---
title: "The Horizon: Where AI Is Heading in the Next 5 to 10 Years"
description: "How fast AI is improving, what changed in the last year, and what credible researchers and builders expect for health, learning, work and science by the early 2030s."
author: Graham
date: 2026-10-15
url: https://powerofai.ca/the-signal/horizon
---

# The Horizon: Where AI Is Heading in the Next 5 to 10 Years

How fast AI is improving, what changed in the last year, and what credible researchers and builders expect for health, learning, work and science by the early 2030s.

Reviewed October 15, 2026.

AI is improving faster than almost any technology before it. Here is the evidence, what changed in a single year, and what the people building and studying it expect by the early 2030s. Every figure links to its source.

## How fast is AI actually moving?

Measurements from independent researchers, official statistics and the companies themselves. Benchmarks are imperfect, so read the trend, not the decimal.

- **~4 months: for AI agents to double the length of task they can finish.** METR measures how long a task (mostly software work) an AI agent can complete on its own. That length has doubled about every 4.3 months since 2023, and faster since 2024. [METR, January 2026](https://metr.org/blog/2026-1-29-time-horizon-1-1/)
- **12+ hours: of expert work, completed about half the time.** Early 2026 models finished tasks that take a human expert around 12 hours, up from about 3.5 hours a year earlier. At high reliability the horizon is shorter, around 1 to 3 hours. [METR results](https://en.wikipedia.org/wiki/METR)
- **19.2%: of Canadian businesses now use AI.** Up from 12.2% a year earlier, according to Statistics Canada’s second-quarter 2026 survey. [Statistics Canada, June 2026](https://www150.statcan.gc.ca/n1/pub/11-621-m/11-621-m2026010-eng.htm)
- **900M: people use ChatGPT every week.** OpenAI’s official figure from February 2026, up from 800 million in October 2025. That is one product among many. [TechCrunch, February 2026](https://techcrunch.com/2026/02/27/chatgpt-reaches-900m-weekly-active-users)
- **5×: more computing power for AI each year.** The compute used to train leading models has grown about five times a year since 2020. Better chips and bigger investments compound. [Epoch AI trends](https://epoch.ai/data/trends)
- **40×: cheaper each year for the same ability.** Reaching GPT-4’s level on PhD-level science questions got about 40 times cheaper per year. Across tasks, price drops ranged from 9× to 900× a year. [Epoch AI, 2025](https://epoch.ai/data-insights/llm-inference-price-trends)
- **Phase III: for a drug discovered with generative AI.** In September 2026, the first patient was dosed in a Phase III trial of rentosertib for a lung-scarring disease. It is not approved yet, but it is the furthest an AI-discovered drug has come. [Insilico Medicine](https://insilico.com/news/isn1009261-insilico-medicine-doses-first-patient-genesis-ipf-3)
- **42/42: reported perfect scores at the Math Olympiad.** At the 2026 International Mathematical Olympiad, two AI systems reported perfect scores, a result only seven of 666 human contestants matched. A year earlier the best AI score was 35. [AFP via Malay Mail](https://www.malaymail.com/amp/news/tech-gadgets/2026/07/23/huawei-xiaohongshu-ai-storm-olympiad-join-maths-elite-with-perfect-100pc-score/228720)

## Things that changed in twelve months

- A year ago: The best AI agents could finish software tasks that take an expert about 3.5 hours, about half the time. Now: They manage tasks of around 12 hours or more. [METR](https://metr.org/blog/2026-1-29-time-horizon-1-1/)
- A year ago: The best AI scored 35 out of 42 at the International Mathematical Olympiad, just enough for gold. Now: AI systems reported perfect scores, and a reasoning model disproved a conjecture Paul Erdős posed in 1946. [TechCrunch](https://techcrunch.com/2026/05/20/openai-claims-it-solved-an-80-year-old-math-problem-for-real-this-time/)
- A year ago: About one in eight Canadian businesses used AI to deliver goods or services. Now: Nearly one in five do. [Statistics Canada](https://www150.statcan.gc.ca/n1/pub/11-621-m/11-621-m2026010-eng.htm)
- A year ago: An AI-discovered drug for lung scarring had promising early-stage results. Now: It is in a Phase III trial with about 320 patients. [Insilico Medicine](https://insilico.com/news/isn1009261-insilico-medicine-doses-first-patient-genesis-ipf-3)
- A year ago: Free AI chat apps mostly answered in text. Now: OpenAI’s newest model is free to everyone and can answer with working charts, forms and small tools. [OpenAI](https://openai.com/index/gpt-6-for-everyone/)

## Where it could be by the early 2030s

These are views about the future from people who build and study AI, attributed and linked. Some are bold, some are cautious. Together they describe a decade that could change how we learn, heal, work and discover.

### Health and medicine

Today: An AI-discovered drug is in a late-stage trial, and OpenAI says more than 230 million people ask ChatGPT health questions every week.

Ahead: Faster drug discovery, earlier diagnosis and expert-level medical guidance available to far more people.

