Good morning, and happy Friday. That was a quick week.

Yesterday, a federal appeals court struck down net neutrality, which was introduced under former President Barack Obama, repealed under President-elect Donald Trump and reinstated under President Joe Biden.

The idea behind net neutrality was to ensure the internet remained equally fast for all users, regardless of the possible corporate interests of the internet providers.

— Ian Krietzberg, Editor-in-Chief, The Deep View

In today’s newsletter:

• 🩺 AI for Good: Stethoscopes for heart failure

• 🚨 Anthropic’s copyright guardrails

• ⚡️ Report: The grid can’t handle AI

AI for Good: Stethoscopes for heart failure

One of the ways to tell if a person might soon experience heart failure involves ejection fraction (EF), a percentage measurement of how much blood the left ventricle of the heart pumps out with each contraction. Scores below 49% could indicate cardiac damage, evidence of possible or pending heart failure. 

But the methods that exist to test for EF aren’t easily accessible during normal visits. Eko Health, a digital health firm, has been working on a solution. It involves AI. 

The details: Eko developed an algorithm, designed to work with its AI-enabled stethoscope, that is capable of detecting low EF scores. 

• The algorithm was trained on more than 100,000 ECGs and echocardiogram pairs from heart failure patients and was both clinically and independently validated. 

• In clinical trials, the algorithm achieved high levels of accuracy and sensitivity, in addition to particular accuracy in detecting heart conditions in pregnant women. 

The algorithm was approved by the U.S. Food and Drug Administration in April. 

“The ability to identify a hidden, potentially life-threatening heart condition using a tool that primary care and subspecialist clinicians are familiar with — the stethoscope — can help us prevent hospitalizations and adverse events,” Dr. Paul Friedman, chair of the Department of Cardiovascular Medicine at Mayo Clinic in Rochester, said. “Importantly, since a stethoscope is small and portable, this technology can be used in urban and remote locations, and hopefully help address care in underserved areas.”

Anthropic’s copyright guardrails

There are two elements to the AI and copyright conflict. One is the bigger question surrounding the baseline construction of generative models; in other words, does the training of AI models on copyrighted data without permission or compensation constitute infringement, or are developers legally protected? 

As this issue gets warred over in courts around the world, there is a separate, far more clear-cut issue of generative AI models producing copyright-infringing output, something that is a well-documented problem. 

For instance:

Many of the lawsuits publishers have filed against developers have to do with both elements. One of these such lawsuits — of eight music publishers against Anthropic, which was filed in October, 2023 — has come to a partial resolution. 

• Anthropic has implemented guardrails to prevent infringing output; the startup has agreed to maintain these guardrails across all its current and future products, something the publishers are happy with. 

• As such, the agreement resolves the injunction the publishers had filed, though it does not resolve — or even address — the rest of the lawsuit. 

Specifically, the publishers have requested “that Anthropic refrain from using unauthorized copies of publishers’ lyrics to train future AI models.” This is still an ongoing part of the suit.

• New research found that California’s grid ran on 100% renewables (solar, wind, water) for a record 98 out of 116 days in 2024, with no blackouts or cost-rises, something that proves that a large-scale grid powered by renewables is possible.

• The recent Tesla Cybertruck explosion, according to 404 Media, highlights issues of corporate surveillance; Elon Musk remotely unlocked the truck for law enforcement, and provided video from charging stations it had visited before the explosion.

• Competition for electricity will define the climate beat in 2025 (Semafor).

• Silicon Valley stifled the AI doom movement in 2024 (TechCrunch).

• Dow reverses course and drops 200 points in volatile start to 2025 (CNBC).

• Tesla sales dropped 1.1% in 2024, its first annual decline in a dozen years (AP).

• The US government announced a ‘historic’ nuclear energy deal (The Verge).

If you want to get in front of an audience of 200,000+ developers, business leaders and tech enthusiasts, get in touch with us here.

Report: The grid can’t handle AI 

We’ve talked often about the electrical, and, therefore, environmental, costs associated with the data centers that run AI applications. And while the environmental impact is clear, there’s another side to this unfolding story: grid instability. 

What happened: A months-long Bloomberg analysis of 770,000 Whisker Labs home power sensors found that more than 75% of highly distorted power readings occurred within 50 miles of significant data center activity. 

