The Modern Efficiency Frontier — A Foundational Paradigm Shift in Society

“Scientia potentia est.”

This is an age-old adage that translates to “knowledge is power.” “Scientia potentia est” traces back to Frencis Beacon expressing “ipsa scientia potestas est” in his work Mediatones Sacrae in 1597, used later by Thomas Hobbes in Leviathan (1668) and even into the early history of the United States, used multiple times by Thomas Jefferson. The crux of the meaning behind this assertion is that humans, by obtaining knowledge about the natural world, humans can gain control of the world and harness its resources for our collective benefit. Underpinning this statement is the fabric of time and evolutionary steps that has propelled the human race since the early days of cavemen. Technological era over era (and I’m not necessarily referring to digital technology as the term has become synonymous with in the modern era) has been built on knowledge of the natural world, from early tools to scholarly theories that provide mathematical representations of the physical world. Most notably, and a thematic statement what will be interwoven within the diction of this piece, is that knowledge comes from collective understanding, not a single action.

The Internet and the Information Age

Eras are easily defined and contextualized through transformative moments, whether an innovation to a historical event. For technology, the last major paradigm shift was the establishment of the internet.

The advent of the internet lead to a massive globalization and democratization of information access and communication. Companies could create pages on the web for customers to access so long as they had internet access (do you remember dial-up through American Online (AOL)?) instead of simply relying on foot traffic to a store. Information could be published for a national audience to read instead of relying on the local news and what was broadcasted over national news networks. Suddenly, the world “shrunk” and the world entered the age of information.

Information available at one’s fingertips was immensely powerful. Suddenly, humans could gain access scholarly articles without leaving ones house and going to a library, obtain travel routes that minimized distance or travel time through a smartphone application, even obtain basic information about businesses without picking up the phone. Within the enterprise, databases were formed to centrally plan mission-critical operations from inventory management, customer relationship management systems, and sharing of knowledge internally on shared drives. Quickly, information shifted from a pull modality, whereby people sought out information, to a push modality, where information was shared directly with us. Smartphones enabled push notifications to give people updates instantaneously, enterprise systems moved to just-in-time modeling, and new data in the form of analytics were starting to get generated. Moreover, should that not suffice, social media came online and catalyzed information dissemination. Now, when a breaking event happened on the news, regardless of local, links to a news article was shared hundreds of thousands of times on the internet with additional commentary. Information overload suddenly came and caused many of the societal problems we see today of anxiety and stress.

Note: These statements are highly oversimplifications of massive interwoven systematic cycles of cause-and-effect that exist within modern society. The purpose of over-generalizing is to make points illustrated later. Readers should note that scholars spend decades studying small subsegments of these systems and I pay a ton of respect to them! Yes, knowledge is also a double-edge sword but I seek to focus on the positive aspects of knowledge within the confines of this article.

With over proliferation of information, the world needed modalities in which to find that information. In came Google with the PageRank algorithm, underlined by the mission to “Organize the World’s Information.” Google and other search engines helped people actively discover information published on the internet and modern search processes — graphical databases, indexes, structured data formats — all helped information retrieval. These algorithms quickly evolved into recommendation engines:

• Facebook created the news feed to see the most “relevant” (quotes intentional) information pertinent to you
• Netflix recommended shows based on prior viewing history
• Advertising companies created micro-profile segments to target you with ads based on shopping or browsing history
• etc

This algorithmic recommendation engine system helped sort through information and push that information directly to you. Power (via knowledge) was gained by large technology companies and institutions mining that information, analyzing it, and transforming it to form knowledge which they used to then both make your life easier but also raised a lot of questions.

The critical characteristic of this newfound power via knowledge was due to scale. Take news, for example. Prior to the internet, if a hurricane caused havoc to your city, you would simply apply a native human reaction to the event — trying to understand how to prepare better for the next impending hurricane, checking with family and friends to make sure they were ok, and calling news outlets to share any information you may had. At human scale, however, one might be able to learn that hurricanes were occurring frequently around the globe during a given time period. With computers, that scale is magnified to gain global trends augmented by information at the edge gathering data about water temperature, climate movements, and overlaying data onto GIS map data to test hypothesis and draw wide-reaching conclusions. Again, access to information yielded systems that provided knowledge which gave humans more power to act.

Artificial Intelligence and Efficiency Frontier

As compute power ramped, breakthroughs in computer science lead to algorithmic efficiencies, and connected devices generated petabytes of data consistently, the world was filled with information. Instead of making information accessible, could computers spot trends at a level far beyond the capabilities of humans both in terms of time and scale to create knowledge. Yes! Enter machine learning and the early promises of Artificial Intelligence. Most notably, supervised machine learning (and eventually unsupervised) argued that teaching a computer “knowledge” through processing labeled information could allow the computer to “understand” a concept via a model.

The classic example I like to use is breast cancer — show a computer an image of 100,000 examples of malicious tumors on a breast and 100,000 without and have the computer learn to make inferences when seeing a never-before-seen example. Moreover, at scale, these computers could detect breast cancer formation prior to any highly skilled doctor as it can see a tremendous more examples than the doctor could in their lifetime, thus saving more lives. The doctor now becomes equipped with an expert breast cancer detector to aid them in diagnosing and treating a patient.

Travel back in time and speak to me back in 2014 when I was touring customers through the IBM Watson Client Experience Center (CEC) in our Astor Place headquarters in NYC and you could see a human-centric approach to demos where a wall of different personas, from pharmacists to lawyers, digitally stood with similar “knowledge” stories on how AI augments their intelligence.

