The idea of hybrid neuro-symbolic architectures (see: @GaryMarcus ) can be framed in terms of Andy Clark's extended mind hypothesis. In this framing, it is the intuition machine (system 1) that does all the work to understand the work of a symbolic system (system 2).
@GaryMarcus It's analogous to a human doing work by using a computer to help do his work. All the thinking remains in System 1 and System 2 is just an external tool (but an extension of the mind).
@GaryMarcus A general intelligence must be grown inside out. But like all biological creatures, it grows by adapting to its environment. To build hybrid systems, it is system 2 that is the environment and it is system 1 that is the general intelligence.
@GaryMarcus There are several mechanisms for system 1 and system 2 to cooperate. I've described that here: medium.com/intuitionmachi…
@GaryMarcus A characteristic of general intelligence is its adaptability. To create a hybrid system, a general intelligence adapts to understanding the processes of a logical system. It is no different in how a human understands a causal system.
@GaryMarcus Hybrid systems are organic systems adapting to an inorganic system. Pictured in the real world: 
@GaryMarcus An intuitive system understands logic in the same manner that it understands a causal system. So, to understand how a DL system might reason, one has to first understand how a DL system understands causality.
@GaryMarcus @threadreaderapp unroll