What it means: Whisker Labs operates a network of hundreds of thousands of simple home sensors that enable the company to monitor the power quality and resiliency of the electric grid across the country. It specifically enables the identification of those areas experiencing something called power distortion, which refers to disturbances that can “reduce energy efficiency, damage home appliances and harm critical grid infrastructure, ultimately leading to higher costs for homeowners.”

• The root of these distortions involves harmonics, the waves of electricity that travel through power lines. The distortion of these waves can impact power reliability and damage appliances; such distortion is also indicative of grid weakness and instability. 

• This monitoring network, according to Whisker co-founder and CEO Bob Marshall, is a vital step at a time when electrification and demand keep increasing while energy sources are changing. “The grid was not designed to accommodate the stress being placed on it today,” he said in November; monitoring it means companies can intervene in areas that are experiencing power quality issues. 

Bloomberg’s analysis found that more than half of the tracked households showing the worst levels of power distortion are within 20 miles of “significant data center activity.” This was particularly heightened in Northern Virginia, the most highly concentrated data center region — known in-industry as “data center alley — in the U.S. 

The Whisker sensors can’t themselves identify a single root cause behind this distortion; Bloomberg’s analysis instead identified a correlation between the rise in data centers and the subsequent spikes in distortion. 

It is no coincidence that this comes at a time when U.S. data center electricity demand has tripled over the past 10 years and is expected to double again by 2028. As a percentage of nationwide electricity consumption, data centers consumed 4.4% of total electricity in 2023, a number that the Department of Energy projected will surpass 12% by 2028. 

The landscape: The North American Electricity Reliability Corporation (NERC) recently released its 2024 long-term reliability assessment, which found that, for the first time in 20 years, we are experiencing “surging” energy demand at a time when fossil fuel plants are being retired and renewable energy sources aren’t coming online as fast as they are needed. 

• The culprit, in part, is AI: “demand growth is now higher than at any point in the past two decades. Increasing amounts of large commercial and industrial loads are connecting rapidly to the BPS (North American bulk power system),” the report reads. “The size and speed with which data centers (including crypto and AI) can be constructed and connected to the grid presents unique challenges for demand forecasting and planning for system behavior.”

• Specifically, less capacity is being generated than “what was projected and needed to meet future demand,” according to the report. “The trends point to critical reliability challenges facing the industry: satisfying escalating energy growth, managing generator retirements and accelerating resource and transmission development.”

Further, the U.S. Government Accountability Office in 2021 published a report that investigated the infrastructure risk posed by a changing climate, one complete with both wildfires and storms of increasing occurrence and severity, a combination that adds a layer of instability to a grid that, before all that, is already facing burgeoning reliability issues. 

The challenge here is one of balancing priorities. In 2022, U.S. electricity production and transmission accounted for a quarter of total greenhouse gas emissions, according to the EPA. Though there are other significant sectors — transportation accounted for 28% of emissions and industry for 23% — we need these numbers to start coming down, and quickly. 

But, because of soaring energy demand for AI (and crypto) projects, demand is outpacing production, an untenable situation that is already leading to delays in the retirement of coal power plants. 

And, beyond AI, as temperatures continue to rise — 2024 is set to be the hottest year on record — air conditioning will become more and more important (and costly to an already beleaguered grid). Such cooling can save tens of thousands of lives each year, according to the International Energy Agency, but “the rapid growth of AC is putting stress on the power grid.”  

Now, machine learning certainly has a role in enhancing operational efficiencies, which is already much-needed, something we’ve also talked about before. But the grid, facing a number of challenges, was not built to handle them all, and it certainly wasn’t built to do so sustainably. 

AI does not exist in a vacuum. So when we think about rapidly expanding integration, we have to center the cost, both short and long-term, of that integration. 

It is not in dollars alone. 

🤔 Your thought process:

Selected Image 1 (Left):

• “Thumb on other image is too smooth and doesn't have a very lifelike bend. foreground and background on that image also don't look like they came from one image.”

Selected Image 2 (Right):

• “First image seemed generated because of the number’s shadow.”

Thanks for reading today’s edition of The Deep View!

We’ll see you in the next one.