Are we there yet?

toward AI-led advancement

A judgment-based index, with low confidence. Not a scientific measure or a countdown.

Updated: December 31, 2022/1,046 dates/14 fields/85 measurable outcomes

What is the Singularity?

The technological singularity is a transition in which technological advancement across many fields compounds faster than human-directed research could sustain or humans can reliably predict.

It is driven by reinforcing loops, in which each advance makes further advances easier, and its strongest engine is artificial intelligence that first matches and then exceeds human capability.

It is a period we could be partway into, not a single moment. The definition does not claim a date, or that it is certain to happen.

How it could come about Full definition and sources

The fields we track

Each field has a numbered milestone on the graph, at its own estimate. The headline 0% is the average of all 14.

Explore the fields

Advancements

Advancements are concrete, real-world products, events and results. Each one is evidence of progress on a measurable outcome, and together they show how far AI has carried a field toward the Singularity.

Mathematics

Software

AI research

Computing hardware

EPower, performance, and area0%

NVIDIA PrefixRL circuits are up to 25% smaller at the same delayJuly 8, 2022 · shown at the 2022 starting point

Robotics

Materials

Chemistry

Biology

Drug discovery

Medicine

Energy

AFusion control0%

DeepMind and EPFL control TCV tokamak plasma with deep reinforcement learningFebruary 16, 2022 · shown at the 2022 starting point

DOperational efficiency0%

DeepMind AI raises data-centre cooling savings to about 30%August 17, 2018 · shown at the 2022 starting point
DeepMind and Trane RL control cuts cooling energy about 9% and 13% at two sites11 November 2022 · shown at the 2022 starting point

Weather and climate

BComputation0%

FourCastNet generates a week-long global forecast in under 2 seconds22 February 2022 · shown at the 2022 starting point

Physics

AEquation recovery0%

AI Feynman recovers all 100 benchmark physics equationsApril 15, 2020 revised preprint · shown at the 2022 starting point

Education

The End?

Probably not. The Singularity is not a finish line. Reaching 100% is where the curve stops being ours to draw, not where progress stops.

Once advancement compounds on its own, there is no obvious reason for it to end. It could go on forever, or at least for as long as humankind is here to live with it.

What waits up there is guesswork. The things people have dreamed about for generations, such as flying cars, teleportation and immortality, sit on the list beside things no one has imagined yet. Some may arrive, some may prove impossible, and some may simply stop mattering.

This is speculation based on the increasing speed of technological advancements humanity is experiencing.

Dates after this point are based on our rate of advancement.

Understand the idea

How could it
come about?

No single breakthrough brings the singularity about. It builds through loops in which each advance makes the next one easier.

AI improving AI. AI tools help build more capable AI, which could lead to systems that match human ability and then exceed it, each helping to build its successor.

Progress compounding within a field. An advance in medicine, materials or energy opens the way to further advances in the same field, with or without AI involved.

Progress compounding across fields. Better chips, energy, materials and robots feed AI and one another, and AI in turn speeds each of them up.

Progress has always built on itself. What would mark a singularity is the pace: acceleration that is sustained, shows up in many fields at once, and no longer depends on people to choose, do and check the work. Whether that happens, and when, remains an open question.

Read the full definition and its sources

People lead

We set the questions, build the tools and check the results.

AI contributes

AI takes on useful work, while people guide and verify it.

AI advances independently

AI chooses worthwhile work, validates it and keeps improving the process.

These are stages of the idea, not dated predictions. Our evidence includes useful assistance and bounded autonomous tasks; it does not establish that the singularity has begun.

Follow the evidence

0%summary estimate

Helping people learn skills they can use on their own.

Explore 6 outcomes

Read the evidence and view the graph

Field percentages summarize their outcome estimates. They are not the share of jobs replaced, or proof that a field is that far from full autonomy.

Keep the number in perspective

An estimate you
can look inside.

The headline is a guide to our research, not a verdict on humanity’s future.

What does 0% mean?

We score each outcome against five questions: can AI do the work, handle unfamiliar problems, choose useful goals, check its results and keep improving without people managing the process?

We average the outcomes within each field, then give all 14 fields equal weight. The result is 0.00%, shown as 0%. This is a new summary index built from the existing research scores.

See the calculation

Field	Outcomes	Outcome average	Headline weight
Mathematics	6	0.00%	1 of 14 equal shares
Software	6	0.00%	1 of 14 equal shares
AI research	7	0.00%	1 of 14 equal shares
Computing hardware	6	0.00%	1 of 14 equal shares
Robotics	6	0.00%	1 of 14 equal shares
Materials	6	0.00%	1 of 14 equal shares
Chemistry	6	0.00%	1 of 14 equal shares
Biology	6	0.00%	1 of 14 equal shares
Drug discovery	6	0.00%	1 of 14 equal shares
Medicine	6	0.00%	1 of 14 equal shares
Energy	6	0.00%	1 of 14 equal shares
Weather and climate	6	0.00%	1 of 14 equal shares
Physics	6	0.00%	1 of 14 equal shares
Education	6	0.00%	1 of 14 equal shares

We calculate with unrounded values and round only the display. Field averages are summaries, not validated field-wide completion measurements.

What are the limits?

The starting point is the period before ChatGPT’s public release in 2022. The endpoint is sustained advancement managed entirely by AI. The distance between those points is not a scientifically established scale.

Missing evidence receives no progress credit, so it can pull the estimate down. A score of zero can mean we do not yet have a qualifying measurement. Small changes in these judgments are not precise measurements of real-world change.

The latest saved assessment is December 31, 2022. Earlier graphs are retrospective estimates, not measurements recorded at the time. This page does not update itself with new research.

Read the full scoring approach