The American Philosophical Association (APA) has introduced the creation of latest prizes for philosophical work on synthetic intelligence.
Particularly, the prizes are for philosophical or philosophically-informed interdisciplinary work on the AI2050 Hard Problems. AI2050 is a undertaking from Schmidt Sciences, a science and technology-oriented philanthrophy, targeted on the right way to notice the potential of AI.
They put it this manner:
It’s 2050. AI has turned out to be vastly helpful to society. What occurred? What are a very powerful issues we solved and the alternatives and potentialities we realized to make sure this final result?
The working checklist of AI2050 Onerous Issues are based mostly round 4 objectives:
-
Develop extra succesful and extra common AI, that’s helpful, secure and earns public belief
-
Leverage AI to handle humanity’s biggest challenges and ship constructive advantages for all
-
Develop, deploy, use and compete for AI responsibly
-
Co-evolve societal techniques and what it means to be human within the age of AI
Every of those objectives is damaged down into units of issues. For instance, the primary, relating to the event of extra succesful and extra common AI, is elaborated on as follows:
1. Solved the science and technological limitations and arduous issues in present AI which might be important to enabling additional breakthrough progress in AI resulting in extra highly effective and helpful AI able to realizing the helpful and thrilling potentialities, together with synthetic common intelligence (AGI).
Examples embrace generalizability, causal reasoning, larger/meta-level cognition, multi-agent techniques, agent cognition, the flexibility to generate new data, novel scientific conjectures/theories, novel helpful capabilities, and novel compute architectures, breakthroughs in AI’s use of assets.
2. Solved AI’s frequently evolving security and safety, robustness, efficiency, output challenges and different shortcomings that will trigger hurt or erode public belief of AI techniques, particularly in safety-critical purposes and makes use of the place societal stakes and potential for societal hurt are excessive.
Examples embrace bias and equity, toxicity of outputs, factuality/accuracy, info hazards together with misinformation, reliability, safety, privateness and knowledge integrity, misapplication, intelligibility, and explainability, social and psychological harms.
3. Solved challenges of security and management, human alignment and compatibility with more and more highly effective and succesful AI and finally AGI.
Examples embrace dangers related to tool-use/connections to bodily techniques, multi-agent techniques, aim misspecification/drift/corruption, dangers of self-improving/self-rewriting techniques, achieve of perform dangers and catastrophic dangers, alignment, provably helpful techniques, human-machine cooperation, challenges of normativity and plasticity.
You may see the complete checklist of working issues, about which the undertaking “makes no declare to being complete, closing, or fastened,” here.
There are two prizes, one for an early-career researcher and one for a longtime researcher. Every prize winner will obtain $10,000. If there are co-winners, they’ll cut up the prize quantity equally. The prizes, funded by Schmidt Sciences, might be awarded on the APA divisional conferences.
The APA now invitations submissions for the inaugural APA AI2050 Prizes. The submission deadline is June 23, 2024.
For extra info on these prizes, together with particulars on standards and eligibility, go to the APA AI2050 Prizes page.
(through Erin Shepherd)
