When we change the efficiency of knowledge operations, we change the shape of society.
AI-augmented knowledge summarization, refactoring, & integration are about to transform the world. Again.
Technology creators, funders, and policymakers must understand how changes in knowledge operations can impact society.
This is intended to be a short primer making a series of claims about these impacts. Each claim could be a book (many are).
1. Knowledge operations matter.
Knowledge operations are ‘actions that involve the conveying or processing of knowledge’. For example, publishing and broadcasting are ‘distribution’ knowledge operations where one ‘conveys’ knowledge to a large audience with minimal feedback. Summarization is a ‘processing’ knowledge operation.
Many core ‘societal activities’—from conflict mediation to identity formation—both (1) require knowledge operations and (2) operate in a world awash with ongoing knowledge operations that influence the outcomes of those activities. Consequently, knowledge operations are fundamental to how we think, how we act, and who we are — individually and collectively. They impact emotion, culture, politics, and power.
2. The past few decades have seen a •dramatic• decrease in the cost of knowledge operations, both by individuals and collectives.
In particular, increases in ‘distribution’ efficiency due to connectivity (e.g. web, messaging, content networks) and selection (e.g. search, recommendations) have had a tremendous impact on the last decade. Arguably, the impact of these efficiency gains has been greater than that of most other societal forces combined, potentially because most forces change more slowly.
3. Changes to the “efficiency of knowledge operations” have led to societal phase changes — dramatic shifts in structures and power dynamics of nations, organizations, and movements.
At one time, organizational, educational, and physical infrastructure was needed to ensure that modern nations could use horses for transportation, communication, and war. The shape of society changed when we transitioned to a more efficient technology for ‘mobility operations’. We saw similar societal phase changes due to changes in the efficiency of knowledge operations resulting from innovations like the printing press, broadcast media, and now social media.
These efficiency changes impact everything from how we make war, to how we love, to how our brains work, to how we find meaning. This is not new.
4. It is challenging for most people (or organizations) to fully appreciate the societal impact of changing the “efficiency of knowledge operations.”
This is why so many were caught flat-footed by the societal impact of Facebook and Twitter, which were considered just “useless” technology by many — until they changed the world. It’s also why the telegraph was expected by many to be a harbinger of world peace.
5. The societal impacts of many new technologies depend significantly on the societal context they are introduced to — and even the methods and timing of their introduction.
One visceral example of this is the development of the atomic bomb and decryption techniques during World War II. The timing and context in which those technologies were introduced had significant impacts on the outcome of the war.
Similarly, new communications technologies in some societies have likely increased polarization, while in others they may have reduced it — potentially as a result of their political structure or culture.
There are many socio-technical and political economy considerations that determine the impact of more efficient knowledge operations in a given context, e.g. (1) the types of costs being reduced, (2) the actors paying (or not paying) the costs of those operations previously and (3) the externalities of these operations and the degree to which they can they be managed.
6. We have some — albeit limited — ability to influence the context, timing, and method of the introduction of new technologies.
We need to make sure that we use this influence wisely — and time is of the essence. New technologies create bifurcation points where societies can go in very different directions. One notable example was the introduction to the car in the United States vs. much of Europe. In the US, policymakers and car companies worked together to destroy mass transit infrastructure; in sharp contrast to what happened in many other countries that invested in mass transit (e.g. different societal contexts).
As we have seen with the rise of radio, TV, and now social media, if we want more efficient knowledge operations to lead to good societal outcomes, we must act to make it so. It is unlikely to happen automatically.
7. AI advances are now enabling communications to be automatically summarized, refactored, & integrated — and will reshape our collective cognition.
Recent AI advances portend a new phase in the increase in the “efficiency of knowledge operations” — one of many we will face simultaneously. Just as with ‘horse infrastructure’, we have built elaborate training regimens and societal structures to manage the tremendous cost and complexity of these knowledge operations. Many of those may soon be rendered irrelevant.
We have also failed to resolve many crucial education, conflict mediation, and governance problems as a result of that computational complexity. New ‘knowledge technologies’ may allow us to to succeed.
- Knowledge operations: actions that involve the conveying or processing of knowledge.
- Conveying: the creation, distribution, and reception of knowledge between entities (inter-entity operations).
- Processing: transforming knowledge in some way to create a new knowledge output (intra-entity operations).
This was inspired by work across many fields, from McLuhanesque media theory, to organizational psychology, to information theory (and the corresponding ‘physics of knowledge operations’). I list a few specific examples below, demonstrating how some of these claims can be framed in the terms of that discipline. It is also worth noting that efficiency is just one of several factors crucial to understand about knowledge operations.
A. Traditional Economic Frame: The functional marginal cost of many operations is being dramatically reduced—often to ~zero.
This is partly because the types of resources needed for those operations have changed from atoms to bits, or from human time to compute time, such that the current limiting resource is no longer necessary.
B. Computer Science Frame: We are significantly decreasing the *computational complexity of knowledge operations.*
In asymptotic computational complexity terms, selection and connectivity have made knowledge operations that were logarithmic with human time (or much worse) into operations that are functionally constant time: O(≥log(n)) → O(1). Machine learning advances will bring these kinds of speedups to many new domains.
Decreasing computational complexity is an even more profound shift than making something 2x more efficient or even 1000x more efficient — it’s changing how the ‘amount of stuff in the system’ impacts efficiency.
C. Behavioral Economics Frame: More efficient knowledge operations impact our ‘bounded rationality’.
Both individuals and groups of people have bounded rationality. Arguably, many (though clearly not all) of the societal problems we face are an indirect result of that bounded rationality making effective decision-making and conflict resolution challenging. Thus, more efficient knowledge operations may significantly enhance our ability to resolve these conflicts. Whether or not this is true depends the relative advantage that increased efficiency provides to those seeking to resolve a conflict vs. those seeking to prevent resolution (among other factors).
Written by Aviv Ovadya at The Thoughtful Technology Project.
You can find Aviv on Twitter @metaviv, this mailing list, or via email.
This is excerpted from a more formal piece in-progress; future excerpts will go into detail around the emergent threats and opportunities relating to specific knowledge operations.
It’s important to note that while this excerpt focused primarily on knowledge, powerful language models can now be trained to “use tools” (e.g. via Toolformer), which allows them to impact far more than just knowledge operations. This will likely have even more revolutionary impacts on society.