Towards Superintelligence Alignment
agendareviewed

Neglected approaches portfolio

Which neglected routes survive unified optimization pressure—and which portfolio bets compound versus diffuse effort across incompatible outer targets?

Introduction

AE Studio’s Neglected Approaches strategy (Berg, Carauleanu, Rosenblatt; Dec 2023 / Jan 2024) treats alignment as a vast, still largely unexplored space where dominant agendas may sit on local maxima. The bet is many low-probability, high-upside shots on goal: parallel pursuit of creative, plausible directions underused by the field, especially where AE’s BCI, neuroscience, and consulting-scale engineering stack is a comparative advantage. AE bootstrapped from a profitable consultancy reinvesting toward human-agency technologies (originally BCI-first); shortening timelines shifted emphasis toward direct technical alignment while keeping the portfolio logic. The published ten-direction menu is explicitly evolving first-pass guesses, not a fixed official agenda—and many items are community-wide neglected ideas AE may not be best placed to implement alone.

Who carries it: AE Studio (lead: consultancy-funded alignment team); portfolio affiliates include aintelope (Gunnar Zarncke / UAD), Steven Byrnes (brain-like safety), Heitzig human-power objective line, SOO programme (Marc Carauleanu et al.)

What they aim to do. Improve exploratory coverage of alignment by funding and executing a diversified set of plausible neglected technical and institutional directions—aiming for superintelligent AI that at minimum does not destroy humanity and ideally increases agency and flourishing.

The hard question. Which neglected routes survive unified optimization pressure—and which portfolio bets compound versus diffuse effort across incompatible outer targets?

What they produce. The portfolio publishes a neglected-approaches strategy and an AI Safety Interventions breadth map, plus object-level lines including UAD and agency-detect, prosociality reverse-engineering (SOO, theory of mind, attention schema), BCI value-mapping and RLNF, provably safe architectures, neuroscience crossed with mechanistic interpretability, field-building, safety-business incubation, and consulting-scale project execution.

Key terms. Recurring terms include neglected approaches, many shots on goal, human agency, BCI, RLNF, prosociality reverse-engineering, self-other overlap (SOO), UAD, brain-like AGI safety, human-power objective, and the AI Safety Interventions catalog.

Related field cruxes. Embedded Agency; Value Learning; Corrigibility; Goodhart Selection; Inner Alignment

What they contribute. Portfolio meta-strategy and an intervention catalog (zarncke2025interventions); Heitzig’s human-power outer target; UAD and agency-detect boundary discovery ; Byrnes’s social-cognition and homeostatic hypotheses; the SOO deception-reduction line; and AE’s ten-direction menu spanning prosociality, TAI-to-BCI-to-researchers pipelines, value mapping, RLNF, provably safe architectures, field-building, safety businesses, consultancy scale-up, neuroscience crossed with interpretability, and neglected policy.

How this project treats it. Portfolio breadth does not imply a single discharge path through the measurement spine; any one outer objective still needs its own adversarial-verifiability argument for Corrigibility and correction-channel integrity .

Map clustering

AISafety.com map listings that roll up to this agenda:

See the coverage matrix for evidence tagged to this agenda, and the glossary for shared terms.