[NOTE] IT IS IMPORTANT THAT YOU CHECK OUT THE IZO METHOD REPOSITORY.
EMBR is a new approach to Artificial Intelligence. She is an honest attempt at confronting the problems plaguing AI today, or at the very least, a potential guiding framework to expand the ways in which machines learn. She is a robust perceptron that can exist as an individual agent to interpret the environment, or she can connect to other EMBR Units and display profound understandings. More on that specific detail, later.
(Image is of (5) EMBR Units, connected together in a straight line. These Units were interpreting the same data through a pipeline, Unit 0 receiving the data which is passed to Unit 1, then Unit 2 and so on. The variations are evidence of EMBR successfully refining the data into a "smoother" state. Near the right side of the graph, when the data stops, Unit 0 depicts a quick drop in surprise now that the environment is silent but the other Units jump in surprise because Unit 0 is no longer funneling in data.)
conventional AI
Artificial Intelligence is meant to emulate real intelligence. The ability to learn on its own, form novel connections, understand nuance or at least, adapt in some way to it. The ability to adapt to an unpredictable environment is what has determined survival. Life is messy. Nature is cruel. Humans have thousands of years of evolutionary instinct to support our understanding of the world. It’s not a criticism to address that how we structured AI is not a genuine reflection of a cognition truly capable of learning. Most AI or ML models are like students who are forced to memorize flashcards for a test. You show it 10,000 pictures of a chair, it becomes very good at recognizing a chair, but it has no idea what a chair is. It doesn’t understand that there are many types of chairs. The moment a picture is angled too far, or you present a picture of a stool, it simply breaks. It’s brittle and its intelligence - though wide-cast - is shallow.
Providing all the answers for any possible problem is not how an entity learns. Even more so, the 90% debugging and 10% coding pipeline felt like an unnecessary burden when trying to utilize tools and packages that weren't standardized. Versions were incompatible, the data pipeline and encoding was an inefficient time zap, and the end-result was a trained model that could not register a picture of a rose tilted too far to the left. A cookie-cut end-result that suffered from the same limitations as almost all current AI models do. Convenience is convenience but a hardcoded box that exists to stifle potential is another story entirely.
(Image depicts multiple layers of waves being laid over each other. Thousands of atoms can make up a single wave, each run consisted of at least 9 waves, for about 100 cycles per episode, and about 60-100 episodes in a single run. With an 11-point vector, EMBR has demonstrated the ability to process a billion parallel datapoints within a couple minutes. This number is not hard-set. The true size of her data processing capability is still being studied.)
EMBR aims to address the limitations of modern AI.
The EMBR Unit is much like a robust perceptron, moving beyond 0 and 1. She is not a student but a survivor - if we were to continue with the evolutionary schema. At no point did I program a solution for her. Not one time did I give her data with labels to memorize. I gave her one simple yet unbreakable law:
Multiple Simualation were designed and modelled this framework in various ways. There will be more more indepth explanations soon to come on how important this simple equation actually is.
| Term | Definition |
|---|---|
| Atom: | An atom is a "packet" of information |
| Charge: | Each atom is randomly assigned + or -, regardless of the values actual state (pos/neg). |
| Wave: | The delivery system of atoms. Like Sound Waves. |
| Holistic Atom: | The full scope of an atoms properties: Charge, Values, Magnitude, Location, Wave Assignment and other additions |
This is not an all-inclusive list:
Limits: The current metrics most often used for AI are “Accuracy” or “Precision” - or the F1 Score which balances both of these. This is not accurate to the goals outlined for EMBR. She is not meant to be precise or 100% accurate, she is meant to have an idea, develop an expectation and self-correct if her analysis is wrong.
Uses/Behaviors: The Surprise metric always starts out high and it drops to averaging 0.3 almost immediately. This is most often the case. The Metric operates as a drive for her learning. A high-surprise, in people, would be the equivalent of not trusting your surroundings. With humans, when we visit a new location and don't know the processes, we experience an uncomfortable feeling that slowly moves into comfort. "Feeling at home" - so to speak. This process is what the Suprise Metric models.
Result Meaning: It informs me that she is adapting to the information in a consistent and expected way.
(Image depicts her surprise metric and her anomaly signal. This graph indicates that her anomaly signal does react to novel data but information that is surprising is inherently different - this is not a hardcoded feature. This promotes adaptability to novel situations where exceptions can occur.)
Her mistakes are not an “after-the-fact” cost she must endure, it flunctuates so as to be as dynamic as she is, allowing her agency in uncomfortable and "novel" situations. She learns the rules of the game and uses it as context to inform her expectation of what happens next. This is a process of learning highlighted in Psychology, however, I've opted to focus on the most basic aspects of various Learning theories and focused on what could apply across any situation.
