AI is Awful
(and also sometimes great)
What you actually need to know to make up your own mind
About this article
Our colleague Tim is tetraplegic. Each morning he uses AI to open his son's curtains.
Somewhere else, an algorithm quietly filters a job application out of a pile, and no one will ever explain why.
Both happen all the time. Both are AI. One gives a person more control over their own life. The other takes it away without explanation. And whether you've typed a prompt into ChatGPT or not, this technology is already shaping decisions about your life — including some you'll never see coming.
Kia ora. I'm Ingrid, and I wrote this with my colleague Ollie. At My Life My Voice we've spent the last year exploring AI in the disability sector — running training for Deaf and Disabled people, and working through our Workforce Futures project to understand what these tools mean for the people who support them.
What we've found is harder to summarise than either side of the loud debate suggests. AI can meaningfully increase independence and control for Disabled people. It can also cause real harm — to communities, to data sovereignty, to people who never signed up to be part of any of this.
This article isn't asking you to pick a side. It's asking whether you understand enough about AI to recognise when it's shaping your life — and to have a real say in how that goes.
Reading Time: Approx 12 minutes. This page includes a lot of information. Some people—especially those using screen readers—may prefer to read it in sections or take breaks.
Last updated: 11 May 2026
Tim Young
Ollie Goulden
Two people you should know about
I mentioned Tim at the top — here's the fuller picture.
Tim's smart home technology isn't a nice-to-have. It's what allows him to control his own environment. He can open his son's curtains, turn on music, unlock the door, call for help if he needs it. These aren't features. They're the difference between independence and not.
Then there's Ollie. He's a wheelchair user, and he also navigates the world with a neurodivergent brain. AI helps him in a different way to Tim — not just with physical access, but with the cognitive load of daily life. Managing support worker schedules. Wording a tricky email. Keeping on top of the executive functioning tasks that can stack up fast. For Ollie, having an AI tool available is like having a thinking partner on hand — it frees up his brain for the things that actually matter.
Tim and Ollie are who this technology is supposed to serve — and they're also who gets left behind when those conversations happen without us.
Keep them in mind as we go through the rest of this. Most of what follows only makes sense once you've got Tim's curtains and Ollie's inbox in your head.
You're probably already using it — without knowing
AI isn't new to your life. You're probably using it without realising. Recommendations on Netflix or Spotify. Your social media feed deciding what you see — and what you don't. Email filtering spam into a separate folder. Fraud detection on your bank account.
You didn't sign up for most of that. It was built into things you already use.
What's changed is that newer tools — like ChatGPT — make AI more visible. ChatGPT, Claude and others are examples of Generative AI: tools that learn patterns from huge amounts of existing text, images and music, then use those patterns to create new content that didn't exist before. Think of it like a remix machine — it doesn't copy, it generates, based on everything it's absorbed.
That's the thing that's new. But AI itself has been shaping your world for a while.
The concerns are real — and serious
All of that said: there are serious problems with AI that aren't going away. Here are the ones we think matter most.
The environmental cost
I want to tell you about my Tuesday, because the maths surprised me.
Last Tuesday I ate a hamburger for lunch, typed a question into ChatGPT to find a gluten free dinner idea for my family, and spent half an hour scrolling Instagram while I wound down in the evening. A pretty ordinary Tuesday. Also one that tells you something surprising about AI and the environment.
That hamburger: roughly 2,500 grams of CO₂ equivalent [i] (references are at the bottom of this page). The ChatGPT prompt: somewhere between 2 and 4 grams, depending on whose estimate you use [ii]. Which means I'd need to send hundreds of ChatGPT queries to match the carbon cost of that single burger.
The Instagram scroll? Thirty minutes generates around 45 grams of CO₂ — the equivalent of about 10 to 20 ChatGPT prompts [iii].
I'm not sharing this to say AI has no environmental cost. It does. I'm sharing it because when we talk about AI and harm, the picture is rarely as simple as the loudest voices suggest.
