01 The statement
Our Mission
One paragraph. The explanation of what drives our work.
The most powerful AI companies on Earth are racing to build systems smarter than humans, and many of the people building them expect to succeed within the decade. No one yet knows how to guarantee that a mind more capable than ours will reliably want what we want. That is the alignment problem, and it remains unsolved. A failure at superhuman scale may permit no second attempt: a system more intelligent than its overseers cannot simply be corrected after the fact, and competitive pressure rewards the labs that worry least. We believe this is the defining technical and political problem of our generation. We also believe it can be solved, through alignment research, rigorous evaluation, serious policy, and many more capable people treating it as their responsibility. Rice AI Alignment exists to do this work: to study the risk honestly, to train the people who will reduce it, and to help ensure that advanced AI goes well for everyone.
— Rice AI Alignment · Rice University · Houston, TX
// AI Safety is no longer a fringe view: “Mitigating
the risk of extinction from AI should be a global priority alongside other
societal-scale risks such as pandemics and nuclear war.”
— the Statement on AI Risk (2023),
signed by the heads of OpenAI, Anthropic & Google DeepMind and hundreds of leading researchers
For the papers, books, and courses behind this paragraph, start with our resources.
02 If we fail
This is what losing looks like.
Below is a simulation of the decade after alignment fails: an advanced system pursues objectives of its own and converts the surface of the Earth into the infrastructure those objectives require. For further reading and inspirations see ai-2040.com
01
Why datacenters?
Compute is the physical substrate of machine intelligence, and almost any open-ended objective is easier to achieve with more of it. A sufficiently capable system optimizing for such an objective can be expected to acquire resources: land, energy, and matter. Advocates like Eliezer Yudkowsky and others stated the concern precisely: “The AI does not hate you, nor does it love you, but you are made out of atoms which it can use for something else.”
02
Why so fast?
For most of the run little appears to change; then the conversion completes within a few years. That is the signature of an intelligence explosion: once AI systems can automate AI research itself, each generation produces a more capable successor on a shorter timescale. Daniel Kokotajlo, a former OpenAI researcher who resigned over the industry’s handling of these risks, charts a decade of this kind in AI 2027. Effective oversight has to exist before that acceleration begins.
03
Why it is preventable
The simulation depicts a failure of alignment and governance, not an inevitability. The levers that change the outcome already exist: alignment research, interpretability, dangerous-capability evaluations, and governance that treats the problem as real. What the field lacks most is people. Preparing them, at Rice and beyond, is the purpose of this organization.
The outcome above is the one we are working to prevent. If you want to contribute to that work, join us.