Wonder
The possibilities of AI nine areas, with results that already exist
This is the hopeful half of the map. Every seed below is real, and names its source and its kind. Each area holds far more than the few shown here, and none of it arrives by itself.
EthicalEconomicEnvironmentalEducationalEmotionalEpistemicEquityExpressiveExistential
- 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
Ethical consent that works
Every artist whose work teaches a machine is asked, credited and paid.
30,000+
artists take part in one opt-in licensing program that now shares 25% of net revenue with them. Reported
$22 million
earned by more than 10,400 people who chose to share their voices on one platform by May 2026, by the company's count. Reported
$1.5 billion
court-approved settlement in July 2026 for about 482,000 books an AI company downloaded from pirate libraries, about $3,000 per work before fees. It pays for how the books were obtained. 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. Reported
Over 92%
of the time, the free tool Glaze disrupted AI style mimicry in its developers' tests with more than 1,000 artists. An independent 2025 study found that simple techniques can get around this kind of protection. Measured
10 billion
pieces of AI-generated content had been watermarked by Google's SynthID by May 2025, by the company's own count. Reported
It takes Licensing that reaches the individual creator, and labels that travel with the work.
Economic shared prosperity
The time and wealth these tools create are shared, so more people can afford to make things.
11 more hours
a week of creative work reported by artists in Ireland's basic income pilot than by a lottery control group. Measured
€1.39
returned per euro spent on that pilot, by a government-commissioned cost-benefit estimate. Estimate
15%
more issues resolved per hour across 5,172 support agents given an AI assistant. Less experienced workers gained most. Measured
40% faster
and rated 18% better: short writing tasks by professionals randomly given ChatGPT, in an experiment with 453 people. The gap between stronger and weaker writers narrowed. Measured
It takes Policy. No such scheme is yet funded by AI's gains.
Environmental a better-watched planet
AI helps us see storms and floods sooner, and do more with less.
97.2%
of 1,320 forecast targets on which an AI weather model beat the leading conventional ensemble forecast. Measured
5 days ahead
AI forecasts of extreme floods on ungauged rivers matched or beat a leading system's same-day reliability. They run in over 80 countries. Measured
33-fold
drop in the energy of a median Gemini text prompt in one year, by Google's own unreviewed figures. Estimate
1,000 times less energy
per forecast is what Europe's weather center reports for its AI forecasting system, in operation since February 2025. By the center's account it is up to 20% better on some measures, including cyclone tracks. Reported
72% to 76%
of the world's industrial fishing vessels do not appear in public tracking systems. Deep learning on satellite images of coastal waters from 2017 to 2021 revealed them. Measured
54% fewer contrails
on 70 test flights where pilots used AI contrail forecasts, judged from satellite images, at the cost of 2% more fuel on those flights. Company-reported and not peer reviewed. Reported
It takes Clean power, published energy figures for every tool, and warnings that reach the people in harm's way.
Educational a patient teacher for all
Every learner has a patient tutor, and every teacher has help.
Over double
the median test gains with a well-designed AI tutor, against an active-learning class, in a randomized trial of 194 Harvard physics students. Measured
4 points
more likely to master a topic: students whose human tutors had an AI assistant. The gain was 9 points with lower-rated tutors. Not yet peer reviewed. Measured
0.31 standard deviations
higher scores for Nigerian secondary students after six weeks of teacher-guided sessions with an AI tutor, in a World Bank randomized trial. Only 759 of 1,328 students sat the final test, and the study is an unreviewed working paper. Measured
66.2% vs 60.7%
Students tutored by an AI that human tutors supervised solved a new problem on the next topic slightly more often than with human tutors alone. The difference's uncertainty range includes zero, and the study is a preprint by the AI's developers. Measured
It takes Guardrails. In another trial a plain chatbot left students worse off once it was taken away.
Emotional connection and care
Tools help us care for one another, and point us back toward each other.
19.6%
higher on an automated empathy measure: peer supporters' replies written with AI feedback. It was 38.9% for those who said support was hard. Measured
51% lower
depression symptoms reported at eight weeks by users of a monitored therapy chatbot in a randomized trial. A waitlist group improved less. Measured
15% vs 6%
rise in referrals at NHS talking-therapy services using an AI self-referral chatbot, against comparison services, with larger rises among nonbinary people and ethnic minority groups. Observational, and the authors work for the chatbot's maker. Measured
15 randomized trials
pooled in a 2023 meta-analysis found that AI conversational agents reduced symptoms of depression and distress, with no significant effect on overall wellbeing. The confidence intervals were wide. Estimate
15 minutes
with an AI companion eased loneliness about as much as talking with a person. The relief did not build over a week of daily use. Measured
It takes Supervision, and designs that strengthen human bonds.
More on AI companions and mental health, with the other side
Epistemic truth and common ground
Tools help us check what is true and find where we agree.
