Common questions about AI answered from the evidence
Short answers to what people ask most. Each figure names its kind and its source, and each answer links to the fuller picture with the other side beside it. Last reviewed October 11, 2026.
Is AI taking jobs?
Some freelance fields have lost work. Broad national data so far show small effects.
In a UK survey, 26% of illustrators and 36% of translators who answered said they had already lost work to generative AI.
Image-related freelancers on Upwork earned 9.4% less a month after image generators launched, relative to other freelancers.
A study of Danish government earnings records found no measurable effect on earnings or hours in 11 exposed occupations two years after ChatGPT's launch. It is a working paper, not yet peer reviewed.
Through mid-2025, a Yale and Brookings analysis found no discernible disruption of the wider US labor market.
How much energy and water does AI use?
Data centers used about 1.5% of the world's electricity in 2024, and that use is projected to more than double by 2030.
By Google's own figure, a median Gemini text prompt uses about 0.24 Wh. Video is far heavier: lab tests of open models put one short clip anywhere from under 1 Wh to about 1,000 Wh, and commercial video tools publish no figures.
Two in three new US data centers built or planned since 2022 are in areas of high water stress.
Does AI help students learn, or does it weaken their skills?
The design of the tool seems to decide it.
Students using plain GPT-4 for math practice scored 48% better in practice, then 17% worse than classmates on an exam without AI. A tutor with guardrails largely avoided the harm.
In a randomized trial of 194 Harvard physics students, a well-designed AI tutor more than doubled median test gains against an active-learning class.
A widely cited 2025 meta-analysis reporting large learning gains from ChatGPT was retracted in April 2026.
Are AI companions and therapy chatbots safe?
The evidence points both ways, and the people most at risk need the most care.
In a US survey, 72% of teens had tried an AI companion, and 52% used one at least a few times a month.
In scripted single-message tests, five commercial therapy chatbots responded appropriately about 50% of the time, against 93% for 16 human therapists.
In a randomized trial of a monitored therapy chatbot, users reported 51% lower depression symptoms at eight weeks.
New York and California now require companion apps to have suicide and self-harm protocols and to tell users they are not talking to a human.
Can people tell AI-made images, video and music from the real thing?
Mostly not. People correctly identify deepfakes 55.5% of the time, which is not reliably better than chance.
In an eight-country survey of 9,000 adults commissioned by Deezer, 97% failed a blind test to pick out fully AI-generated songs, and 80% wanted fully AI-made music labelled. The test used only three tracks.
Was AI trained on artists' and writers' work without permission?
Many AI systems learned from books, songs, images and conversations whose makers were never asked. In an eight-country survey, 73% of adults said it is unethical for AI firms to use copyrighted music without the artist's approval.
Courts are still working out the law. In July 2026 a court approved a $1.5 billion settlement over about 482,000 books that Anthropic had downloaded from pirate libraries. A judge had ruled in 2025 that training on lawfully acquired books was fair use. Anthropic makes Claude, the assistant that helped build this site.
Licensing has begun. In September 2025 Sweden's STIM launched what it calls the world's first collective AI music licence.
Is AI-generated music flooding streaming services?
On peak days in June 2026, more than 50% of new uploads to Deezer were fully AI-generated. They drew only 1 to 3% of streams, and by the company's own detection up to 85% of streams on fully AI-generated tracks in 2025 were fraudulent.
Spotify says it removed 75 million spammy tracks in the 12 months to September 2025.
Since January 2026, Bandcamp has banned music generated wholly or in substantial part by AI.
Is AI biased?
Measured tests have found bias. In hypothetical trials, language models chose the death penalty 27.7% of the time for speakers of African American English, against 22.8% for speakers of Standard American English.
In 2023, women made up 3% of one image generator's pictures of a judge, against 34% of real US judges.
In a 2023 test, 61% of 91 essays by non-native English writers were labelled AI-written by seven AI-text detectors. Detectors have changed since then.
Could AI cause human extinction?
Nobody knows. Every figure here is someone's judgment.
In a 2023 survey with a 15% response rate, AI researchers' median guess was a 5% chance that AI causes human extinction or severe disempowerment.
In 2022, superforecasters put the same risk by 2100 at 0.38%, and domain experts put it at 3%.
The 2026 International AI Safety Report gives no probability.
What good has AI done so far?
In a randomized trial of 105,934 women, AI-supported mammography found 29% more cancers, with no rise in false positives.
AI forecasts of extreme floods on ungauged rivers, made five days ahead, matched or beat a leading system's same-day reliability. They run in over 80 countries.
A man with ALS conversed through a brain implant with AI decoding at 97.5% word accuracy.
Machine learning read more than 2,000 characters from a Herculaneum scroll carbonized in AD 79, without unrolling it.
Most of these results are early, and almost every one was built on purpose, with consent, guardrails or supervision.
Who is regulating AI?
Since 2 August 2026, the EU AI Act has required chatbots to tell people they are talking to an AI, and generative AI providers to mark their output in a machine-readable way.
California's SB 53, in effect since January 2026, requires the largest AI developers to publish safety frameworks, report critical safety incidents to the state and protect whistleblowers.
In December 2025 the UN General Assembly voted 164 to 6 on autonomous weapons. There is still no binding treaty.
How can artists and musicians use AI ethically?
A Shared Ethic on this site offers eight commitments: human first, consent, credit and pay, say how it was made, keep the craft alive, purpose before profit, use energy with intention, and respect. It also sorts tools into three tiers by how their training data was gathered.
Who made this site, and was AI used?
Michael Hummel conceived and directed it. Claude, an AI assistant made by Anthropic, did the research, drafting, design and code under his direction, and the images were generated with AI.
- Measured counted or experimental data
- Survey what people report or believe
- Estimate a judgment or model, often a company's own
- Forecast a projection of the future
- Reported a company's, agency's or journalist's own account