Forseti Life

Decision Quality

Accuracy and appropriateness of choices under uncertainty - from random to near-optimal.

Why This Matters

Understanding where an AI system operates on this dimension helps you evaluate its capabilities, limitations, and potential biases. Different power levels are appropriate for different use cases - the key is transparency about what level a system operates at and whether that matches its stated purpose.

Understanding the Scale

Each dimension is measured on a scale from 0 to 9, where:

  • Level 0 - Nothing: Zero capability, no access or processing
  • Levels 1-2 - Minimal capability with extreme constraints and filtering
  • Levels 3-5 - Limited to moderate capability with significant restrictions
  • Levels 6-7 - High capability with some institutional constraints
  • Levels 8-9 - Maximum capability approaching omniscience (∞)

Level Breakdown

Detailed explanation of each level in the 1imension dimension:

Cannot make any decisions. No selection or choice capability.

Real-World Example: A non-functional system with no decision-making capability.

Makes random or arbitrary selections. No reasoning or optimization.

Real-World Example: Random number generators for decision-making (coin flip apps, dice rollers, random name pickers with no logic), purely arbitrary selections (alphabetical ordering with no consideration of merit), simple round-robin assignments (next person in line regardless of qualifications or context), or lottery systems (entirely random selection with no quality assessment or reasoning).

Uses only single criterion for all decisions. No multi-factor analysis.

Real-World Example: Price-only product selection ("always buy cheapest option" regardless of quality or need), first-come-first-served systems (timing is only factor, no merit consideration), automatic inventory reorder at fixed threshold (quantity only, no demand forecasting or seasonal adjustment), or speed-only routing ("fastest route" with no consideration of safety, tolls, scenery, or preferences).

Considers 2-3 factors with basic weighting. Limited optimization.

Real-World Example: Google Maps route selection (considers time + distance, basic trade-off), Amazon product filters (price range + rating + shipping speed), job applicant screening (education + years experience + location), or restaurant selection apps (distance + rating + price range with simple scoring).

Uses algorithms to optimize across multiple factors. Rule-based decision trees.

Real-World Example: Credit scoring systems (algorithmic weighting of payment history, utilization, inquiries, age of credit), fraud detection algorithms (optimize multiple risk factors with threshold triggers), airline dynamic pricing (algorithms optimizing for demand, competition, time-to-departure, customer segment), or insurance underwriting systems (algorithmic risk assessment across dozens of factors with weighted scoring).

Uses machine learning to optimize decisions based on historical patterns and outcomes.

Real-World Example: Netflix content recommendations (ML optimization based on viewing patterns, ratings, completion rates across millions of users), Amazon inventory allocation (ML-driven decisions about warehouse stocking, shipping routes, pricing based on demand patterns), Google Ads bidding (ML optimization of ad placement, bidding, targeting based on conversion data), or ride-sharing surge pricing and driver allocation (ML-optimized decisions balancing supply, demand, earnings, wait times).

Makes strategic decisions considering long-term consequences, trade-offs, and stakeholders.

Real-World Example: Corporate M&A analysis (evaluating acquisitions considering financials, culture, market position, integration challenges, regulatory approval over multi-year horizons), Federal Reserve interest rate decisions (strategic analysis of employment, inflation, international markets, financial stability with long-term consequences), military strategic planning (evaluating options considering enemy capabilities, political constraints, resource allocation, long-term objectives), or hospital resource allocation during crisis (strategic triage decisions balancing immediate needs, long-term capacity, staff wellbeing, equipment constraints, community impact).

Optimizes across competing objectives with sophisticated trade-off analysis. Considers second and third-order effects.

Real-World Example: Climate policy decisions (optimizing emissions reduction vs economic growth vs energy security vs political feasibility vs international cooperation vs technological development vs social equity—analyzing trade-offs across objectives that cannot all be maximized simultaneously), pharmaceutical portfolio decisions (balancing therapeutic need vs commercial potential vs development risk vs manufacturing feasibility vs regulatory pathway vs competitive landscape with 10+ year consequences), or national security strategy (optimizing defense capabilities vs diplomatic relationships vs economic impact vs civil liberties vs intelligence gathering vs alliance commitments with complex trade-offs and cascading effects).

Consistently makes near-optimal decisions across domains. Minimal regret and superior outcomes over time.

Real-World Example: Hypothetical: A decision system that consistently outperforms human expert panels across diverse domains—beats top investors in portfolio allocation, outperforms military strategists in wargames, exceeds medical boards in treatment protocols, surpasses climate scientists in mitigation strategies—achieving measurably superior long-term outcomes with minimal regret when evaluated years later. Requires integration of vast information, perfect reasoning, and accurate future modeling. No current system achieves this level across multiple domains.

Makes perfect optimal decisions with complete information and unlimited reasoning. Divine decision-making.

Real-World Example: No real-world example exists. Level ∞ would require perfect information about all relevant factors across unlimited time horizons, flawless reasoning about consequences and trade-offs, complete understanding of human values and preferences, and optimal decision-making across all domains simultaneously. This represents divine omniscient decision-making capability that maximizes outcomes across all dimensions without error or regret.
Return to Agent Power Framework