Gary Marcus explains his conception of AI:
“Deep understanding” is a term that Davis and I used throughout our 2019 book, Rebooting AI, literally the conceptual focus of the entire book:
In short, our recipe for achieving common sense, and ultimately general intelligence, is this: Start by developing systems that can represent the core frameworks of human knowledge: time, space, causality, basic knowledge of physical objects and their inter-actions, basic knowledge of humans and their interactions. Embed these in an architecture that can be freely extended to every kind of knowledge, keeping always in mind the central tenets of abstraction, compositionality, and tracking of individuals.
Develop powerful reasoning techniques that can deal with knowledge that is com-plex, uncertain, and incomplete and that can freely work both top-down and bottom-up. Connect these to perception, manipulation, and language. Use these to build rich cognitive models of the world. Then finally the keystone: construct a kind of human-inspired learning system that uses all the knowledge and cognitive abilities that the Al has; that incorporates what it learns into its prior knowledge; and that, like a child, voraciously learns from every possible source of information: interacting with the world, interacting with people, reading, watching videos, even being explicitly taught.
Put all that together, and that’s how you get to deep understanding.
It’s a tall order, but it’s what has to be done.
This specification passes far beyond any simple functional notion of intelligence (a la Turing) and insists that it fundamentally reproduce the conditions of human intelligence. The problem is that we don’t fully understand the latter. In speaking about human intelligence we have a tendency to describe essential constitutive features that are less fundamental than emergent. Hinton adopts a less human-focused and idealist perspective, arguing that intelligence emerges as an emergent property of complex neural systems, not as some elusive property that coherently anticipates or precedes them. In this manner, he adopts both a more materialist and structural perspective. It is not as though AI is intelligent because it duplicates human intelligence but because it demonstrates features of intelligence that have a firm basis in neural architecture and processes, as well as a more general and platform agnostic character. Hinton’s neural network’s approach signals that there is not simply one privileged form of intelligence but multiple ones. Hinton describes distinguishing features of digital intelligence: it is immortal (not dependent on any single, unique biological identity); it can also replicate itself much easily and swiftly (radically altering the possibilities, pace and overall scope of learning). However, the question remains, what makes it ‘intelligent’ if it is founded upon linear algebra and stochastic symbol processing? Here, it is best not to get too hung up on the details of implementation. Neural networks certainly do employ numerical means different to our own, but this indicates less a lack of intelligence than a different intelligent framework. We focus only on the alien discrete mechanisms, which are actually as opaque to us as the electro-chemical processes in our own brains, and tend to ignore the structural-systemic characteristics. The latter are not simply any shallow mimicking of human intelligence but reveal emergent homological features. Most importantly, AI has the capacity to evolve an overall model of the universe of symbolic meaning. The hugely multi-dimensional character of this model enables sophisticated processing and generation of information, as well as the coordination of independent action. Of course, this is largely restricted to the sphere of symbolic intelligence, but we can expect the perceptual and materially engaged capacities of AI systems to increase and become more nuanced in future. Ironically, despite his avowed humanism, Marcus employs a strongly binary logic: intelligence is conceived as either human or not intelligent at all. Hinton, on the other hand, encourages the thinking of multiple forms of intelligence – some of which may be (worryingly) superior and better adapted to survival than our own.