About 20%
drop in belief in a chosen conspiracy theory after a conversation with an AI, among 2,190 believers. It lasted at least two months. Measured
22.8% to 38.6%
Share of small groups whose members all took the same side, before and after AI-mediated discussion, in one UK experiment. Measured
70%
of the time, crowd raters preferred an AI system's suggested source over the existing one, for the Wikipedia citations it judged most likely to be unverifiable. Measured
24% to 28%
more accurate predictions from people who could consult an advanced AI forecasting assistant, against a control group using a weaker model, in an experiment with 991 people. One outlier question strongly affected the results. Measured
It takes Systems built for accuracy and agreement instead of attention. No AI has yet stopped a war.
Equity access for everyone
Every language and community is served, and owns what it shares.
200 languages
translated by one model released free for non-commercial use, many with few written resources. Measured
92%
reported accuracy of te reo Māori speech recognition built by a Māori-led broadcaster, which keeps the data under a licence requiring its permission for any use. Reported
16% fewer
diagnostic errors, and 13% fewer treatment errors, by clinicians randomly given an AI safety-net tool across about 40,000 visits in 15 Kenyan clinics. Unreviewed preprint, partly written by OpenAI staff. Measured
4% to 21% fewer
poor people were wrongly left out when Togo targeted COVID cash aid with machine learning on phone and satellite data, compared with targeting by geography. About half of the poorest were still missed. Estimate
It takes Communities holding their own data, tools and say.
Expressive more voices, new forms
More people can make art, and lost voices and works return.
10.7%
more novel: stories by less-creative writers who were offered up to five AI ideas. Measured
2,000+ characters
read with machine learning from a Herculaneum scroll carbonized in AD 79, without unrolling it. Reported
97.5%
word accuracy for a man with ALS conversing through a brain implant with AI decoding, voiced to sound like he did before the illness. Measured
Nearly 60%
of a man's words were understood by listeners when a brain implant turned his attempted speech into a synthesized voice with a delay of about one fortieth of a second, against 4% without it. He has ALS, and he could also change intonation and sing short melodies. One participant. Measured
A 1970s cassette
held John Lennon's voice tangled with a piano. Machine learning separated them, and the Beatles' 'Now and Then' won the 2025 Grammy for Best Rock Performance. The singing is Lennon's own. Reported
42 recordings
of Randy Travis's earlier vocals trained an AI model that, with his family's blessing, let the country singer release a new song in 2024, eleven years after a stroke took his singing voice. Another singer supplied the base vocal. Reported
It takes Keeping craft alive, so that gains for each of us do not narrow what all of us make.
Existential longer lives, a wiser hand
Illness is found earlier, lives last longer and better, and we learn to steer powerful tools together.
29% more
cancers found by AI-supported mammography (6.4 vs 5.0 per 1,000) in a randomized trial of 105,934 women, with no rise in false positives and 44% less reading work. Measured
200 million
predicted protein structures in the AlphaFold database, used by over 3 million researchers by its maker's count. Reported
Nearly 99%
of banned ozone-depleting substances phased out under the Montreal Protocol. The world has restrained a technology before. Measured
100+ experts
wrote the 2026 International AI Safety Report, with a panel nominated by over 30 countries and intergovernmental organizations. Reported
98 mL gained
in lung capacity over 12 weeks on the highest dose of rentosertib, a lung-fibrosis drug whose target and molecule were found with generative AI, while capacity fell 20 mL on placebo. The trial had 71 patients, was designed to test safety, and the drug is not approved. Measured
Millions of features
were extracted from inside one AI model by researchers, many of them interpretable, and turning one up changed the model's behavior. It is an early step toward reading AI from the inside. Anthropic makes Claude, the assistant that helped build this site. Measured
Over 2,000, then almost none
More than 2,000 nuclear test explosions took place between 1945 and 1996. Since the test-ban treaty opened for signature, only India, Pakistan and North Korea have tested. The treaty is still not in force, and in 2025 the US said it would prepare to resume testing. Measured
80%
of US adults said in a 2025 Gallup survey that the government should keep rules for AI safety and data security even if that slows AI development. Survey
It takes Independent trials, fair access, and rules that keep pace with the tools.
More on AI safety, power and the long view, with the other side
The world this points toward
A seed is small. What it becomes depends on the soil, the season and who tends it. So it is with these tools. Picture them carrying the tedious weight so that people have more time for each other... medicine that finds illness early, a patient teacher for every child, every language heard, artists asked and paid and making work only they could make. Picture sharing that comes easily because fewer people fear going without, and arguments that end in common ground. None of this is promised. Each piece has already begun somewhere.
Is it feasible?
No one knows. The evidence does show that each piece already exists somewhere at small scale, and that people chose to build it that way. A future like this would be grown one decision at a time, by companies, governments and each of us.
What stands between here and there
- Almost every seed here was planted on purpose, with public money, consent, guardrails or supervision. None grew from a tool left to itself.
- Most results are early. Many are short trials, single sites, one company's product or a company's own count.
- The same tools carry the risks on the companion sheet. A tutor with guardrails helped, and a plain chatbot harmed.
- Benefits reach the well-resourced first unless access is built in.
- Finding cancer earlier is not yet proof of longer lives, and no AI-designed drug has been approved.
Where each of us can start
- Choose tools and makers that ask, credit and pay.
- Say how your work was made.
- Keep practicing your craft, and teach it.
- Share what you can, and support pay and security for those who cannot yet.
- Ask of every new use whether it brings people closer together.