Real Benefits, Real Concerns: The Full Picture
Companion to Issue #14 — the fifth and final class in our AI Fundamentals series. My goal here is informed confidence — not fear, and not uncritical enthusiasm. This guide walks through everything the newsletter couldn't fit: where AI stands today, what it could realistically do for us, what it costs, what it means for jobs, how to protect yourself, and who's supposed to be setting the rules.
1. Where AI Stands Right Now
Seventy-five years of consumer technology followed the same pattern: something amazing arrives, and within a decade or two it becomes ordinary. Television, the personal computer, the smartphone, GPS — all of them went from remarkable to unremarkable. Generative AI (ChatGPT, Claude, Gemini) is the most recent leap. “Agentic AI” — software that can take actions through connected tools, not just answer questions — is the wave arriving right now. It's real, but it's early.
Six Companies Shaping the Headlines
Anthropic (Claude) — $65B+ annualized revenue pace (reported Aug 2026)
OpenAI (ChatGPT) — 1B+ weekly active users (Aug 2026)
Google (Gemini) — 1B+ monthly users (Aug 2026)
Microsoft (Copilot) — 30M+ paid Microsoft 365 Copilot licenses (Jul 2026)
xAI (Grok) — 600M monthly users across X + Grok combined (Jan 2026)
Meta (Muse) — launched Sept 8, 2026; standalone user count not yet verified
These numbers are a dated snapshot, not a ranking — the measures differ (users vs. revenue vs. licenses), and user populations overlap across products. Checked September 18, 2026; recheck before reusing.
2. The Real Trade-off: Benefits and Concerns
Both stories about AI are true at the same time. It genuinely helps, and it genuinely raises real concerns. Holding both without flinching is the whole point of “informed confidence.”
What AI Could Help Us Achieve — opportunities, not guarantees
Productivity — accomplish more with less time and effort.
Medicine — help detect disease and develop better treatments, supporting (not replacing) professional judgment.
Science — accelerate discovery and help solve difficult problems.
Prosperity — create economic value and new opportunities, not guaranteed to be distributed equally.
Safety — help identify hazards and prevent harm.
What We Need to Get Right — the five concerns
Environmental — managing the demands AI places on electricity, water, and communities.
Privacy — protecting personal information and controlling who has access to it.
Humanity — preserving human judgment, relationships, and control.
Jobs — helping people navigate changing work and, in some cases, displacement.
Trust — limiting misinformation, impersonation, and scams.
These are opportunities and risks to manage, not predictions that either the benefits or the harms are guaranteed or evenly shared.
3. The Real-World Cost of the Cloud
AI doesn't run on thin air. Every chatbot answer and every generated image runs through a data center — a facility full of computers and the systems that power and cool them. That has real, physical costs worth understanding.
Forecast share of U.S. electricity used by data centers in 2030: 9.5–15.3% — that range includes all data centers, not just AI, and it's a forecast, not an observation. (Source: Berkeley Lab, 2025 update.)
Electricity & emissions — growing power demand can add emissions when fossil fuels supply the electricity. The response being pursued: efficient systems and cleaner power.
Water for cooling — some cooling systems use substantial water, adding pressure where supplies are limited. The response being pursued: water-saving cooling and careful location choices.
Local impacts — large facilities can bring noise, land-use changes, and added infrastructure needs. The response being pursued: local planning and clear community safeguards.
“Solution” labels identify responses being pursued — not guarantees. Sources: IEA, U.S. DOE, ACM (2025). Impacts vary by location, energy source, and cooling design.
A Bipartisan Political Flashpoint
Data centers have become political in a way that's worth knowing about, regardless of which party you support. A September 2026 Republican strategy memo (Tony Fabrizio / Fabrizio Lee) — written to advise GOP candidates on messaging, not a neutral study — reported polling showing broad opposition to unconditional data center buildout across party lines: 65% opposed construction without protections, while 61% favored it with enforceable protections in place.
A few honest caveats: this is a campaign messaging document, not an independent academic survey, and it explicitly did not test whether this issue would change how anyone actually votes. What's genuinely useful here isn't the politics — it's the underlying question worth asking about any data center proposed near you: who pays for the infrastructure, and what protects local electricity bills, water supply, and the environment?
4. Two Concerns About Humanity
“AI and humanity” covers two very different worries, and it's worth keeping them separate — they call for different responses.
Erosion of humanity (what could we lose?) — AI could gradually weaken qualities we value if we lean on it too heavily: our own skills, motivation, independent judgment, integrity, relationships, and sense of purpose. This is about how we choose to use — and depend on — AI, not a prediction that it will happen.
Destruction of humanity (could we lose control?) — the more extreme concern: that powerful AI could enable catastrophic harm, through deliberate misuse or through systems acting beyond effective human control. Catastrophic risk and literal extinction are not interchangeable claims — and no current system has demonstrated the ability to cause that kind of harm. This is a category of concern, not a forecast.
