This morning, the White House released a formal "Action Plan” with suggestions and proposals across nearly all facets of the AI industry. While the bulk of the document is standard fare (albeit devoid of any particular metrics or cost sheets), there are numerous massive red flags with how the current administration is planning to handle AI’s development over the next few years.
I believe two of these particular areas are particularly critical missteps:
- Erasing core safeguards (misinformation, DEI, and climate) from NIST’s AI risk management framework
- Sidelining environmental concerns of AI development via deregulation and/or a dramatic deemphasizing of existing emission reduction strategies
Gutting NIST’s AI guidelines
In the doc, the administration is explicitly ordering NIST to purge critical ethical considerations from their AI risk management framework (lauded in 2013 as a cornerstone for responsible AI, emphasizing that AI systems should be “valid and reliable, safe, secure and resilient, explainable, privacy-enhanced, and fair – with harmful bias managed”). There is a belief among the admin that removing these guidelines (which they falsely claim are “ideological”, as always) will help restore free speech and remove bias in the output of these models.
- Misinformation. The plan orders NIST to scrub the term entirely even as experts warn generative models crank out convincing falsehoods at scale, opening the door for election meddling and public‑health hoaxes. The models themselves, without the evils of human manipulation, still sport a 15% hallucination rate as of last year.
- DEI. Removing DEI (an initiative that a has become unreasonably politicized) strips requirements for diverse teams and fairness checks; real‑world studies already show AI-based face recognition, lending, and hiring tools discriminate against women and minorities. Losing diverse data requirements perpetuates real, measured biased outputs. Time and time (and time) again.
- Climate change. Deleting climate change references and metrics lets corporation off the hook, and serves to delegitimize the horrors around modern climate science predictions. Retaining climate-based measurements is a vital incentive that drives more efficient models.
This guidance is a critical misunderstanding of how these models work, and is ironically encouraging the very thing they feel they are protecting against. These “idealogical” guardrails are the very tools that lower risk; take them out and models will see more hallucinations, deeper discrimination, and an even more bleak climate outlook.
Tossing out environmental considerations
Here’s the current reality of AI and energy use:
- Electricity. Data centers now suck up over 1% of global electricity. By 2026, their consumption could match countries like Germany or Sweden. AI-specific workloads already chew through 10-20% of U.S. data center energy (multiply this by millions more AI queries daily).
- Carbon emissions. Microsoft’s emissions alone jumped 30% thanks to AI-driven cloud computing, leading to criticisms on the abandonment of long-state sustainability efforts.
- Water usage. By 2027, AI-driven data centers might guzzle a trillion gallons of water annually. Arizona’s Microsoft data centers alone could consume 50 million gallons of drinking water every year.
The Action Plan calls for:
- Fast-tracking data center permits by skipping critical environmental reviews.
- Rolling back protections under the Clean Air Act and Clean Water Act.
- Opening federal lands to unchecked data center construction and ensuring older, dirtier power plants stay online longer.
The energy draw of casual ChatGPT may be overstated, but the focus from the corporations providing these services should be on researching and developing smaller, more energy-efficient models (not encouraging bulkier beasts).
The energy draw of data centers is absolutely not overstated, and is growing increasingly more and more concerning as we learn about their effects. This type of construction regulation is the area of AI development that is in need of government oversight the most, and the current administration seems to only be interested in cranking up the scales.
What this means for AI development
Innovation must be paired with accountability and sustainability; I’ve been shouting this from the rooftops since 2022. We have to tackle misinformation head-on, actively address bias and inequity, and aggressively drive efficiency and sustainability. The alternative is distrust now and disaster later.
- Trust erosion. If AI systems flood society with misinformation or deepfake-driven scandals, public trust collapses. The promises of AI are useless if no one trusts it enough to produce them.
- Bias and social backlash: Unchecked AI inevitably creates biased outcomes. Think wrongful arrests, discriminatory lending, and unfair hiring practices. These outcomes invite litigation, regulation, and public backlash.
- Environmental and energy crises. Ignoring AI’s environmental footprint now guarantees massive energy and water problems later. Brownouts, water shortages, and skyrocketing carbon emissions are becoming more and more clear as the years go on.
- Global reputation damage. Europe and Asia increasingly set tough AI regulations around fairness and sustainability. The U.S. risks falling behind, becoming the exporter of unethical, energy-guzzling AI.
“Build, Baby, Build!” is a short-sighted (and rather uncreative) motto. Building AI under the terms outlined in the Action Plan will result in catastrophes (like the one forming in Memphis) that we won’t truly understand the full impact of for years to come.
There is a responsible way to build AI and embrace its best promises. This is not it.