And now, the real journey begins!
There is a drive in the naturally born creatures of this world. Instinctively, creatures are driven to survive with sensory perceptions that include combinations of visual, auditory, motor, olfactory and other senses at birth which are not learned. Evolution may have played a part in forming the gene over time, but the creature conceived and formed with these capabilities at birth.
At birth, the creature is introduced into an environment that is ready to receive the newcomer. Temperate is the atmosphere supporting conditions within a narrow band of tolerable comfort. A perfect mixture of oxygen, nitrogen, hydrogen and other elements within a narrow band of necessary density and combinations to provide the creature’s embodiment with gas replenishment. Over and over, these basic automatic responses and many others occur without any thought. Viable is the creature and the welcoming environment for which the creature is introduced.
Whilst these automatic responses occur, naturally occurring automatic desires beyond the creature’s control drive actions to seek food, seek warmth, seek safety and achieve satisfaction. Given, these desires occur within a preformed and welcoming environment for which the creature was introduced. By intent, what the creature seeks the environment provides.
Desires now drive the creature. Assessing the surroundings, the creature attempts to relieve the effort required to seek and adaptive control begins. With control gradually freeing time is thought evolving or was thought simply another instinct that the mind had more time to utilize? Thoughts began and continued efforts increased for more adaptive control. At first forming tools, forming locations, farms and other adaptions to relieve efforts of those desires once seeked and gain more control. Now, cities, countries, financial institutions, businesses and what is next?
More and more time is available.
With this time begets ever more complex thoughts of love, war, greater control, and other desires to be complacent or to conquer become the makings of novels and further intents of humankind. Is this what we mean by intelligence? Merriam Webster defines intelligence as “the ability to learn or understand or to deal with new or trying situations”.
The preamble above is a basic journey of infancy from hunter-gatherer to current society. How does this relate to “Machine Intelligence, not Artificial Intelligence and What Really is Machine Learning?” Intelligence by definition is “the ability to learn or understand or to deal with new or trying situations” and this definition of intelligence is apparent when studying and considering history with the adaptive control that occurs over time. During these journeys to control, mistakes are made and time combined with motivation are the contributing factors to adjust and adapt. When is fitness fully achieved? Jumping forward in the journey to adapt and control, we arrive at the computer and the early assumption of what is artificial intelligence and what is understood by machine learning.
Artificial intelligence in the strong form was and continues to be perceived as a creature in an artificial form meaning not born of this earth having self-awareness while akin to a human in some combination of resemblance, actions or desires. Artificial intelligence in the weak form is very useful and combined with sensory perception to an environment that is not self-aware but is process controlled with boundaries such as self-driving cars, robotic medical assistants or other fascinating inventions. What is important is that one is not the other and will not learn what the other does e.g. the self-driving car cannot, nor desire to be a medical assistant and vice versa. The need to adapt and control is not to expand beyond the engineered boundaries of intent but to operate efficiently and effectively within those boundaries increasing fitness through feedback enrichment. The beauty of weak intelligence is that these devices do not have to think about what they want to be when they grow up. When researching robot capabilities, would you want them to be self-aware or simply reactive and operate without error for their engineered intent? The robot’s weak intelligence is not forming desires, not molding environments to better suit their existence and not evaluating other needs outside of their intended realm so is there any intelligence, “deal with new or trying situations”, beyond the boundaries of that which is programmed to be adaptive.
What is instinctive or adaptively learned outside of these boundaries?
There is also weak intelligence in purely diagnostic applications. These are the majority of applications advertised as Machine Learning where data is analyzed and there is sentiment or other textual response with little or no action occurring. You have a response so what do you do with it now? In the area of well-defined domains such as medical imaging that change very slowly, simple data sets such as text messages, frequency searches deriving intent and a few others, the output is quite helpful. However, in the areas that are subjective to opinion, business rules, compliance, regulatory or extreme complexity, there may be mixed results or results that are only accurate until opinion, business rules, compliance, regulatory or ever more extreme complexity breaks that implementation. By default, machine learning is not learning as there is limited ability to “deal with new or trying situations”. In most cases, think about how one generally arrives at the machine learning solution. First, human reviews the information available in the domain the solution is required. Second, pick a tool as they are not all the same. Third, establish a corpus of information to establish baselines. Fourth, train the system with combinations of established procedures for the selected tool and repeat by feeding with reference corpus and continue with supervised and unsupervised learning enrichment until the machine responds with useful and validated answers. In simple terms, answers are formed mostly by comparing clusters using Bayesian and/or Gaussian Mixture probability which is basically a statistical guess. Here is one of many academic references. In some industries and business areas, this works well form the start, in some business areas it works until things change and, in some business areas, it depends on how much time and money you have. For those areas that change slowly, searching for frequency or evaluating simple data sets, machine learning in its current form is very useful. However, solutions are not trying to adapt to other areas and context is established by a human from the “outside-in” and is accurate provided the human and the information is representable and the underlying problem does not significantly change. What is not happening is that the system is not adaptive to new conditions but instead is programmed with data for an intended problem area. Moreover, where is the economy of scale if there is a different machine learning solution at cost for each different area or maybe even within a given area for each solution.
Would this approach always be in the red? Perhaps a more appropriate name for what we currently call machine learning is Information Refinement, Categorization and Search but that is not as interesting is it? What is more interesting to consider is what if you put the machine learning algorithms atop a unified adaptive control platform, then we may be approaching machine learning’s intent. The problem with the term machine learning is mostly an issue when considering starting from the “outside-in” for a specific intent but what if one started from the “inside-out” and achieved something more universal?
A machine is not intelligent but if the machine had an adaptive program that could become thousands of solutions at once, could that machine eventually become intelligent? Humans with their born instincts, are these aligned with the definition of intelligence? These instincts allow the human to adapt and seems to meet the definition “deal with new or trying situations”. If software were instinctively adaptive and able to “deal with new or trying situations”, would that not be another type of intelligence? With that said, could we then establish the term Machine Intelligence where there is no intent for the machine to become self-aware but there is every attempt for the machine to be viable in the cybernetic environment and adapt to change whereby the machine is capable of “dealing with new or trying situations” in that environment? Where a human is the creature born into the ecosystem of this earth, the adaptive machine is formed by humankind into the cybernetic ecosystem. There is not an intent to form an artificial intelligence but there is every intent to form a machine intelligence that further assists human’s adaptation by utilizing the machine’s adaptation. Human instructs machine; machine supervises itself according to human’s instructions. The machine responds to many humans and many different instructions all at once providing automated solutions for other humans to use. The machine’s supermodel has the ability and capability to adapt to human’s ever-changing needs relevant to the cybernetic environment that machine resides. Initial context forms. Other humans then utilize the machine’s solutions for legal, insurance, financial, government and other needs where data is captured against a unified model and automatic operational context forms. A machine’s adaptation to needs is similar to the adaptation of needs of the early human with the environment and useful intent being different. It is easier to go with the grain and can be very tough or even choking when going against it.
More to come with Universal Machine Learning, Supermodel and Genetic Mapping and One that Could Rule them All!
About the Author:
Richard Yawn is a technology leader, architect, developer, inventor and visionary with experience providing solutions across industries and functions in insurance, manufacturing, government and legal.