Each ChatGPT prompt also uses water — estimates range from about one-fifteenth of a teaspoon to several teaspoons’ [iv]. One prompt isn't the issue. Billions of them, every day, are.
The headline numbers are reassuring at the level of a single dinner question. They become much less reassuring when you scale them up to a planet's worth of dinner questions.
When the cost lands somewhere else
Carbon and water are one part of the story. But environmental cost isn't an abstract average — it shows up in specific places, to specific people. The data centres that power AI — including the ones that answered my dinner question — are already changing life for real communities.
In the Mexican state of Querétaro, where Microsoft, Google and Amazon have built large data centres, the communities surrounding them face strict water rationing, some households receive water just three days a week. In the village of La Esperanza, near a Microsoft facility, about fifty people fell ill in a hepatitis outbreak after water outages left residents unable to maintain basic hygiene [v]. Microsoft disputes that its facility is linked to the water shortages or the outbreak [vi].
In Ireland, data centres now consume more than 20% of the country's electricity — a figure projected to rise to 32% by 2026. South Dublin County Council passed a motion in late 2025 calling for a nationwide moratorium on new data centres, except those powered entirely by renewable energy, saying communities were being forced to absorb the costs of someone else's digital expansion [vii].
The companies behind this infrastructure profit. The communities dealing with the consequences mostly don't.
Misinformation — it’s already happening
AI makes it easy to produce convincing false content at scale. Not future-tense. Now.
We're talking about targeted scams that sound exactly like your bank. Political content engineered to change how you think and vote. In January 2024, a deepfake recording of President Biden's voice was used to tell voters in New Hampshire not to vote in the presidential primary. It was completely fabricated. Thousands of people received it before it was identified [viii]. The tools to do this are cheap, accessible, and getting better fast.
When you're not sure what's real, it's harder to make good decisions. That's not an accident.
Bias — when AI decides about you
AI learns from the past. And the past wasn't fair.
When AI is used to screen job applications, assess people for benefits, or decide what healthcare someone receives — it doesn't start from scratch. It starts from patterns built on historical decisions. If those decisions were biased (and they were), AI can repeat and scale that bias, without anyone in the room being aware it's happening.
For Tim, AI is the tool that gives him control over his environment. For Ollie, it's a thinking partner that lifts the cognitive load of executive functioning. For someone else — someone whose application gets quietly filtered out, or whose benefit claim is flagged by an algorithm — AI might be something that just happens to them, with no explanation and no recourse.
All of these are true at once. Whether AI works for you or against you depends a lot on which side of the system you're standing on.
This isn't just about whether you use AI
When people talk about AI, the conversation usually focuses on tools like ChatGPT — should you use them, are they safe, are they worth it?
But there are actually two separate things worth pulling apart.
Using generative AI is a choice. Whether you try out ChatGPT, Claude, Copilot — and what you use them for — is genuinely up to you.
Being affected by AI is not a choice anymore. AI is already built into systems that shape everyday life. Decisions about benefits and ACC claims. Job application screening. What appears at the top of your social media feed — including the half hour I was scrolling on Tuesday. How financial services assess your risk.
Whether or not you've ever typed a prompt, AI is already influencing the information you see, and in some cases, the decisions made about you.
So, the question isn't just “do I want to use AI?” It's also: “do I understand enough about it to know when it's affecting me?”
This is about control — and who gets a say
Some of the most important AI conversations aren't really about technology. They're about power: who controls data, who decides how it's used, who benefits, and who gets to say no.
In Aotearoa New Zealand, Māori have been leading conversations about data sovereignty — tino rangatiratanga over data. That means communities having the right to determine how information about their own people is collected, used, and who benefits from it. When AI systems are trained on Māori voices, images, te reo, or cultural knowledge without consent — and then used to make decisions about healthcare, welfare, or policing — that's not a tech problem. That's a power problem.
This is not just a Māori issue. But Māori have been at the forefront of naming it clearly, and this conversation matters for all of us. If the systems shaping what we see, who gets hired, and who receives support are built without our input — do we have any real say in how that goes? Do we understand it well enough to push back when something goes wrong?