How to Respond to Each
For erosion — keep people at the center: keep people responsible for judgment, relationships, care, and consequential decisions; use AI for higher-order work while still exercising your own skills and creativity; prepare for new roles and opportunities as work changes.
For destruction risk — build and test safeguards: review and test safeguards before deployment, and as capabilities grow; limit what AI can access and do, especially in high-consequence situations; maintain human oversight and a genuinely tested ability to pause or stop harmful operations.
A stop button alone isn't enough — people still have to recognize trouble and step in before harm becomes irreversible.
5. AI Will Change the Work — What Happens to the Worker?
These three categories describe tasks within a job, not entire jobs. Most real jobs are a mix of all three.
Human-led — people perform most tasks and remain responsible for judgment, trust, and action. Example roles: firefighters, electricians, judges.
AI-augmented — people use AI to improve their own work. Example roles: teachers, clinicians, engineers.
AI-automated — software or robots perform defined tasks with limited human involvement. Example tasks: document processing, assembly, sorting.
These are examples of human-led responsibility today — not a claim that any role is immune from technology forever. Some roles may shrink, disappear, or shift; transition support for the people affected matters.
The Opportunity Side: Higher-Order Work
Using AI to support more demanding, meaningful work — framing problems, weighing consequences, relating to people, and directing tools responsibly: solving harder problems, exercising judgment, building relationships, and directing AI tools while remaining accountable.
New roles may emerge, but creation doesn't necessarily match displacement in timing, location, or required skill. Training, support, and time to adapt are what make this transition work for real people.
6. Protect Your Data. Check Before You Act.
Personal information: keep passwords out of chats; adjust privacy settings in every app you use; grant AI tools only the access they need — and disconnect access when you're done; share sensitive details only when it's genuinely necessary.
Suspicious messages: use AI, or another trusted tool, to help you assess a suspicious message; verify through a phone number or website you already know is real — never one supplied by the suspicious message itself; remember that a reassuring answer is not proof.
Try It Yourself: Investigate a Suspicious Message
Next time you get an email or text that feels off — a delivery notice, a bank alert, an “urgent” request — try this before you click anything:
Redact any real personal details (names, addresses, account numbers, tracking codes) before you paste anything into an AI tool.
Ask the AI to identify specific warning signs in the message, and to help you research the vendor or service it claims to be from.
Open official sources yourself, independently — don't rely only on what the AI tells you.
Never click the links or submit payment information as part of checking it out. If in doubt, throw it out.
7. AI Can Sound Certain — and Still Be Wrong
A quick refresher on three questions worth asking about anything AI tells you:
Accuracy — a confident answer can still be wrong. Check important facts.
Evidence — a citation may be invented or misleading. Open the source — does it actually support the claim?
Authenticity — images, voices, and messages can be fabricated. Confirm through an independent, trusted channel.
The Higher the Stakes, the More You Verify
Casual questions (movie casts, past World Series winners) — a quick check may be enough. School & research (college essays, research papers) — check original sources and citations. High-stakes decisions (medical claims, legal briefs) — verify independently and consult a qualified professional.
This is a teaching guide, not a measured formula. Independent sources means outside confirmation — not just asking the same AI a second time.
8. Who Sets the Rules for AI?
This is a framework for discussion, not a legal compliance checklist — these rules don't exist everywhere yet, and different places already have different laws. Two levels are typically in play:
Organizational policy (schools, colleges, universities, banks, and companies) — defines allowed and prohibited uses, sets expectations for students, customers, faculty, and employees, and protects data by requiring disclosure, human review, and accountability.
Government regulation (national and local policy discussions) — classifies AI by capability, use, and potential harm; requires testing and independent review for higher-risk systems; establishes certification, oversight, monitoring, and enforcement.
A certification process can't guarantee permanent safety — ongoing review is what actually matters over time. This section is a discussion framework, not legal advice.
What to Watch. What You Can Do.
More capability is coming — more capable agents, and eventually more capable robots. What matters is whether reliability, permissions, and oversight keep pace with that capability. That's the thing worth watching, more than any single headline.
Three Things You Can Do Starting Today
Protect your data. Set limits on what any AI tool can access.
Match your verification effort to the stakes — and never treat a reassuring AI answer as proof.
Keep practicing your own skills and judgment. AI should support your thinking, not replace it.
This closes out our five-class AI Fundamentals series: Introduction to AI, AI & ChatGPT, AI for Everyday Life, AI for Creativity & Hobbies, and AI Safety & The Future. More classes and topics are on the way — keep learning, keep exploring.
Talk soon,
Gary
Founder, AI Unlocked
Browse the Newsletter Archive →
Written for AI Unlocked · aiunlocked.us
Your trusted guide on the AI journey