Stephen Wolfram on the Wolfram Language, Computable Knowledge, and the Future of AI

This podcast clip features Stephen Wolfram discussing the core philosophy and practical applications of the Wolfram Language and Wolfram Alpha. He explains the Wolfram Language as a high-level, symbolic computational language designed to represent and operate on 'things that exist in the world or things that we can imagine and construct,' rather than just intrinsic computer operations. This fundamental difference allows it to encompass a vast range of real-world concepts, from boolean expressions to geographical data and image identification, through its 6,000 primitive functions. Mathematica is presented as an instance of this language, used for technical computations, while Wolfram Alpha leverages the language and its underlying knowledge base to answer natural language queries.

Wolfram highlights the distinction between the Wolfram approach and traditional AI, particularly the 'learning from scratch' paradigm. He argues that the Wolfram Language and Wolfram Alpha embody a 'knowledge-based' AI, which computationally encodes the accumulated knowledge of human civilization, rather than attempting to derive everything from raw data. This approach, he contends, is often more efficient and effective, especially for tasks requiring broad world knowledge. He also discusses the emerging synergy between this symbolic, knowledge-based approach and modern machine learning techniques, particularly in areas like image identification and Natural Language Processing (NLP), where statistical methods complement the precise computational understanding of Natural Language Understanding (NLU).

The conversation delves into the historical context of AI, noting that earlier 'expert systems' were too modest in scope, whereas Wolfram's vision required 'biting off the whole problem.' He recounts the daunting task of building the Wolfram Alpha knowledge base, inspired by his 'Principle of Computational Equivalence,' and the realization that even a vast reference library is finite. The development process involved iterative implementation across numerous domains and leveraging input from world experts to achieve 'true expert level knowledge' across general civilization knowledge, aiming to automate answers to any question based on this knowledge.

Finally, Wolfram touches upon the broader implications and challenges, including the slow adoption of profound ideas, the necessity of leadership and a coherent vision for long-term projects, and the 'infinite tails' of ambitious endeavors. He also explores the future applications of computable knowledge, such as 'computational contracts' and the critical need to encode ethics into 'AI ethics modules' for automated content selection and other societal functions. He cautions against a singular, universal AI ethics module, suggesting a more distributed, 'brand-specific' approach to avoid undesirable societal outcomes.

the point of often language is that what the language is talking about is things that exist in the world or things that we can imagine and construct not it's not it's not sort of it's it's aimed to be an abstract language from the beginning

I would say by far the highest level computer language that exists and it's really been built in a very different direction from other languages

the goal of Wolfram language is to have the language itself be able to cover this sort of very broad range of things that show up in the world

the fact that right in the language it knows about all the volcanoes in the world that knows you know computing what the nearest ones are it knows all the maps of the world and so on fundamentally different idea of what a language is

in our civilization we have learnt lots of stuff we've surveyed all the volcanoes in the world we've done you know we've figured out lots of algorithms for this or that those are things that we can encode computationally and that's what we've tried to do and we're not saying just you don't have to start everything from scratch

the interplay between these things between this kind of knowledge of the world that is in a sense very symbolic and this kind of sort of much more statistical kind of things like image identification and so on and putting those together by having this sort of symbolic representation of image identification that that's where things get really interesting

NLU defined beautifully as converting their query into computation come into a computational language which is a very well first of all super practical definition a very useful definition and then also a very clear definition

what we're trying to do is a thing that for better or worse requires leadership and it requires kind of maintaining a coherent vision over a long period of time

our goal over the next year or two is to ingest everything that's in here and that's you know it seemed very daunting but but in a sense I was well aware of the fact that it's finite

if there's a question that can be answered on the basis of general knowledge and a civilization make it be automatic to be able to answer that question

it is a necessary feature of attempting to automate more in the world that we encode more and more of ethics in a way that gets sort of quickly you know is able to be dealt with by computer

it's not possible to decide or it might be possible but it would be really bad for the future of our species if we just decided there's this one AI FX module and it's going to determine the the the practices of everything in the world