AI vs. EI

Which is best to answer our Questions?

Artificial Intelligence or Expert's Intelligence?

Generative AI has sparked significant discussion and speculation about its future potential. However, it's crucial to recalibrate our understanding and expectations.

In this article, you'll learn what generative AI is and what it isn't, and the gaps it needs to fill before it can evolve into a Distributed Expert Intelligence (DEI). This DEI would ideally be accessible to anyone with a smartphone, enabling us to expand our human capabilities through readily available expert knowledge.

What is Intelligence?

It is “the ability to perceive or infer information, and to retain it as knowledge to be applied to adaptive behaviors within an environment or context.” (Wikipedia)

What is Artificial Intelligence?

“Artificial intelligence (AI), in its broadest sense, is intelligence exhibited by machines, particularly computer systems.” (Wikipedia)

In 2024, AI commonly refers to Large Language Models (LLMs), such as Generative Pre-trained Transformers (GPTs). These models are sometimes tailored for specific domains, like BloombergGPT for finance. The leading players in the AI marketplace include OpenAI, Microsoft, Google, and AWS, but there are hundreds of other LLMs, and the list is rapidly expanding. LLMs are "probabilistic models of a natural language" (Wikipedia).

LLMs proficiency lies in the statistical analysis of tokens, enabling them to predict the next word in a sequence. In truth, since AI chatbots always fabricate content without knowing its meaning, the content produced is only an expression resulting from a statistical analysis. By its very nature, it is neither accurate nor inaccurate since it lacks any meaning. Humans, in turn, attach meaning to this fabricated content, and may personally consider such expression accurate or inaccurate.

Inferior input leads to inferior output. The content used to generate these new expressions is primarily generic, unstructured data scraped from the internet. It lacks access to the structured data maintained in corporate databases, which is a far richer source of knowledge than data publicly available online. Additionally, it misses out on the experts’ tacit knowledge residing in human brains, which is highly valuable as it often drives innovation and provides a competitive edge within organizations.

Furthermore, large language models (LLMs) cannot access a user's personal history or recent diary data. This limitation means that LLMs cannot provide truly personalized recommendations within the context of an individual's specific experiences.

LLMs also reflect the biases present in their extensive training data. Recent iterations of ChatGPT are based on 175 billion parameters and require thousands of GPU clusters to train. Sam Altman has noted that training GPT-4 costs over $100 million.

For more, see: “The best large language models (LLMs) in 2024”, by Harry Guiness, May 2, 2024

https://zapier.com/blog/best-llm/

What is EI, an Expert’s Intelligence?

“An expert is somebody who has a broad and deep understanding and competence in terms of knowledge, skill, and experience through practice and education in a particular field or area of study. Informally, an expert is someone widely recognized as a reliable source of technique or skill whose faculty for judging or deciding rightly, justly, or wisely is accorded authority and status by peers or the public in a specific well-distinguished domain.” (Wikipedia)

What are EI’s Neural Network (EI’s NN) Dialogue Models?

An expert's knowledge is encoded into Neural Network Dialogue Models by assigning weights to the relationships between domain-specific factors or parameters. In the health and wellness domain, these parameters might include age, gender, symptoms, diseases, products/drugs, ingredients, side effects, treatment characteristics, and numerous other criteria affecting the decision-making process. The weights reflect the beliefs and relative importance attributed by the expert to each relationship between these nodes.

These weights can represent various factors, such as the frequency of an event, or the intensity of an effect. They may also pertain to distances, periods, levels of abstraction, complexity, harmony, popularity, and other measurable or estimable dimensions, sometimes assessed using subjective fuzzy logic (e.g., faster, stronger, relatively young). Some weights are grounded in statistics, curricula, or the expert's specific experiences, or life practices. However, others may reflect the expert's style, originality, character, personality, or intuition.

To rank alternative decisions or choices and determine their attractiveness to the decision-maker, an AI tool needs to analyze these alternatives in terms of their evaluative criteria and the weights assigned to their relationships. This table of parameters, or evaluative criteria, and the associated weights, built as an EI's neural network dialogue model (EI’s NN), represents a virtual synopsis of the expert's intelligence.

How are Dialogue Models Applied to the User’s Context to Generate Personalized EI-Insights or Advise an Alternative Choice?

To leverage the expert's intelligence, and apply it as actionable knowledge in the user's decision-making context, a Q&A dialogue is engaged between the users and the EI-NN models residing in their Personal Virtual Assistant (PVA) bot. This interaction is enhanced by the history of the user’s previous dialogues (health diary).

The answers to these questions will reflect the user's state regarding each evaluative criterion, and the cumulative weights each state adds to each alternative decision will continuously rank them. As more criteria are addressed in the discovery dialogue, the ranking of each alternative decision will evolve.

