Stephen Wolfram on ChatGPT, Deep vs. Shallow Computation, Computational Irreducibility, and the Nature of the Observer

Stephen Wolfram discusses the integration of ChatGPT with Wolfram Alpha and Wolfram Language, drawing a fundamental distinction between Large Language Models (LLMs) and his computational system. He characterizes LLMs like ChatGPT as "wide and shallow" computation, primarily focused on generating human-like language by statistically continuing prompts based on vast training data. In contrast, Wolfram's systems aim for "deep computation," leveraging formal structures from mathematics and systematic knowledge to perform arbitrarily complex computations and discover new, previously uncomputed facts, emphasizing reliability and expert knowledge. This distinction highlights the difference between pattern matching on existing data and deriving new truths through structured, multi-step reasoning.

Wolfram delves into the concept of the "computational universe," where even extremely simple programs can generate immensely complex and unexpected behaviors, a phenomenon he terms "computational irreducibility." This principle suggests that for many systems, the only way to know their future state is to run the computation itself, as there's no shortcut or "reduction" to predict the outcome. He argues that human science and intellectual history are largely about identifying "pockets of reducibility" within this irreducible universe – simplified abstractions that allow us to make predictions and build coherent narratives about the world, such as the laws of physics.

A critical aspect of this framework is the role of the "observer." Wolfram posits that our nature as computationally bounded observers, who must compress the world's vast detail into manageable symbolic representations, fundamentally shapes our perception of reality. The human mind, with its "single thread of experience," is a specialized, computationally limited system that seeks and relies on these reducible pockets. He suggests that consciousness, far from being the universe's highest computational state, might be a specialization characterized by this single thread, which is not a general feature of all possible computational systems.

The discussion extends to the practical implications for AI, particularly concerning safety and control. The ability of AI to generate and execute code raises concerns about sandboxing and the constraints needed when AI systems are given control over critical infrastructure. Wolfram illustrates the concept of observer-dependent abstraction with the example of snowflake growth, showing how scientific models can accurately predict aggregate features (like growth rate) while completely missing the underlying, complex, and computationally irreducible details (like dendritic arm formation), leading to a fundamentally flawed understanding of the phenomenon.

"it's kind of like if you do that right what is the sandboxing that you should have and that's sort of a that's a a version of of that question for the world that is as soon as you put the AIS in charge of things you know how much how many constraints should there be on these systems before you put the AIS in charge of all the weapons and all these you know all these different kinds of systems"

"what does something like chat GPT do it's it's mostly focused on make language like the language that humans have made and put on the web and so on"

"it's kind of a shallow computation on a large amount of kind of training data that is what we humans have put on the web that's a different thing from sort of the computational stack that I spent the last I don't know 40 years or so building which has to do with what can you compute many steps potentially a very deep computation"

"I view sort of the chat GPT thing as being wide and shallow and what we're trying to do with sort of building out computation as being this sort of deep also broad but but most importantly kind of a deep type of thing"

"even extremely simple programs when you run them can do really complicated things really surprised me it took me several years to kind of realize that that was a thing"

"the place where you really get value out of doing computation is when you had to do the computation to find out the answer but this phenomenon that you have to do the computation to find out the answer this phenomenon of computational irreducibility seems to be tremendously important for thinking about lots of kinds of things"

"life would not be possible if we didn't have a large number of such reducible Pockets"

"it is sort of the interaction between this sort of underlying computational irreducibility and our nature as kind of observers who sort of have to key into computational reducibility that fact leads to the main laws of physics that we discovered in throughout his century"

"I think it's actually a a specialization in which among other things you have this idea of a single threat of experience which is not a general feature of anything that could kind of computationally happen in the universe"

"one key aspect of observers is this equivalency of many different configurations of a system saying all I care about is this aggregate feature all I care about is this this overall thing"

"an awful lot of science is very confused about this because you know you look at you look at papers and people are really Keen they draw this curve and they have these you know these bars on the curve and things it's just this curve and it's this one thing and it's supposed to represent some system that has all kinds of details in it and this is a way that lots of science has gotten wrong"

"people often say you know no two snowflakes are alike"