[Dario Amodei, Anthropic (2024)](https://www.darioamodei.com/essay/machines-of-loving-grace): AI-enabled biology could compress the progress of the next 50 to 100 years into 5 to 10, while warning these predictions will be wrong in the details.

[Epoch AI (2025)](https://epoch.ai/blog/what-will-ai-look-like-in-2030): A more measured view: few, if any, drugs approved by 2030 will have benefited from today’s AI tools.

### Learning and teaching

Today: In a World Bank pilot in Nigeria, six weeks of AI-supported tutoring produced learning gains comparable to nearly two years of typical schooling.

Ahead: A patient, personal tutor for every student, and more time for teachers to do the human part of teaching.

[Bill Gates (2025)](https://www.beckershospitalreview.com/disruptors/ai-will-end-scarcity-of-medical-expertise-bill-gates-says/): Over the next decade, great tutoring and great medical advice will become free and commonplace.

[World Bank (2025)](https://blogs.worldbank.org/en/education/From-chalkboards-to-chatbots-Transforming-learning-in-Nigeria): The Nigeria pilot outperformed 80% of comparable education programs.

### Work and small business

Today: Nearly one in five Canadian businesses use AI, and Ottawa is funding adoption and 10,000 student placements.

Ahead: Many tasks inside jobs will change. New roles appear as others shrink, and the people who learn the tools early gain the most.

[World Economic Forum (2025)](https://weforum.org/press/2025/01/future-of-jobs-report-2025-78-million-new-job-opportunities-by-2030-but-urgent-upskilling-needed-to-prepare-workforces/): Expects 170 million jobs created and 92 million displaced by 2030 across all trends, with nearly 40% of job skills changing.

[Epoch AI (2025)](https://epoch.ai/blog/what-will-ai-look-like-in-2030): Projects a 10 to 20% productivity improvement within tasks that AI touches by 2030.

### Science and discovery

Today: AI agents have flagged an unexplained system in bacterial DNA, and AI-produced math results are being published with machine-checkable proofs.

Ahead: AI working alongside scientists as a tireless research partner: reading everything, testing ideas and proposing experiments.

[Demis Hassabis, Google DeepMind (2025)](https://www.etvbharat.com/en/!technology/ai-could-cure-all-diseases-in-10-years-says-google-deepmind-chief-hassabis-enn25042202229): Within 10 to 15 years we may have “a real crack at solving all disease,” and drug development could shrink from years to months.

[AI Futures Project (2026)](https://blog.aifutures.org/p/q1-2026-timelines-update): Moved its median forecast for AI that can automate coding earlier, to between mid-2028 and mid-2030.

### Everyday assistants

Today: The newest models are free to use, the smarter Siri works in English in Canada, and assistants are arriving in glasses, TVs and speakers.

Ahead: Assistants that can carry out real errands and projects for you, at a cost that keeps falling.

[Sam Altman, OpenAI (2025)](https://blog.samaltman.com/the-gentle-singularity): In the 2030s, intelligence and energy will become wildly abundant, and the cost of intelligence should approach the cost of electricity.

[Metaculus forecasters](https://www.metaculus.com/questions/5121/date-of-artificial-general-intelligence/): The community’s median forecast for general AI, under a demanding definition, sits around 2031.

### Accessibility

Today: Be My Eyes connects more than a million blind and low-vision people with AI that describes the world around them, and launched a foundation to keep it free.

Ahead: Live description, translation and reading aloud are becoming standard features of phones, glasses and speakers.

[Be My Eyes (2026)](https://doubletaponair.com/be-my-eyes-surpasses-1-million-users-and-launches-foundation-to-keep-ai-tools-free/): More than 1 million users in 163 countries, with support to keep AI assistive tools free.

## Optimism works best with clear eyes

- AI keeps improving rapidly but unevenly, failing at some simple tasks, and safeguards are not keeping pace with the technology. [International AI Safety Report 2026, chaired by Yoshua Bengio of Mila](https://www.aap.com.au/aapreleases/cision20260203ae77099)
- AI is “like a tsunami hitting the labor market,” and the tasks it removes are often the ones that make up entry-level jobs. [Kristalina Georgieva, IMF (2026)](https://fortune.com/2026/01/23/imf-chief-warns-ai-tsunami-entry-level-jobs-gen-z-middle-class/)
- Real-world change may come more slowly than capability: new technologies spread through the economy over decades, not years. [Arvind Narayanan and Sayash Kapoor, Princeton (2025)](https://knightcolumbia.org/content/ai-as-normal-technology)
- Estimates AI may raise overall productivity by less than 1% over ten years, far below popular forecasts. [Daron Acemoglu, MIT (2024)](https://www.nber.org/papers/w32487)

## The best way to understand the future is to use it today

- [Try one useful task](/start): Pick something from your own week and see what AI does with it.
- [Find your path](/for/teachers): Guides for teachers, students, employees, small business owners, families and everyday life.
- [Learn the habits that matter](/learn): Context, follow-ups and checking: the skills that make AI useful.
- [Find the right tool](/finder): Match the job to the app, from chat to agents that take action.

Forecasts are attributed to the people and organizations who made them. They are views about the future, not promises.