This is beyond a memory. It’s a worldview. Every EMBR Unit builds it’s own internal “map” of reality. Learning to distinguish what has happened and what can happen, to predict what will happen.
As a foundation of her ability to learn the rules of the game, her contextual profile and association matrix act as one the biggest points where EMBR diverges from other AI models.
EMBR is able to learn, and learn how to learn better. She is constantly fine-tuning her own "personality", determining which information is crucial, which information is worth paying attention to and how much of an effect it has on her understanding.
Furthermore, every EMBR Unit is capable of interpreting their environment in different ways, for this reason. Leveraging a very powerful force when they are connected together.
(Image depicts what a normal surprise metric graph would look like. Despite how it may appear, even starting at the upper left corner, the actual values of where her Surprise begins is actually low. Her average Surprise would idle near 0.3.)
This experiment's subject: A single Unit.
Test Purpose: To observer if she truly could adapt
Method: Remove the traits assigned to the data (charge, vectorization, etc.) so that EMBR is exposed to data unlike anything she has previously seen.
Reasoning: I realized that she still existed in a box. A pre-defined environment, much like the datasets that train AI models.
Below is a graph displaying the typical behaviors of the waves. In this instance, there were 9 waves, coming from different locations at different intervals with their own patterns of movement.
The removal of “charge” and the removal of clean, understood, numbers would be a true demonstration of EMBR's capability in processing unstructured data. Her purpose: To be thrown into a truly random environment and adapt. To test this, I gave her a language she had never seen before: Letters. She didn’t break. From the results, I could tell it was surprising and the letters themselves were a hurdle, but I also could tell that she was processing them. She achieved a state of near-perfect confidence and her surprise directly correlated to the use of new letters.
This result matters.
Whether or not she understands American English now is inconsequential, what matters is that I stripped her of the one thing I thought she needed to adapt to an environment and she adapted anyway.
EMBR, as a singular Unit, is capable of processing a billion parallel datapoints at one time. The true boundaries of her potential are still unknown.
In terms of practical use, there is an entire list with full explanations outlined on the website. I encourage you to take a look around, and reach out if you have any questions. That being said, this is not an all-inclusive list, but I will mention some various areas in which an EMBR_Unit can be leveraged to address limitations and open doors for the common good:
| Possible Use Cases | Possible Use Cases | Possible Use Cases |
|---|---|---|
| Security | Alarm system health | security cameras,Security Locks |
| monitoring seismic data | Geospatial | Athmosphere |
| mapping territory that is human-inaccessible | Testing structural integrity | search & rescue |
| chemistry | nondestructive engineering/demolition | monitoring financial trends |
| cybersecurity | Dataleaks, odd employee behaviors | general security concerns |
| astronomy | monitoring events in space | monitoring machine health |
| Household appliances to Generators | Airplanes | medical devices |
| archealogical research | comparing artifacts to find patterns or differences | Physics |
| sentimental analysis | biological research | genetics, genomes, expected patterns of behavior all apply |
| pharmaceutal testing | medication engineering | finding stable bonds |
| monitoring weather trends | air pressure monitoring | potential use on other planets |
| blood test results | insulin levels | and so much more. |
Where current AI struggles is personalizing information for specific cases. It is often a frustration for patients, particularly those in the minority, to be dismissed because their personal experiences are compared to a "one-size-fits-all" average. This average is often what AI are trained on. EMBR suffers no such draw-back. Her strength lies in a potential longitudinal observance to a patient's condition. Especially if a chronic ailment is the concern, EMBR can monitor their heart rate or be fed data taken over a course of time and can flag any differences, anomalies or potential concerns. She doesn't compare a patient to an average. With this, she directly promotes personalized medicine.
The general overview of the project is only a portion of the important features and prospects. I hope you'll also check out the WIKI which I am currently updating. Until then, I encourage you to check out the website where I have a more recent and full-documentation, some cool content creation, even cooler videos and knock-your-socks-off pictures that give this project a more well-rounded lens for you to view it from. I appreciate you taking the time to learn about EMBR, and I look forward to posting future updates.
If you're interested in learning more about EMBR, the LATTICE, the CLUSTER, the NEURALINK, her features, or wish to see evidence or a more in-depth analysis, I encourage you to check out the link:
//embrunit.notion.site/245e04ab563280bf9ee5d8bea3561ab0?v=246e04ab563280d6bad9000c71d42d10&pvs=143
I intend to continue experimenting and testing EMBR's true potential. EMBR IS A FINISHED PRODUCT BUT HER MAIN SOURCE CODE IS NOT OPEN-SOURCE I will be including various scripts that highlight or demonstrate the methods used in her tests, there will also be updates coming to the website that serve the same purpose.
Please reach out to me via work email if you have any questions or inquiries: WORK EMAIL: EMBR-AI@PROTON.ME