challenges to overcome. In the trailer we can see a group of oddly-limbed avatars strolling around a hillside, solving puzzles, exploring strange caves and setting off signal flares. Cooperation will be a big part of the game. Players will need to work together to navigate through the terrain and keep in contact with one another while out and about. 

“Our favourite part of playing online coop games is when they give you enough direction for the group to keep up a good momentum, but it’s relaxed enough that you’re mostly just able to enjoy spending time with your friends. We hope Big Walk gives players that direction, and space, to focus on and enjoy those group dynamics, whether you’re playing with one friend or as a big group,” Jake Strasser, developer at House House, said in a Thursday release. Big Walk is slated to arrive on Steam and the Epic Game Store in 2025.     

OpenAI's spot atop the generative AI heap may be coming to an end as Google officially introduced its most capable large language model to date on Wednesday, dubbed Gemini 1.0. It's the first of “a new generation of AI models, inspired by the way people understand and interact with the world,” CEO Sundar Pichai wrote in a Google blog post.

“Ever since programming AI for computer games as a teenager, and throughout my years as a neuroscience researcher trying to understand the workings of the brain, I’ve always believed that if we could build smarter machines, we could harness them to benefit humanity in incredible ways,” Pichai continued.

The result of extensive collaboration between Google’s DeepMind and Research divisions, Gemini has all the bells and whistles cutting-edge genAIs have to offer. "Its capabilities are state-of-the-art in nearly every domain," Pichai declared. 

The system has been developed from the ground up as an integrated multimodal AI. Many foundational models can be essentially though of groups of smaller models all stacked in a trench coat, with each individual model trained to perform its specific function as a part of the larger whole. That’s all well and good for shallow functions like describing images but not so much for complex reasoning tasks.

Google, conversely, pre-trained and fine-tuned Gemini, “from the start on different modalities” allowing it to “seamlessly understand and reason about all kinds of inputs from the ground up, far better than existing multimodal models,” Pichai said. Being able to take in all these forms of data at once should help Gemini provide better responses on more challenging subjects, like physics.

Gemini can code as well. It’s reportedly proficient in popular programming languages including Python, Java, C++ and Go. Google has even leveraged a specialized version of Gemini to create AlphaCode 2, a successor to last year's competition-winning generativeAI. According to the company, AlphaCode 2 solved twice as many challenge questions as its predecessor did, which would put its performance above an estimated 85 percent of the previous competition’s participants.

While Google did not immediately share the number of parameters that Gemini can utilize, the company did tout the model’s operational flexibility and ability to work in form factors from large data centers to local mobile devices. To accomplish this transformational feat, Gemini is being made available in three sizes: Nano, Pro and Ultra. 

Nano, unsurprisingly, is the smallest of the trio and designed primarily for on-device tasks. Pro is the next step up, a more versatile offering than Nano, and will soon be getting integrated into many of Google’s existing products, including Bard.

Starting Wednesday, Bard will begin using a especially-tuned version of Pro that Google promises will offer “more advanced reasoning, planning, understanding and more.” The improved Bard chatbot will be available in the same 170 countries and territories that regular Bard currently is, and the company reportedly plans to expand the new version's availability as we move through 2024. Next year, with the arrival of Gemini Ultra, Google will also introduce Bard Advanced, an even beefier AI with added features.

Pro’s capabilities will also be accessible via API calls through Google AI Studio or Google Cloud Vertex AI. Search (specifically SGE), Ads, Chrome and Duet AI will also see Gemini functionality integrated into their features in the coming months.

Gemini Ultra won’t be available until at least 2024, as it reportedly requires additional red-team testing before being cleared for release to “select customers, developers, partners and safety and responsibility experts” for testing and feedback.” But when it does arrive, Ultra promises to be an incredibly powerful for further AI development.

After years of production delays, Tesla CEO Elon Musk took to a dimly-lit stage on Thursday to hand-deliver the first batch of Cybertruck EVs to their new owners. The company has also, finally announced pricing for the luxury electric truck. Prospective buyers can expect to pay anywhere from $60,990 to $100,000 MSRP (and potentially $11,000 less after rebates and tax credits). The company has launched an online configurator tool for those interested in placing an order of their own.     

The event caps nearly half a decade of design and development that started in 2019 and which has been subject to intense hype, promotion and scrutiny throughout.

Developing

The Biden White House recently enacted its latest executive order designed to establish a guiding framework for generative artificial intelligence development — including content authentication and using digital watermarks to indicate when digital assets made by the Federal government are computer generated. Here’s how it and similar copy protection technologies might help content creators more securely authenticate their online works in an age of generative AI misinformation.

