# §31 Supercapabilities do not entail “superpowers”

*Reframing Superintelligence: Comprehensive AI Services as General Intelligence — K. Eric Drexler, FHI Technical Report #2019-1 (Markdown edition). Section 31 of 40. Source label: `abilities-not-powers`. Rendered PDF: https://ora.ox.ac.uk/objects/uuid:9c05427a-6390-4b42-9c55-ee45f73a26ad*

**Thesis.** By definition, any given AI system can have “cognitive superpowers” only if others do not, hence (strategic) superpowers should be clearly distinguished from (technological) supercapabilities.

**Cites sections:** §10 *R&D automation dissociates recursive improvement from AI agency*

**Cited by sections:** §II, §32

---
## Summary

*Superintelligence* (Bostrom 2014) develops the concept of “cognitive superpowers” that potentially include intelligence amplification, economic production, technology development, strategic planning, software hacking, or social manipulation. These “superpowers”, however, are defined in terms of potential strategic advantage, such that “at most one agent can possess a particular superpower at any given time”. Accordingly, in discussing AI strategy, we must take care not to confuse situation-dependent super*powers* with technology-dependent super*capabilities*.

## AI-enabled capabilities could provide decisive strategic advantages

[Application of AI capabilities to AI R&D](→§10) could potentially enable swift intelligence amplification and open a large capability gap between first- and second-place contenders in an AI development race. Strong, asymmetric capabilities in strategically critical tasks (economic production, technology development, strategic planning, software hacking, or social manipulation) could then provide decisive strategic advantages in shaping world outcomes (Bostrom 2014, 91–104).

## Super*powers* must not be confused with super*capabilities*

Bostrom (2014, 93) introduces the concept of a “superpower” as a property of “a system that sufficiently excels” in one of the strategically critical tasks, stating that “[a] full-blown superintelligence would greatly excel at all of these tasks,” and later explains that “excels” must be understood in a relative sense that entails a strong situational asymmetry (Bostrom 2014, 104):

> […] superpowers […] are possessed by an agent as superpowers only if the agent’s capabilities in these areas substantially exceed the combined capabilities of the rest of the global civilization [hence] at most one agent can possess a particular superpower at any given time.

To avoid confusion, it is important to distinguish between strategically relevant capabilities far beyond those of contemporaneous, potentially superintelligent competitors (“superpowers”), and capabilities that are (merely) enormous by present standards (“supercapabilities”). Supercapabilities are robust consequences of superintelligence, while superpowers—as defined—are consequences of supercapabilities in conjunction with a situation that may or may not arise: strategic dominance enabled by strongly asymmetric capabilities. In discussing AI strategy, we must take care not to confuse prospective technological capabilities with outcomes that are path-dependent and potentially subject to choice.

## Further Reading

- §20 *Collusion among superintelligent oracles can readily be avoided*

- §32 *Unaligned superintelligent agents need not threaten world stability*