What problem are they working on?
- Surveys of leading AI researchers suggest a significant probability of human-level machine intelligence in this century.
- A human-level machine intelligence might not be aligned to our values, which could cause it to perform actions negatively affecting humanity.
- We do not currently know how to define the goal function of such powerful optimization processes in a way that is consistent with human values.
- Survey of expert opinion suggests about 1 in 3 chance that development of powerful artificial intelligence will be bad or extremely bad for humanity.
Why is this problem important?
Charity interventions are typically designed in terms of years or decades. But all the individuals existing in the current decade are vastly outnumbered by the individuals that will exist in the decades, centuries, and millennia to come. If we consider only the next thousand years, the ratio “present decade: future decades” is already 1:100. And yet, in relation to cosmic timescales, a thousand years is almost nothing, meaning that future individuals might outnumber presently existing ones by an even larger factor.
Risks of dystopian futures caused by artificial intelligence are both pressing and under-explored. MIRI has a team of researchers and its work has been endorsed by prominent AI researchers. Although there are some uncertainties connected with MIRI’s work, we believe that MIRI is an outstanding organization that can have an immensely positive impact in the near and far future.
What do they do?
The Machine Intelligence Research Institute (MIRI) focuses on overcoming technical obstacles to making smarter-than-human artificial agents safe and robust. They do this by developing formal tools for the clean design and analysis of general-purpose AI systems. Additional activities include research workshops and conferences.
Does it work?
MIRI develops formal tools to solve AI control problems in-principle, ensuring that future programmers have the foundation to address these problems in practice. This approach has been successfully used by researchers in the past, and helped create the foundations for modern computer science. The Open Philanthropy Project recently granted MIRI $500,000 to continue, expand and improve it technical research.
We recommend MIRI because:
- MIRI’s work has the ability to prevent vast amounts of future suffering.
- The AI-control problem is underexplored and underfunded considering its importance.
- MIRI is using an effective and tested approach to solving AI-control problems.