At My Life My Voice, we think about this through a self-determination lens. AI can increase independence and access — Tim's curtains and Ollie's inbox are evidence of that. But it can also reinforce exclusion if the people it's supposed to serve aren't included in shaping it.
“Nothing about us without us” applies here as much as anywhere.
So why are people using AI?
With all of that said, it's completely valid to want nothing to do with AI. That is a real choice.
Tim Young demonstrating how he uses his voice to control his blinds at home.
But there's a reason AI is being taken up quickly — and for many people, it's not because they got swept up in hype. It's because these tools genuinely enable greater independence and self-determination for Disabled people and Tangata Whaikaha Māori.
For Deaf and hard of hearing people, real-time captions and transcription open up everyday conversations that used to be inaccessible. For blind and low-vision users, AI can describe an image, read a label, or guide someone through an unfamiliar street. For people with learning disabilities, it can break complex information into plain language. And for people like Tim, smart-home tools turn an unfamiliar environment into one you can actually control — opening curtains, turning on music, unlocking a door.
In the disability support sector, AI is also showing up in the workflow — drafting and updating support plans, summarising meeting notes into action points, producing Easy Read or plain language versions of documents, and helping people get started on a tricky email. (The meeting-notes one is the bit I quietly rely on most — it's saved my brain more times than I can count.)
These aren't conveniences. For a lot of people, these tools change what's possible.
The benefits are real. They just aren't evenly distributed — and they don't cancel out the risks.
The better question
It's tempting to want a simple answer. AI is good. AI is bad. Pick a side.
But that framing doesn't do justice to Tim, who relies on this technology to control his own home. Or to Ollie, who uses it to manage the cognitive load of daily life. Or to the families in Querétaro whose water was cut off while a data centre profited nearby. Or to the job applicant whose CV was filtered out by an algorithm trained on a biased past.
We're all part of the same story.
A more useful set of questions: Who is using AI, and for what purpose? Under what rules? Who benefits, and who is harmed? Who has a say?
When we ask those questions, instead of arguing about whether AI is “good” or “bad,” we start to see what actually matters.
Self-determination starts with information
Here's something simple that's easy to miss.
You can choose not to use generative AI — and that is a valid, important choice. But it doesn't mean AI won't affect you. It will still shape decisions made by organisations, systems, and platforms around you.
When we don't understand what's happening, we don't just opt out. We hand over control. That's true whether we're talking about an individual navigating a benefits system or a community whose data is being used to build tools they had no part in designing.
Self-determination isn't just about making choices. It's about making real choices — grounded in actual understanding of what you're choosing between.
Where does that leave you?
My Tuesday — the hamburger, the dinner question, the Instagram scroll — was full of decisions I made without really thinking about them. None of them felt like decisions about AI. But most of them were, in one way or another.
That's kind of the point.
There's no consequence-free option here. If you use AI, you can access its benefits — the captions, the smart-home control, the thinking partner — and you'll be more likely to notice when something is going wrong. That comes with responsibility: checking output, protecting your data, staying alert to its limits. If you don't use AI, you avoid some of those direct risks. But your benefit application is still being screened by it. Your feed is still being shaped by it. The decisions made about you don't pause while you opt out.
Either way, this technology is helping Tim open his son's curtains, and helping Ollie get through his inbox, and changing what running water looks like for families in Querétaro, and quietly deciding which CVs get read. It's already shaping the world Disabled people, Tangata Whaikaha Māori, and our wider communities are living in. Those communities deserve to be part of the conversation. So do you.
You don't have to become an expert. You don't have to start using tools you don't trust. You don't have to take a side in a debate that mostly happens at full volume on the internet. But understanding what's actually happening — and who gets to shape it — is how you keep a real choice in your hands instead of someone else's.
This isn't about liking or not liking AI. It's about whether you understand it well enough to have a say.
That choice only exists if you're informed enough to make it.
