Stephen Wolfram on the Computational Universe, Wolfram Alpha, and the Path to Artificial General Intelligence

Stephen Wolfram, creator of Wolfram Alpha and Mathematica, discusses his perspective on achieving Artificial General Intelligence (AGI) by leveraging systematic computational knowledge rather than solely mimicking human brains. He recounts an early interaction with AI pioneer Marvin Minsky, highlighting the initial skepticism and eventual recognition of Wolfram Alpha's practical utility in answering complex, real-world questions. Wolfram emphasizes that the success of Wolfram Alpha stems from a deep integration of natural language understanding, vast data sources, and the ability to compute things from that data, often by 'cheating' and applying centuries of exact scientific knowledge and algorithms rather than pure reasoning.

The core of Wolfram's approach lies in the Wolfram Language, a symbolic language that allows for the coherent representation and manipulation of both abstract mathematical concepts and real-world entities. He illustrates this with examples ranging from analyzing random graphs and processing images to computing geographical distances and modeling 3D anatomical structures. A key philosophical shift for Wolfram was realizing that "mere computation" could achieve what he once thought required building a brain-like system. This led to a bottom-up development strategy for Wolfram Alpha, building capabilities across thousands of specific domains and then identifying common frameworks, rather than starting from a grand, global ontology.

Wolfram also delves into modern machine learning, demonstrating how Wolfram Language can symbolically represent, inspect, and train neural networks using datasets like MNIST. He shows how to analyze intermediate layers of a neural network to understand its feature extraction process. This integration of machine learning within a broader computational framework underscores the system's versatility and power. The discussion extends to the fundamental question of what is computable, leading to his work on cellular automata and the concept of the computational universe.

His seminal discovery of Rule 30, a simple cellular automaton generating immense, seemingly random complexity from trivial rules, serves as a powerful metaphor. It suggests that the natural world and even advanced AI might emerge from simple underlying programs, and that searching the "computational universe" for effective programs can be more fruitful than trying to design them from first principles. This perspective offers a profound implication for AGI: rather than reverse-engineering the brain, AGI might be achieved by systematically exploring and leveraging the vast space of possible computations and integrating them into a coherent, knowledge-based programming system.

"I kind of seen for years you know question answering systems that tried to do sort of general intelligence question answering and so at Marvin and so I was going to show Marvin you know Wolfram Alpha he looks at it and he's like okay that's fine whatever said no Marvin this time it actually works you can try real questions this is actually something useful this is not just a toy."

"I think the most important thing is just knowing a lot of stuff about the world is is really important to actually being able to to understand natural language in a useful situation."

"One of the things that we were doing in in Wolfman alpha was to kind of cheat relative to what had been done in previous AI systems which was instead of using kind of reasoning type methods we're just saying okay we want to compute where the ISS is going to be well we've got a bunch of equations of motion that corresponds to differential equations we're just going to solve the equations of motion and get an answer."

"My conclusion was if you want to do something like that the only realistic path to being able to do it was to build something much like a brain and so I got interested in neural nets and I tried to do things with neural nets back in 1980 and nothing very interesting happened."

"Many years later for reasons I can explain I kind of came back to this and realized actually it wasn't true that you had to build a brain like things sort of mere computation was sufficient."

"One of my principles in building Wolfram Alpha was not to start from a grand theory of the world that is not to kind of start from some global ontology of the world and then try and build down into all these different domains but instead to work up from having you know hundreds then thousands of domains that actually work."

"The thing that sort of made building wolf now for possible was this language wolf and language which started with Mathematica which came out in 1988 and has been sort of progressively growing since then."

"What we learn from this is out in the computational universe of possible programs it's possible to get even with very simple programs very rich complicated behavior."