An economic reckoning, legislative loopholes, and the AI-earth relationship.
S1E3 | Highlights and impact of this week's top three tech themes.
Hi, Friends —
Welcome to [our digital disco]! I’m excited to have you here. Keep scrolling for key themes in tech news this week, as well as misc. outputs from my brain (Snack Time). You can also check out last week’s newsletter here.
Notable Themes
☞ Loopholes & Privacy Protection (or lack thereof).
The FBI admitted to previously purchasing US phone location data without obtaining a warrant. FBI Director Christopher Wray stated that the FBI is not currently doing so, and now relies on a “court-authorized process” to obtain location data from companies gathered specifically for advertising purposes.
Why does it matter? The FBI isn’t alone. It has become increasingly common for US government agencies to purchase personal location data from data brokers, sidestepping the legal need for a warrant. This is because the 2018 ruling of Carpenter v. United States — which held that government agencies need a warrant to access cell location data from cell phone companies — left open a loophole that allows the government to obtain this information by other means. Many are concerned about the privacy of everyday citizens, as well as the transparency of these government practices.
Pros: At the time of its decision, the Carpenter ruling was largely recognized as a win for digital privacy advocates. In theory, it recognized that modern technology has fundamentally changed the nature of privacy, and that individuals should have a reasonable expectation of privacy in their location data. The ruling also reinforces the principle that law enforcement agencies must have a specific, justifiable reason for accessing personal information.
Cons: We are reminded that a law, when narrowly interpreted, can quickly become obsolete in the face of new technologies. The vast amount of data collected by Google and other data brokers is not protected to the same constraints as that from cell phone companies — effectively giving law enforcement agencies carte blanche access to gather personal data that they would not otherwise be able to. Federal agencies including US Customs and Border Protection and the Defense Intelligence Agency have taken advantage of this loophole, state and local authorities are known to acquire cellphone-tracking software, and the Department of Homeland Security reportedly purchased the geolocations of millions of Americans from private marketing firms. US lawmakers have struggled to pass a comprehensive privacy law.
☞ SVB & an Economic Reckoning.
Silicon Valley Bank (SVB), a major lender for startups, was shut down late last week. In short: as startups experienced difficulty equity capital raising late 2022 into 2023, SVB needed to raise cash for deposit withdrawals. The bank pursued a capital raise and sold $21B in securities, leading to a downgrade by Moody's. VCs and startups rushed to withdraw funds totaling to $42B, resulting in SVB's closure by regulators on March 10, 2023. Silicon Valley Bank’s unexpected failure — the most significant bank collapse since the global financial crisis — has sent reverberations throughout the tech economy.
Why does it matter? Last week, Silicon Valley Bank was the favorite bank of startups supported by the Bay Area’s most prominent venture firms. Today, SVB is an imploded institution, run by regulators in the US and by HSBC in the UK (HSBC acquired SVB’s British arm for £1). The events have made many founders, investors, and spectators wary of tech’s unpredictable economic environment. Founders who did not bank with SVB may also be impacted by delayed or rescinded investments due to the bank's collapse. This could have a significant impact on the startup ecosystem in Silicon Valley and beyond, as access to funding is critical for the growth and success of new companies.
Pros: On Sunday, the Department of the Treasury, Federal Reserve and FDIC guaranteed all deposit accounts at Silicon Valley Bank, ensuring that depositors would have access to their money and no losses would be borne by taxpayers. Given over 90% of SVB deposits were uninsured, the move aimed to prevent further chaos and widespread fallout from the bank's collapse. Moreover, given SVB’s niche focus in the tech sphere, many financial professionals don’t believe SVB’s end will largely spillover into the broader banking system.
Cons: The resulting chaos could lead to more centralization in the AI field, as large corporations pick off the struggling startups and buy them for cheap. Many companies who banked with SVB will also experience issues such as delays in payroll processing, which may cause issues for fast-growing tech startups. While the diversification of holdings in mainstream banks may provide some stability, the unknowns created by the current unprecedented time pose a threat to the stock market.
☞ Environment & Computation.
Artificial intelligence, particularly generative AI, requires a massive amount of energy to process, store, and search internet content. This energy comes at a significant cost: training a single model can consume more electricity than 100 US homes in a year. In hand with the electricity required to power, operate, and cool specialized hardware, the rise of AI is strongly tied to a rise in energy consumption and — you guessed it — carbon emissions. But it doesn’t have to be this way. AI is an accelerant: its technological capabilities are increasing exponentially, making it cheaper and more accessible to a wider range of businesses and industries. When purposefully applied to fighting the climate crisis or optimizing energy-saving methods, AI could have a massive impact on reducing our carbon footprint.
Why does it matter? The field lacks transparency, making it difficult to measure AI’s environmental impact. (E.g., Does a data center get its electricity from renewable energy, such as wind farms, or from fossil fuels, which produce much higher emissions?) Much of the focus on AI carbon emissions is directed at the upfront data cost, and may fail to provide an accurate understanding of AI’s carbon footprint. A recent study by AI startup Hugging Face found that the carbon emissions of its own large language model doubled when they included other emissions, including the energy used to manufacture computer equipment and to maintain and run the model after training.
Pros: As models get larger, many AI companies are also working on improvements to make them run more efficiently. Microsoft and Google are leading aggressive sustainability initiatives to reduce emissions. AI is also being leveraged to support the environment. Many conservationists value technological tools to monitor, protect, and alleviate ecological damage. In the African savannah, for example, researchers implemented an AI-powered camera to identify African forest and savannah elephants — recently classified as ‘critically endangered’ and ‘endangered’ — to monitor and protect the wildlife. The device sends real-time data to local rangers and communities, who can respond to crises significantly faster than with former tools.
Cons: One third-party analysis estimates that the training of GPT-3, which supports OpenAI’s ChatGPT, consumed enough energy to emit enough energy to send single person roundtrip between New York and San Francisco 550 times. And given ChatGPT’s data cuts off in late 2021 (partly to reduce computing requirements), this model will need to be retrained. This obscene energy consumption speaks to just one large model’s impact — a model which kickstarted an arms race in AI. As startups and Big Tech race to produce the leading products, the demand for more powerful computing resources may increase, leading to higher energy consumption and greenhouse gas emissions. This competitive nature may also result in shorter product lifecycles, leading to more frequent hardware upgrades and disposal of outdated or obsolete AI hardware (which often contain hazardous materials).
Snacktime
📓 Reading: What Conversation Can Do for Us — A New Yorker piece on being human and leaving agendas at the door.
♬ Listening to: Billy Joel’s Vienna. Its nostalgia lies on the idea of missed opportunities and the regret that can come with them, and serves as a reminder to slow down and enjoy life.
✰ Thinking about: Postmodernism. The postmodern art movement highlights art history — in some ways, the artist edging one’s own space into a conversation that already exists — and rejects the notion of artistic advancement. Given generative models combine historic artworks and styles, should we consider AI-generated art to fall into this movement? If not, how does the act of a human synthesizing what already exists differ from that of a machine? And, if I may answer my own question with another question: How does the entropy in a human’s mind differ from the entropy of a model?
Next up
✎ OpenAI’s pricing strategy and take on stakeholder value. And whatever is happening in the news. And whatever rant I go on over the next six days.
✿ As always — any and all feedback is welcome! In the meantime: give someone a hug and say an ‘I love you’ this week. Make the world a little happier.