Technology by humanity for humanity
Powerful AI could put humanity’s future beyond our control
Leading AI researchers and the heads of major AI companies have warned that AI poses a risk of human extinction. They have called for reducing this risk to become a global priority alongside pandemics and nuclear war.
Evidence
The Good Tech Institute is conducting key research into Primary Intelligence. We have made significant progress towards intelligence that is deterministic, fully understandable and fully controllable, while remaining fully intelligent.
Three fundamental problems with today’s powerful AI
-
We cannot reliably predict its behaviour
Generative AI can behave non-deterministically: the same inputs and conditions can produce different results. Its behaviour in unfamiliar situations is also difficult to predict. When these systems act independently through software or other connected systems, an unexpected decision can cause harm before a person can intervene.
-
We do not fully understand how it works
Even the developers of powerful AI do not fully understand the internal processes behind its capabilities and decisions. Without that understanding, dangerous behaviour can be difficult to identify before it occurs, and failures can be difficult to explain or correct.
-
We cannot guarantee it will remain under human control
Safety training and guardrails can reduce failures. They do not provide a guarantee that a system will always stay within its intended limits. Safeguards can be bypassed, and new situations can expose failures that earlier checks did not reveal.
Greater power makes these problems more dangerous
As AI becomes more capable and more autonomous, failures can have greater consequences. The risk is that humanity could lose control of systems powerful enough to cause irreversible harm on a global scale. That is a threat we believe must be addressed before people become dependent on such systems.
Evidence
Our response is Primary Intelligence
Our research is working towards all three requirements together: deterministic behaviour, complete understandability and complete human control. The goal is to achieve them while preserving the system’s full intelligence and ability to solve complex problems.
Work with us on this research
We are looking for allies in academia and organisations working on AI safety, policy, regulation and advocacy. We welcome universities, researchers, nonprofit organisations, grassroots movements and private companies. We also welcome funders who want to support this work.
Our purpose is to build technology that benefits all of humanity. Our immediate focus is intelligence, with a wider interest in education and health.
Sources
These sources support the statements about AI risk on this page. They do not describe or assess the Good Tech Institute’s research.