A quick history of watermarking

Analog watermarking techniques were first developed in Italy in 1282. Papermakers would implant thin wires into the paper mold, which would create almost imperceptibly thinner areas of the sheet which would become apparent when held up to a light. Not only were analog watermarks used to authenticate where and how a company’s products were produced, the marks could also be leveraged to pass concealed, encoded messages. By the 18th century, the technology had spread to government use as a means to prevent currency counterfeiting. Color watermark techniques, which sandwich dyed materials between layers of paper, were developed around the same period.

Though the term “digital watermarking” wasn’t coined until 1992, the technology behind it was first patented by the Muzac Corporation in 1954. The system they built, and which they used until the company was sold in the 1980s, would identify music owned by Muzac using a “notch filter” to block the audio signal at 1 kHz in specific bursts, like Morse Code, to store identification information.

Advertisement monitoring and audience measurement firms like the Nielsen Company have long used watermarking techniques to tag the audio tracks of television shows to track and understand what American households are watching. These steganographic methods have even made their way into the modern Blu-Ray standard (the Cinavia system), as well as in government applications like authenticating drivers licenses, national currencies and other sensitive documents. The Digimarc corporation, for example, has developed a watermark for packaging that prints a product’s barcode nearly-invisibly all over the box, allowing any digital scanner in line of sight to read it. It’s also been used in applications ranging from brand anti-counterfeiting to enhanced material recycling efficiencies.

The here and now

Modern digital watermarking operates on the same principles, imperceptibly embedding added information onto a piece of content (be it image, video or audio) using special encoding software. These watermarks are easily read by machines but are largely invisible to human users. The practice differs from existing cryptographic protections like product keys or software protection dongles in that watermarks don’t actively prevent the unauthorized alteration or duplication of a piece of content, but rather provide a record of where the content originated or who the copyright holder is.

The system is not perfect, however. “There is nothing, literally nothing, to protect copyrighted works from being trained on [by generative AI models], except the unverifiable, unenforceable word of AI companies,” Dr. Ben Zhao, Neubauer Professor of Computer Science at University of Chicago, told Engadget via email.

“There are no existing cryptographic or regulatory methods to protect copyrighted works — none,” he said. “Opt-out lists have been made made a mockery by stability.ai (they changed the model name to SDXL to ignore everyone who signed up to opt out of SD 3.0), and Facebook/Meta, who responded to users on their recent opt-out list with a message that said ‘you cannot prove you were already trained into our model, therefore you cannot opt out.’”

Zhao says that while the White House's executive order is “ambitious and covers tremendous ground,” plans laid out to date by the White House have lacked much in the way of “technical details on how it would actually achieve the goals it set.”

He notes that “there are plenty of companies who are under no regulatory or legal pressure to bother watermarking their genAI output. Voluntary measures do not work in an adversarial setting where the stakeholders are incentivized to avoid or bypass regulations and oversight.”

“Like it or not, commercial companies are designed to make money, and it is in their best interests to avoid regulations,” he added.

We could also very easily see the next presidential administration come into office and dismantle Biden’s executive order and all of the federal infrastructure that went into implementing it, since an executive order lacks the constitutional standing of congressional legislation. But don’t count on the House and Senate doing anything about the issue either.

“Congress is deeply polarized and even dysfunctional to the extent that it is very unlikely to produce any meaningful AI legislation in the near future,” Anu Bradford, a law professor at Columbia University, told MIT Tech Review. So far, enforcement mechanisms for these watermarking schemes have been generally limited to pinky swears by the industry’s major players.

How Content Credentials work

With the wheels of government turning so slowly, industry alternatives are proving necessary. Microsoft, the New York Times, CBC/Radio-Canada and the BBC began Project Origin in 2019 to protect the integrity of content, regardless of the platform on which it’s consumed. At the same time, Adobe and its partners launched the Content Authenticity Initiative (CAI), approaching the issue from the creator’s perspective. Eventually CAI and Project Origin combined their efforts to create the Coalition for Content Provenance and Authenticity (C2PA). From this coalition of coalitions came Content Credentials (“CR” for short), which Adobe announced at its Max event in 2021. 

