Stephen Wolfram: Computational Universe | MIT 6.S099: Artificial General Intelligence (AGI)

Wolfram Language, developed from Mathematica, is a symbolic language where data (like graphs, images, or even cities) is represented symbolically and can be directly manipulated and computed. It integrates diverse knowledge domains and algorithms, allowing for high-level, precise computational thinking.

Wolfram Alpha uses a 'bottom-up' approach, starting with hundreds or thousands of specific domains and building up frameworks from common themes, rather than a global ontology. Each domain requires careful, often human-expert-driven, canonical representation of data.

The computational universe is the space of all possible programs. Wolfram demonstrated that even very simple programs, like cellular automata with minimal rules, can generate incredibly rich and complex behavior. This suggests that sophisticated computation can be 'mined' from this universe without explicit, step-by-step human engineering.

Computational irreducibility is a phenomenon where the only way to determine the outcome of a computation is to run it. This implies that there are limits to how much we can foresee or understand complex systems without actually executing them, a concept fundamental to understanding both natural systems and advanced AI.

AI systems like neural networks develop their own internal 'words' and criteria for making distinctions, which are optimal for their task but don't align with human-invented language or historical methods of explanation. This makes automatically generated proofs or image identification processes hard for humans to follow because the underlying logic uses concepts for which humans lack intuitive understanding or words.

Wolfram considers artificial intelligence our 'first example of an alien intelligence.' He argues that understanding AI's 'thinking,' purpose, and goals confronts us with similar questions to those we might ask about extraterrestrial intelligence, as their internal logic and distinctions diverge from human thought processes.

A symbolic discourse language, proposed by Wolfram, would serve as a precise, formal language derived from Wolfram Language to express human goals and ethics (like an 'AI Constitution') to AI systems. This is an attempt to bridge human values with the computational world, essential for aligning AI with human purposes.

Wolfram believes molecules can and do compute in the sense that their interactions can generate sophisticated outcomes from simple rules, similar to cellular automata. This view suggests that nanotechnology could shift from merely miniaturizing traditional computers to assembling molecular components for novel computational processes, like 'algorithmic drugs.'

Instead of focusing on low-level coding, Wolfram suggests teaching computational thinking using high-level languages like Wolfram Language, which automate much of the mechanical aspects of computation. This allows children to focus on formulating computational questions and exploring what they want to compute, fostering intuition about how computational systems operate.

Wolfram suggests that human purposes are historically contingent and constantly evolve. While a future where trillions of simulated 'souls' play video games might seem 'depressing' from today's perspective, he argues that at any given point in history, people find meaning in what they do. The interaction with AI will likely lead to new, meaningful purposes, even if they seem alien to past generations.