Plain-English thinking on AI and automation.

Every major technology shift — from the Luddites smashing looms to fears around computers and factory automation — has triggered similar anxieties. History shows the reality is usually more nuanced and ultimately human: technologies automate specific tasks, not entire jobs, while reshaping roles around uniquely human strengths like judgment, creativity, taste, empathy, communication, and coordination.

AI follows this pattern. It excels at routine cognitive work (data entry, basic analysis, drafting, summarization), but struggles with the full complexity, context, and accountability most jobs require. Many roles are being augmented and transformed rather than eliminated. Recent evidence from 2025–2026 studies shows AI-adopting companies and industries experiencing productivity gains, wage growth in many sectors, and more job reshaping than outright replacement.

New opportunities are emerging around AI itself: prompt engineering and orchestration, model evaluation and red-teaming, AI ethics and governance, high-quality data curation, synthetic data creation, specialized domain expertise for training, and integration of AI into existing workflows. Sectors with heavy routine work (clerical, basic coding, customer support) face more disruption and transition challenges — especially for entry-level positions — but the broader trend points to transformation, new demand, and economic growth rather than mass unemployment.

The key variable isn’t AI taking jobs. It’s how quickly people and organizations adapt, upskill, and redesign work around human-AI collaboration.

Current systems are powerful pattern learners and predictors trained on massive amounts of data. They can be astonishingly useful, but they do not possess consciousness, lived experience, emotional depth, or consistently reliable reasoning.

In my view, much of what feels like growing intelligence comes less from the raw models themselves and more from the harnesses and frameworks we build around them — sophisticated prompting techniques, agentic scaffolds, tool integration, memory systems, and retrieval pipelines. These structures make AI appear significantly smarter and more capable than the underlying model would be on its own.

Progress is real, and capability jumps deserve attention. But timelines for AGI are speculative, and claims of imminent consciousness or takeover are often closer to marketing, fundraising, or science fiction than sober engineering.

Bias is real. Large language models absorb patterns from vast human-generated training data, cultural internet content, editorial choices by developers, and safety alignment processes. This can lead to noticeable skews on politically charged topics.

However, these biases are not monolithic or locked in one direction. Different models show different tendencies depending on their training data, fine-tuning, and guardrails. Some lean progressive, others more neutral or even contrarian. Open-source and local models (like Llama, Mistral, Qwen, or Phi) can be further customized by users to reduce unwanted slant.

The idea that AI bias is uniquely dangerous overstates the risk. Humans are biased too. Often far more consistently and emotionally. The real danger comes from over-reliance on any single system, not from AI itself.

Better approaches include:

• Transparency around training data and alignment methods
• Adversarial testing and red-teaming
• Pluralism — using multiple models and comparing outputs
• User control through local/open models, fine-tuning, and custom system prompts
• Ongoing research into debiasing techniques and synthetic balanced datasets
In short, bias is a solvable engineering and data problem, not an inevitable path to ideological capture. The healthiest path forward is more choice, more openness, and less dependence on any one company’s worldview.

Generative AI is a powerful creative collaborator. It excels at rapid ideation, drafting, iteration, scaling production, and handling repetitive or tedious tasks like generating variations, marketing assets, or initial prototypes. It can produce derivative and competent work at incredible speed.

However, truly original, emotionally resonant, and culturally meaningful creativity still relies on deeply human elements: lived experience, personal intent, emotional depth, cultural context, moral judgment, and refined taste. AI lacks these. It remixes patterns from existing data rather than drawing from genuine insight or soul.

History supports this view. Cameras didn’t kill painting, they helped birth new movements like Impressionism and street photography. Synthesizers didn’t end music, they expanded what musicians could create. The same pattern is unfolding with AI. Many artists, writers, musicians, and designers are already using GenAI to move faster, explore more directions, eliminate busywork, and amplify their vision.

Audiences continue to crave human authorship, authenticity, trust, and unique perspective. The most successful creators will be those who master AI as a tool while doubling down on what machines can’t replicate: personal voice, emotional truth, and intentional storytelling.

AI has made high-quality synthetic media far easier and cheaper to produce, so the risk of targeted manipulation, confusion, and erosion of trust is genuine. Especially in the final days before voting.

That said, recent election cycles (including the high-stakes 2024 U.S. and global elections) did not deliver the feared wave of AI-generated content that swung outcomes at scale. Incidents occurred, but they remained limited in measurable impact.

Traditional misinformation channels — partisan media, memes, influencer networks, and longstanding human-driven disinformation — continue to dominate. AI tools sometimes amplify these older tactics, but they haven’t replaced them.