CR attaches additional information about an image whenever it is exported or downloaded in the form of a cryptographically secure manifest. The manifest pulls data from the image or video header — the creator’s information, where it was taken, when it was taken, what device took it, whether generative AI systems like DALL-E or Stable Diffusion were used and what edits have been made since — allowing websites to check that information against provenance claims made in the manifest. When combined with watermarking technology, the result is a unique authentication method that cannot be easily stripped like EXIF and metadata (i.e. the technical details automatically added by the software or device that took the image) when uploaded to social media sites (on account of the cryptographic file signing). Not unlike blockchain technology! 

Metadata doesn’t typically survive common workflows as content is shuffled around the internet because, Digimarc Chief Product Officer Ken Sickles explained to Engadget, many online systems weren’t built to support or read them and so simply ignore the data.

“The analogy that we've used in the past is one of an envelope,” Chief Technology Officer of Digimarc, Tony Rodriguez told Engadget. Like an envelope, the valuable content that you want to send is placed inside “and that's where the watermark sits. It's actually part of the pixels, the audio, of whatever that media is. Metadata, all that other information, is being written on the outside of the envelope.”

Should someone manage to remove the watermark (turns out, not that difficult, just screenshot the image and crop out the icon) the credentials can be reattached through Verify, which runs machine vision algorithms against an uploaded image to find matches in its repository. If the uploaded image can be identified, the credentials get reapplied. If a user encounters the image content in the wild, they can check its credentials by clicking on the CR icon to pull up the full manifest and verify the information for themselves and make a more informed decision about what online content to trust.

Sickles envisions these authentication systems operating in coordinating layers, like a home security system that pairs locks and deadbolts with cameras and motion sensors to increase its coverage. “That's the beauty of Content Credentials and watermarks together," Sickles said. "They become a much, much stronger system as a basis for authenticity and understanding providence around an image” than they would individually." Digimarc freely distributes its watermark detection tool to generative AI developers, and is integrating the Content Credentials standard into its existing Validate online copy protection platform.

In practice, we’re already seeing the standard being incorporated into physical commercial products like the Leica M11-P which will automatically affix a CR credential to images as they’re taken. The New York Times has explored its use in journalistic endeavors, Reuters employed it for its ambitious 76 Days feature and Microsoft has added it to Bing Image Creator and Bing AI chatbot as well. Sony is reportedly working to incorporate the standard in its Alpha 9 III digital cameras, with enabling firmware updates Alpha 1 and Alpha 7S III models arriving in 2024. CR is also available in Adobe’s expansive suite of photo and video editing tools including Illustrator, Adobe Express, Stock and Behance. The company’s own generative AI, Firefly, will automatically include non-personally identifiable information in a CR for some features like generative fill (essentially noting that the generative feature was used, but not by whom) but will otherwise be opt-in.

That said, the C2PA standard and front-end Content Credentials are barely out of development and currently exceedingly difficult to find on social media. “I think it really comes down to the wide-scale adoption of these technologies and where it's adopted; both from a perspective of attaching the content credentials and inserting the watermark to link them,” Sickles said.

Nightshade: The CR alternative that’s deadly to databases

Some security researchers have had enough waiting around for laws to be written or industry standards to take root, and have instead taken copy protection into their own hands. Teams from the University of Chicago’s SAND Lab, for example, have developed a pair of downright nasty copy protection systems for use specifically against generative AIs.

Zhao and his team have developed Glaze, a system for creators that disrupts a generative AI’s style of mimicry (by exploiting the concept of adversarial examples). It can change the pixels in a given artwork in a way that is undetectable by the human eye but which appear radically different to a machine vision system. When a generative AI system is trained on these "glazed" images, it becomes unable to exactly replicate the intended style of art — cubism becomes cartoony, abstract styles are transformed into anime. This could prove a boon to well-known and often-imitated artists especially, in keeping their branded artistic styles commercially safe.

While Glaze focuses on preventative actions to deflect the efforts of illicit data scrapers, SAND Lab’s newest tool is whole-heartedly punitive. Dubbed Nightshade, the system will subtly change the pixels in a given image but instead of confusing the models it's trained with like Glaze does, the poisoned image will corrupt the training database its ingested into wholesale, forcing developers to go back through and manually remove each damaging image to resolve the issue — otherwise the system will simply retrain on the bad data and suffer the same issues again.

The tool is meant as a “last resort” for content creators but cannot be used as a vector of attack. “This is the equivalent of putting hot sauce in your lunch because someone keeps stealing it out of the fridge,” Zhao argued.

Zhao has little sympathy for the owners of models that Nightshade damages. “The companies who intentionally bypass opt-out lists and do-not-scrape directives know what they are doing,” he said. “There is no ‘accidental’ download and training on data. It takes a lot of work and full intent to take someone’s content, download it and train on it.”