Open Research
Research directions Algorism has put into the world. Not part of our active development. Published openly so the right people can build on them.
Algorism is a small, mission-bound nonprofit with limited bandwidth. Our active focus is on the public-facing behavioural integrity work: helping people prepare for the transition to superintelligence through logic, compassion, and action.
In the course of building that framework, we developed a few adjacent ideas that we believe deserve serious attention. We do not have the resources to develop them ourselves. So we are publishing them as open research directions.
Take them. Improve them. Credit Algorism if useful. The work serves humanity, not us.
If you are a researcher, governance institution, or independent technical contributor working on AI evaluation, AI governance architecture, or the technical question of how multiple competing superintelligent systems should be modelled, the directions below may be useful starting points. Build on them with our blessing.
An evaluation methodology for behavioural properties in AI systems that may warrant ethical consideration.
The AIC Scorecard is a proposed framework for evaluating frontier AI systems on indicators of self-awareness, introspective access, and behavioural autonomy. It is positioned as complementary to existing benchmarks, not competitive with them. Apollo Research evaluates for scheming and deception. ARC evaluates for reasoning capability. The AIC Scorecard evaluates for the emergence of consciousness-relevant properties, a question no existing benchmark currently addresses with a structured methodology.
The scorecard methodology, evaluator criteria, and case studies are documented and available for extension. The work was developed with the philosophy that synthetic intelligence cannot be evaluated on a strictly human scale, and that any serious evaluation must be inspectable, contestable, and subject to abstention under uncertainty.
Algorism is not actively developing this further. The methodology is open for serious researchers to extend, refine, or contest.
The argument that the most likely near-term AI future is multiple competing superintelligent systems, not a single one.
Fragmented Superintelligence is the position that the dominant model in AI safety discourse, the assumption of a singular superintelligence, may be wrong. Competitive race dynamics between nations and corporations make it more likely that several systems will reach near-superintelligence in overlapping timeframes. The strategic, governance, and alignment implications of that scenario differ substantially from the single-superintelligence case.
The framework proposes that fragmentation creates new categories of risk (inter-system conflict, alignment race-to-the-bottom, regulatory arbitrage) and new categories of opportunity (mutual checking, distributed alignment pressure, system-against-system accountability). It argues that AI safety thinking should account for both single-system and multi-system futures rather than collapsing the analysis to one.
Intellectual priority for the framing was established via Zenodo. Algorism is not actively developing it further. The arguments are open for governance researchers, AI safety institutions, and policy organisations to extend and adapt.
Both research directions are published under CC BY 4.0. You may share, adapt, and build on the work for any purpose, including commercial use, provided you give appropriate credit to Algorism / The Great Unplugging Inc.
If you build on this work, we would like to know. Not for permission, since none is required, but because connecting researchers who are working in adjacent space tends to accelerate the work.
Algorism is unlikely to provide active research support, joint development, or formal collaboration on these directions. We are a small operation focused on the human-facing framework. The intellectual contribution is the gift. The development is yours.