Several factors help mitigate the threat:

• Improving detection tools and watermarking/provenance standards (like C2PA)
• Platform policies and rapid response systems
• Growing public skepticism and media literacy
• Legal measures in many jurisdictions targeting deceptive election deepfakes

Democracy has always struggled with manipulation and falsehoods. From print-era propaganda to social media virality. Generative AI adds new capabilities and urgency, but it doesn’t fundamentally break systems that were already imperfect. Vigilance is essential, yet panic overstates the novelty and underestimates society’s ability to adapt.

U.S. data centers power the internet, cloud computing, and the rapid growth of AI, yet they account for roughly 0.2% or less of national freshwater withdrawals and consumption. Direct on-site water use for cooling is even smaller — often estimated at around 0.04% to 0.05%.

To put this in perspective:

• Agriculture uses about 70% of U.S. freshwater, with corn irrigation alone consuming 10x to 16x more water than all data centers combined.
• All U.S. data centers together consumed roughly 17–20 billion gallons directly in recent years (around 47–55 million gallons per day on average).
• This is far below household use, industrial processes, golf courses, and residential lawns (which together consume trillions of gallons annually).

While the national numbers are modest, local impacts can feel more significant in water-stressed regions (e.g., parts of Arizona, Texas, or Virginia) where multiple large facilities cluster. The industry is responding with efficiency improvements, reclaimed wastewater, air/liquid cooling alternatives, and water-positive commitments from major players like Google and Microsoft.

Context matters: AI and digital infrastructure deliver enormous economic and societal value, while water challenges are real but solvable through better siting, technology, and policy. Just as we’ve done with other high-impact sectors.

A large hyperscale data center might withdraw 1 to 5 million gallons of water per day, which naturally makes for scary headlines comparing it to a small town’s usage. However, withdrawal is not the same as consumption.

• Often 20–30% or more of the withdrawn water is returned to the local water system as treated wastewater.
• Many facilities use reclaimed wastewater or non-potable sources rather than drawing from municipal drinking water supplies.
• Newer data centers are adopting air cooling, liquid cooling, closed-loop systems, and advanced efficiency tech that dramatically reduce water use.
• While hyperscale facilities sit at the high end, many typical centers consume far less — with some reports showing around 80,000 gallons per day or lower.

When viewed nationally, these numbers remain modest relative to other sectors. The industry is actively improving: major players are investing in water-positive goals, alternative cooling methods, and smarter siting to minimize local strain. Local concerns in arid regions are valid and deserve scrutiny, but the blanket “AI is drying up America” narrative often exaggerates scale while ignoring engineering progress and relative impact.

In water-stressed regions such as parts of Arizona, Virginia, Oregon, and Texas, clusters of large data centers can put noticeable pressure on local water supplies and represent a visible portion of municipal demand. These localized impacts deserve serious attention, strong community oversight, and transparent planning.

That said, the scale is often more modest than headlines suggest. For example, data centers in Maricopa County, Arizona (home to greater Phoenix) have been estimated to use roughly 0.12% of total county water, compared to 3.8% for golf courses in the same area.

Key context:

• Data centers compete for water like any other industrial or commercial user. They do not receive automatic special priority.
• Many facilities are shifting toward reclaimed wastewater, dry cooling alternatives, and efficiency improvements to reduce their footprint.
• Local governments can — and should — enforce strict water usage standards, impact fees, and public reporting requirements.

Local challenges are legitimate and solvable through better policy, technology, and siting decisions. They do not justify broad national panic about AI “destroying” America’s water supply. As with past industrial growth, smart management and innovation can address these issues without halting digital progress.

The industry is making fast progress on Water Usage Effectiveness (WUE). Operators are deploying a wide range of improvements, including:

• Air cooling and free cooling in colder climates
• Direct-to-chip and immersion liquid cooling
• Higher cycles of concentration in cooling towers
• Greater use of reclaimed wastewater, non-potable sources, and even seawater in coastal areas

Major players like Google, Microsoft, and Equinix are publicly reporting meaningful reductions in water intensity per kilowatt-hour and per AI workload. Many have set ambitious water-positive goals, returning more water to local communities than they consume.

Importantly, the surge in AI demand is not just increasing water use — it is driving faster innovation in cooling technology and resource efficiency. As models and infrastructure scale, the industry is responding with better engineering rather than simply using more resources.

A large portion of the water impact attributed to data centers is indirect — it comes from the cooling water used by power plants that generate the electricity data centers consume. This isn’t unique to AI or data centers. Every home, factory, office, and EV charger shares this same indirect water footprint.

Here’s the encouraging part:

• Wind and solar power use virtually zero water during operation (unlike coal, natural gas, and nuclear plants).
• Major data center operators are increasingly signing long-term Power Purchase Agreements (PPAs) for carbon-free energy, directly funding new renewable projects.
• As the share of wind and solar on the grid grows, the indirect water impact of powering data centers steadily shrinks.

By driving demand for clean energy, data centers and AI are not only reducing their own water footprint, they’re helping accelerate the transition to a lower-water, lower-carbon electricity grid that benefits everyone.