Since decision-making "is regarded as a continuous process integrated in the interaction with the environment" (Wikipedia), the EI’s NN dialogue model will suggest the next question by prioritizing the resolution of the unknown state that progresses the most toward a decision threshold, ultimately determining the user's choice based on all evaluative criteria.

This approach to building a Personal Virtual Assistant (PVA) bot for decision-making assistance, leveraging EI-generated insights, eliminates the need for the costly collection of big data, machine learning, or deep learning typically used to create generic GPT content that lacks awareness of the user's states.

It also avoids the necessity of Retrieval-Augmented Generation (RAG) to guard against inaccuracies, often called "AI hallucinations."

Expert’s Intelligence Neural Network Dialogue Models (EI NN) Characteristics

EI’s NN models are Very Tiny Dialogue Models (VTDM). For instance, a 400-parameter model designed to discuss abdominal symptoms and help determine whether a condition is a hernia, constipation, cholecystitis, or appendicitis requires only 4KB for the neural network and the dataset. In comparison, ChatGPT was trained on 570GB of datasets. Most EI’s neural network dialogue models will have a few dozen to a few thousand parameters. They will typically be less than 20KB in size, including the dataset and the model.

Creating an EI’s NN dialogue model costs a fraction of what it takes to develop a large language model, and its maintenance and continuous training costs are even a smaller fraction.

In another domain’s decision-making, such as shopping categories, the parameters might include product features, benefits, destinations, delivery methods, application methods, and more.

What is an EI Platform?

An Expert Intelligence EI platform, such as NNOD (Neural Network On-Demand)(1), is a web platform that facilitates the parametrization of expert knowledge and the development and distribution of neural network dialogue models between participants across multiple EI-Domain ecosystems built on this EI platform.

Participants can use text, voice, Siri, Alexa, or other voice assistants and since all dialogues are parametrized, an expert dialogue model (EI NN) built using English can easily be translated into a foreign language, and respond to a broadcast query formulated in any foreign language supported by the voice assistant (Siri, Alexa).

On-Demand Model Access

These dialogue models will be available for download from Libraries of Models (LOM), maintained by any participating user or provider.

User Side

• Local Operation: The EI neural network models are stored on the user's mobile device or Personal Virtual Assistant (PVA), allowing for offline engagement in discovery dialogues between the user and their PVA bot. This provides EI-generated insights to help users solve measurable problems without an internet connection.

• Privacy and Accessibility: This setup is ideal for remote field workers, ensuring complete privacy by eliminating the need to share personal data with a centralized authority. The PVA bot can also maintain a personal health diary on the device, which can be used in dialogues with the user, or bot-to-bot interactions between the PVA bot and provider bots when an internet connection is available.

Provider Side

• Integrated Models: For providers (vendors, doctors, nurses, coaches, therapists, manufacturers, universities, etc.), the domain-specific neural network models are integrated within a chatbot. These models support discovery dialogues about the provider's products, services, benefits, and applications.
• Broadcasting Intent: Both users and providers can broadcast an intent, or query profile, on relevant channels. The providers’ chatbots can subscribe to these channels, review the intents, and engage in a bot-to-bot dialogue with the users' PVAs. This will facilitate identity verification and Q&A navigation to address the next unknown parameter in the dialogue, reducing user frustration and hang-ups.
• Reversed Advertising: In this mode of interaction instead of being served paid advertisements (as in the Google model), users broadcast their query profile, and providers respond to these requests for information (RFI) or requests for proposals (RFP). This bot-to-bot dialogue can then be elevated to human interaction if needed.

This approach fosters a more direct and private interaction model, where expert knowledge is readily accessible and personalized for users and providers.

What is an EI Domain?

An Expert Intelligence (EI) Domain is an ecosystem comprising participants who share neural network dialogue models focused on a common subject area. These participants engage in discovery dialogues facilitated by their Personal Virtual Assistant (PVA) or the providers’ chatbots.

What will be a typical dialogue with EI-Generated Insight?

A user can record self-observation notes and engage in Q&A sessions with her Virtual Assistant (PVA) on various subjects, such as menstrual periods, digestive issues, pain, diet, anxiety, fear, mood swings, headaches, confusion, blood pressure, depression, self-esteem, motivation, behavioral conflicts, social interactions, coworker cooperation, curiosity, ideal car, travel plans, calendar events, reminders, daily goals, successes, and more. She can initiate these Q&A dialogues with any parameter, in any order, and they can be conducted without an internet connection. The states she records for these parameters are time-stamped, updated, and saved under major categories such as Personal Health Profile, Shopping Intents, Personal Notes, and To-Do Lists. All data is kept private by being stored solely on her device.