Ma te huruhuru ka rere ai te manu. With feathers, a bird can fly.
Information is your feathers. We hope this article gives you a few more of them.
👉 Want to understand how AI actually works? Read our Introduction to AI article, watch the video, or watch the NZSL video.
👉 Ready to think about using it safely? Read our AI Safety article, watch the video, or watch the NZSL video.
👉 Want to stay updated on our work? Register here.
References
[i][i]Carbon footprint of a beef hamburger (~2.5 kg CO₂e): The Carbon Cost of One Beef Burger, Plant Based Minutes (https://plantbasedminutes.com/article/the-carbon-cost-of-one-beef-burger); Beef Carbon Footprint, CO2 Everything (https://www.co2everything.com/co2e-of/beef). Note: estimates vary by methodology, with some studies placing the figure higher (around 2.8–3.3 kg CO₂e).
[ii]Carbon footprint of a ChatGPT query (~2–4g CO₂): The Real Carbon Footprint of ChatGPT: 4.32g CO₂ Per Query, Piktochart (https://piktochart.com/blog/carbon-footprint-of-chatgpt/) — based on November 2022 query volumes; What's the carbon footprint of using ChatGPT? and August 2025 update, Hannah Ritchie (https://hannahritchie.substack.com/p/carbon-footprint-chatgpt; https://hannahritchie.substack.com/p/ai-footprint-august-2025) — Ritchie estimates around 2–3g per query and notes per-query energy use may be even lower than previously thought.
[iii]Carbon footprint of 30 minutes of Instagram scrolling (~45g CO₂): The carbon impact of Instagram app features, Greenspector (https://greenspector.com/en/6168-2/) — newsfeed scrolling measured at 1.549 gEqCO₂/minute.
[iv]Water use per AI query: Sam Altman: ChatGPT queries consume 0.000085 gallons of water, Data Centre Dynamics (https://www.datacenterdynamics.com/en/news/sam-altman-chatgpt-queries-consume-034-watt-hours-of-electricity-and-0000085-gallons-of-water/); How much water does AI consume? The public deserves to know, OECD.AI (https://oecd.ai/en/wonk/how-much-water-does-ai-consume) — note: estimates vary considerably depending on methodology, and Altman's figure does not include training costs.
[v]Community impacts of data centres in Mexico: Resistance blooms in Mexico's data centre valley, Context/TRF (https://www.context.news/ai/long-read/resistance-blooms-in-mexicos-data-centre-valley); Querétaro Data Center Boom Triggers Water, Power Backlash, Mexico Business News (https://mexicobusiness.news/cloudanddata/news/queretaro-data-center-boom-triggers-water-power-backlash).
[vi]Microsoft's response: Microsoft denies Mexico data center linked to water shortages, local illnesses, and power outages, Tom's Hardware (https://www.tomshardware.com/tech-industry/microsoft-denies-mexico-data-center-linked-to-water-shortages-local-illnesses-and-power-outages-stomach-bugs-and-even-hepatitis-reported-in-region-as-1-5-gigawatt-ai-data-center-buildout-looms).
[vii]Data centres and electricity in Ireland: Data Centres in Ireland: The State of Play, IIEA (https://www.iiea.com/blog/data-centres-in-ireland-the-state-of-play); South Dublin County Council passes motion calling for moratorium on new data centres, TheJournal.ie (https://www.thejournal.ie/data-centres-ireland-5-6812073-Sep2025/).
[viii] Biden deepfake robocall, New Hampshire 2024: FCC fines company behind Biden deepfake robocall, NBC News (https://www.nbcnews.com/politics/2024-election/fcc-fines-company-behind-biden-deepfake-robocall-rcna148593) — call monitoring service Nomorobo estimated between 5,000 and 25,000 calls were placed.
This article was written by our team, with support from AI. The ideas, analysis, and final decisions are our own. We reviewed the information and references carefully before publishing, because AI can make mistakes.
Supported by the Workforce Futures Fund |Tahua Rāngaimahi Anamata