Alpha- & Beta- AI; An Introduction
By James Carlson
See White Paper on Alpha-AI & Beta-AI
Artificial Intelligence (AI) isn’t a fantasy, it’s a present reality. One reality of AI is that there are no machines thinking, no neural networks, and no ‘sky-net’; these are all fantasies. What is real are Machine Learning (ML), Nodal Networks (Deep Learning - DL), and many networks that allow for Generative AI to do data research. Given all the hype, we can utilize AI in new and different ways that likely have never been done before. Hence, I introduce to you a new field of AI research that I call Alpha-AI and Beta-AI.
It is vain to attempt to change the predominance of false ideas. It is best to present new and improved ideas that people will use to replace them. Instead of flailing about to complain about those flailing about with nonsense, let’s replace the false ideas of what AI is; that makes real sense. Combining old and new tools for science and discovery, we can utilize AI in the real world that advances our understanding of the real world.
So, let’s start with an introduction to the Science of Systems Thinking. This is where we part from the fantasy of thinking machines, neural networks, and other networks that don’t exist. Let’s do some thinking for ourselves before we use the tools of AI. The 5 characteristics of any given ‘system’ are:
· Parameters
· Relationships
· Initial Conditions
· Inputs and Outputs (Open System)
This is where we can begin to use tools that will improve upon the functions of AI in the analysis of data to discover the characteristics of any given system.
Next, let’s introduce the idea of Design of Experiments (DOE). This is an old tool of science that has been used for the statistical analysis of data generated by various tests on any given system. To simplify the matter, DOE begins with 2 or 3 parameters and provides a truncated set of test runs to arrive at a polynomial (equation) representing the interdependent relationships between the parameters tested. This then provides equations and relationships for a system under test that improve upon our understanding of how they work and what they do.
Now, let’s introduce what AI actually is. First, we have Machine Learning (not thinking). This is a very important starting point for AI as it associates inputs with outputs to arrive at a process of analysis. Given a series of inputs, a computer program (algorithm) is trained to respond to various outputs. With a trained algorithm, testing on new inputs provides a trained output and the training is complemented with positive/negative reinforcement. This is not machine thinking, it is machine learning. And it is buggy at best.
Next, we have Deep Learning (DL), which is a subset of Machine Learning (ML). DL provides more information than ML can. One example is in the analysis of Nodal Networks (not neural) that provide various parameters and the relationships between them. Each node may be considered a parameter or perhaps a scalar on each parameter. If A maps to B as a functional equation (A = F(x,y,z) = B), then the result of the AI analysis provides an explicit equation that shows the relationships between the various parameters (2x2+y/3 – z-3 = B). This is the analysis that AI brings to us with the AI tool of DL. DOE can take this statistical analysis further.
DL, when properly used, provides the parametric relationships that defines any given system under test. DL utilizes statistical analysis and is often considered to be a statistics engine where that analysis occurs. And as it provides for the identification of parameters and their relationships, it is uniquely suited for the work of DOE and Systems Thinking. This then is the unique combination of old and new tools that I call Alpha-AI.
Generative AI (GenAI) is a subset of DL and provides tools the user needs to access networked data stores. So, with a Data Store and a Data Analysis, there is ultimately a Data Utility or application that is the result of this AI process (AI is all about the DATA-DATA-DATA). This is a process whereby AI delivers useful information for our growth and advancement. GenAI does this along with DL to provide access to the networked data that is available on the open Internet or with Data Centers via local Intranets. Data provides the foundation; AI provides the analysis. Now let’s turn to the use of Alpha-AI along with Historical Data, which I call Beta-AI.
Beta-AI is only one application of Alpha-Ai whereby we capitalize on the data already in hand for further analysis and utility. The key to Beta-AI is the use of historical data that often has assumptions associated with it. As our knowledge grows from traditional science, old ideas of how systems work are often discarded. Similarly, Beta-AI makes use of old data that came from previous experiments and challenges the assumptions of previous experiments while using the data that came from them for system discovery and equation formulation.
The theory behind a test often determines what the test parameters are, and the data that is a result of the test may have a limited set of parameters to study. As such, all data is theory-laden, a famous quote that someone else invented. I use it here to suggest that all data sets have meaning for the greater reality we all share apart from the original assumptions of any given test. By challenging the assumptions and getting a truncated set of data from various sources, we may be able to compile a more complete data set that will enable us to make discoveries of system characteristics and the equations that govern these systems.
Again, as data sets provide immediate assistance to the advancement of any particular scientific endeavor, they have utility apart from the original assumptions of the test that generated them. This then is the beginning of Beta-AI where the use of historical data provides insights into systems thinking that original assumptions may not have considered. Challenging the experiment is not the purpose of Beta-AI but challenging the data to goes beyond the experiment is.
Given the thousands of data stores across the globe, we have billions of data points that are available for the use of Alpha-AI (hence, Beta-AI). But these data stores are sequestered, user-account-accessible, Intellectual Property (IP), and so on. If/when the data is available, Beta-AI can champion that data and provide for real discovery beyond past assumptions. Beta-AI has the ability to make leaps in our understanding of various systems, what characterizes them, and how they work in a more deterministic analysis.
This is where the rubber meets the road. We have systems all over the place. When we know exactly how they work (Parameters, Relationships, Initial Conditions, Inputs/Outputs), then we say it is a deterministic system. But with human learning, we have systems that are not fully understood, and we cope with that lack by calling it a stochastic system. The term stochastic refers to a probabilistic analysis of how a system works and what the likely output of that system might be. As probability and statistics overlap, so too does AI and Stochastic Systems.
Examples of stochastic systems are:
· Weather Prediction - Meteorology
· Economics
· Medical Arts
· Politics/Warfare
Given mankind's long history with intrinsically stochastic systems, we have learned to use short pithy sayings to deal with them. These constitute much of our culture as morals and traditions. Examples are:
· Red at night, sailors delight…
· Save your money for a rainy day
· Pay down your debt; live within your means
· Starve a cold and feed a fever
· History vs Abstraction (Burkean conservatism)
· Let slip the dogs of war and cry havoc
While morals are born of our God-given conscience and represent universal truths, traditions vary from nation to nation and culture to culture. We celebrate our differences in traditions and hope to learn from them all. But the ability to put together a deterministic understanding of how these stochastic systems work begs us to laugh at those who claim such. Recall masks during Covid when it was told that they either worked 100% or 0%. The extremes are rarely true; and while they do work, they only work a little. We laugh at those who know absolutely how a given system works when everyone else only knows it in part. Now, we have the opportunity to move forward towards a deterministic understanding of these systems with Alpha- & Beta- AI; and that's no laughing matter.
Comments