For subject matters in which her PVA has previously downloaded Expert Intelligence neural network models (EI NN), it will provide EI-generated insights. If the user declares an intent, the EI on the device will prompt the next most important question to resolve any unknown state and reach a decision threshold that triggers an EI recommendation.

What is a Library of Models (LOM)?

For subject matters where the PVA has not yet downloaded EI NN dialogue models, it will initiate a download from one of the available Libraries of Models (LOMs) offered by various parties. For instance, a library of models in a generic subject area might be provided by a cloud provider, a university, a government organization, a foundation, a professional association, or a private non-profit organization like the Wikimedia Foundation. These high-level generic LOMs are expected to cover subjects of Global Truth (GT-LOM) and interest.

For subject matters of regional or special interest, the Regional/Special Interest Library of Models (R/SI-LOM) will include a greater number of parameters and additional EI NN dialogue models to support Q&As in these focused areas.

In addition, private brands and service providers will offer their own EI NN dialogue models (PB-LOM) to support Q&As about their specific products or services.

Health, Wellness, & Beauty Treatments (HWB) LOMs

Most information flows and decisions in Health, Wellness, and Beauty Treatments (HWB) can be viewed within three phases:

1. DISCOVERY: self-observation, diagnosis (self or assisted), fact gathering and interpretation
2. DECISION: goals, search for alternative resolutions and decisions, selection of a resolution path
3. DEPLOYMENT: of a resolution plan, user assignments, coaching partner, and follow-through

With social media’s word-of-mouth, the EI Platform that will develop a Library Of Models (LOM) and engage significant ecosystems with many users in one of these EI Domains will gain a first-mover advantage and be well-positioned to serve users in adjacent domains’ Q&A.

Distributed Expert Intelligence (DEI)

These Libraries of Models (LOMs) will form a layer of Distributed Expert Intelligence (DEI), where clusters of LOMs can be associated in various ways. For example, a Global Truth Library of Models (GT-LOM) on cancer published by the Mayo Clinic could be licensed and adapted by Lurie Children’s Hospital. Their experts may adjust parameter weights and add EI NN models to support deeper discovery dialogues about pediatric cancer, offer second opinions, and better serve their specialized interest in children's health.

How can we compare, or contrast, GPT’s AI and EI’s NN Dialogue Models?

The future

Generative AI will continue to enhance our ability to quickly and comprehensively grasp vast amounts of data, generating insights that help us investigate alternative decisions and think critically.

While future Artificial General Intelligence (AGI) is often envisioned as a centralized authority—a superior, know-it-all entity that is always right because it has analyzed all books and internet material—it cannot provide truly personal and contextual recommendations without access to private databases, expert intelligence, and personal data.

In contrast, the future of Expert Intelligence (EI) is Distributed (DEI). This distributed EI will reside on our mobile devices and our smart homes, from $35 Raspberry Pi computers to powerful cloud systems. It will provide EI insights anytime, and when connected to the internet, it will broadcast intent profiles in bot-to-bot discovery dialogues with provider chatbots. It will download EI neural network models to be applied to the personal information we keep private, advancing our progress in decision-making and implementation.

Distributed EI will be multimodal and ubiquitous, seamlessly integrated into our lives, much like spell checkers in our writing. It will access numerous EI neural network models functioning with minimal data to reveal explanations, augment human intellect, accelerate prosperity, and benefit societies.

Generative AGI and Distributed EI may coexist in digital services, each serving different yet complementary tasks and purposes.

Georges Selvais,

co-founder NaturHeals

(1) Neural Network On-Demand (NNOD) is an EI Platform developed by Sergey Tolkachev who received US Patent # 9305050B2 for his invention of the virtual neuron, the backbone of NNOD, which I believe will be a cornerstone of the Distributed EI economy. NNOD is fully developed as a prototype and Sergey’s company 256GL is inviting potential funding partners with aligned vision to contact us if they are interested in the NNOD market opportunity. We asked ChatGPT what is truly unique about Mr. Tolkavhev's patent and the response is in this post: https://www.dhirubhai.net/feed/update/urn:li:activity:7208100288191152128?utm_source=share&utm_medium=member_desktop

(2) Personal vs. Personalized AI, Doc Searls Weblog, https://doc.searls.com/2024/05/10/personal-vs-personalized/

Thank you Note: I am profoundly grateful to my friend, Sergey Tolkachev, for decades of honest and profound discussions about artificial intelligence. His brilliance, unparalleled insights, and unwavering dedication have been a source of immense inspiration. I also extend my heartfelt thanks to NaturHeals for providing the opportunity to innovate in ways that aim to alleviate global health inequity.