Over time, we learned that Artificial Intelligence was simply Probability and Statistics, but at scale, human intelligence could be captured in statistically significant manners.

Enter Generative Artificial Intelligence (“GenAI”) with ChatGPT in November 2022. What structured and unstructured data did for Supervised and Unsupervised machine learning in the “Conversational Intelligence Era” of Artificial Intelligence, language data did today in the formation of Large Language Models. Large Language models took a highly difficult and task of modeling human language, which is highly nuanced and complex with idiosyncrasies and highly dependent on context, and created a modality both to represent this information (moving away from simple language modeling of n-grams to attention mechanisms and vector embedding representations) that gave rise to a new source of data. Now, natural language could be ingested and new “knowledge” could be formed in the patterns of human language (and computer code) as machines started to create similarities in topics (again, probability and statistics) thus codifying “knowledge” (quotes intentional as AI is not “knowledgeable” in understanding per se). Organizations started mining any available source of information, from social media posts to music lyrics to corporate documents simply to try and build these large models. As a result, the probabilistic models allowed for newly generated results that were creationary in nature, less consumptive.

Note: I, again, understand the complexities and “political” climate surrounding LLMs from data ownership and privacy, to recreation of likeness and deepfakes. This blog post, again, is meant to illustrate a broader point but readers should be mindful of all competing sides, both pros and cons, to this technological advancement.

New modalities of teaching computers started to form, giving rise to questions such as: (1) Can computers understand code semantics and generate computer code, after all…high level programming is written in natural language as opposed to compiled machine language and (2) what if computers could understand “actions” such as navigating through a user interface or could translate natural language into codified actions?

The former is highly impressive but the latter is what is transformative. The newest paradigm, from Microsoft to Rabbit, is layering instruction via natural language to allow people to achieve more. In essence, Generative AI is helping to democratize efficiency.

This democratization comes from reducing actions into statements made in natural language, a near common-denominator for most of humanity. From “Natural Language Computing and the Democratization of Innovation”

Why natural language? Why distil computational power into the form of words? As mentioned previously, words are powerful and are the inherent modality of communication among human beings, verbal and non-verbal communication. Our ability to comprehend, innovate, and communicate happen naturally with words. While prompt engineering and Large Language Models certainly have their shortcomings, the concept of interfacing and “computing” with natural language introduces a new realm of possibilities (cognizant of language availability and privilege I have of speaking English), unlocked by human potential and the power of words, and breaks down the barrier to access for all.

Let me provide some truly transformative examples that I find extremely compelling:

• Coding “Copilots” — using natural language to generate documentation, write routine computer code, or even decompose a problem into steps with performing the aforementioned tasks as a result. This is one of the promises of GitHub Copilot Workspaces.
• Content Summarization — a knowledge worker receives hundreds of emails or notifications on a routine basis as normal course of action for their job. With a click of a button, that user can get a summary of the discussion and any actions required to be taken, saving time and mental effort. This is available within Microsoft Outlook today as part of Microsoft Office 365 Copilot
• Topic Transformation — a student does not understand a complex mathematical topic such as eigenvectors and can ask a system to analogize the key concepts into a subject matter familiar to them or even transform the content between modalities, from text to diagrams to audio playback to help a student learn more efficiently. Khan Academy is doing this today alongside Microsoft.
• Knowledge Search — looking for information, someone can easily search for a topic and get an answer, no more “10 blue links” and can ask follow up questions or delve in deeper on a topic, even create a “web page view” of the material — this is Googles Search Generative Experience.

All of this power stemming from underlying knowledge fueled by information that allow people to work more efficiently and focus on complex tasks.

(I will leave the subject about autonomous agents as an exercise for the reader). Moreover, in “theory” each individual gains access to these various models to then achieve more than their physical time or ability allows them to. As I alluded to in “The Sovereign Entrepreneur”

Ben’s theory is that, in an idealized world of an abundance of high-functioning, disparate verticalized (and/or generalized) Artificial Intelligence models, a single individual with an idea can drastically reduce the time to a Proof of Concept (or “Hello World”) along with a business model, marketing plan, and pitch deck without requiring a team of people to support them. I’m not disclaiming the power of collaboration and the human mind, rather, the angle taken here is that an individual can accomplish more from the start before getting others involved. Put another way, experimentation to gain early feedback, persistent with “The Lean Startup Methodology” becomes more attainable and presents quicker cycles with AI. It gets back to the notion of removing the blank canvas problem but at a much larger scale.

All of that is predicated on the “Council of Experts Leader-Agent Model”

What occurs within the framework is that a trusted ecosystem forms of trustworthy data and expert models that collectively can be accessed by engineers to build complex AI-powered solutions, with monetization by all players in between. This framework may certainly require evolution over time, however, lays out a logical manner in which to communally build expertise and achieve a layer of “AGI” (quotes intentional).

Which, in part, is that the general trend is leading to in the industry, although confirming to a dominant design pattern of chatbot plugins (neutral sentiment on).

Humans are naturally flawed beings and we as a society have a massive hurdle to overcome — to breech our individualistic nature for survival and behavioristic results of having limited resources (time, possessions, etc) and historical mental processes of inequity that we need to prevent from pouring into models and infrastructure. There does exist a world whereby individual competitive advantages intersect with ever-growing subject-matter knowledge to accelerate innovation even further, but that requires (as I stated at the beginning of the piece) collective knowledge. I truly believe that AI is transformative and this future fuels my passion for this truly transformative technology. I hope I am seeing AI’s true potential and not creating fiction, rather, detailing a reality that makes the statement “Scientia potentia est” a statement about the human experience and progress in society.