What are Artificial Intelligence and Machine Learning?

From Webster’s Dictionary:

“Definition of intelligence

1a(1) : the ability to learn or understand or to deal with new or trying situations : reason 

also : the skilled use of reason

…”

This is the definition that most engineers and scientists use and understand, i.e. the ability to learn about and understand complex systems, processes, and problems sufficiently to design and improve and optimize processes, solutions, methods, tools, or products.  In simpler terms, it is the cognitive ability to make scientific progress.

With respect to science and engineering, the strict meaning of the words “artificial intelligence” is non-biological intelligence, or an intelligent computing device or program.  Intelligent computers or programs do not currently exist.  All of the intelligence behind any existing computer application comes from the mind of its developer.  First he must derive or invent a new or improved solution to some problem.  Then he must translate that mathematical or physical solution into a computer program that contains step-by-step instructions (code) for the computer to obtain the solution automatically (upon execution).  Coding the solution to a problem in a computer program (the computer science part) can be as difficult or more difficult than determining the solution to the problem (through science and engineering).  Neither the program nor the computer it runs on are intelligent.  The computer processes the developer’s code to implement his solution algorithm automatically, exactly as instructed by the developer to achieve its objective in obtaining the program outputs (=solution) from its inputs (=problem description), one line of code at a time.  The computer simply follows the instructions in the program code that represents the knowledge of the developer of how to solve the problem using the simplest and most efficient logic and methods that he knows or derives.  We construct programs that solve complex problems because the program can solve the problem or perform some function much faster and more accurately than people can.  Neither the computer nor the program “knows” how to solve the problem, or even what the problem is or the meanings of the inputs and outputs.  Only competent developers and users have that knowledge.  Our computers and programs are simply tools that we use to significantly advance human knowledge and capabilities.

Many sources cite 3 levels or types of artificial intelligence: narrow, general, and super.  General artificial intelligence is the above strict definition that most of us have understood and used for decades (an intelligent computing device or program).

“Narrow” artificial intelligence is entirely created by people and is no different than computer programming.  A good definition of narrow or weak AI is "a computing device or program that appears to be intelligent, but is not".  Some define it as a computer application that mimics intelligent human behavior. Examples commonly cited include pattern and voice recognition, interactive voice assistants and search engines. It is all very advanced interactive programming by very intelligent developers, but it is not AI, because no computer or computer program is intelligent. By most definitions of "narrow" or "weak" AI, examples also include the abacus, the slide rule, the electronic calculator and just about every computer program ever written that saves time over manual computation or processing.

Claims of achievement of “narrow” artificial intelligence are meaningless.  Most sources admit that General AI is a subject of research and does not currently exist.  Some refer to AI as the Computer Science field of simulating artificial intelligence to make applications appear to be intelligent (rather than actually being intelligent, which is impossible today and for the foreseeable future).

“Super Artificial Intelligence” is defined as surpassing human intelligence in all domains. As we always believed that general AI would do. Those inventing these terms often refer to all computer programming as AI that began in the 1970's!

Since we are apparently free today to make up term definitions (and entire sciences!) to suit our purpose, we define “Artificial Specific Super Intelligence” as the ability of a machine or program to exceed the abilities of humans to solve or perform any specific problem or task (rather than in general on all subjects).  This, again, has existed since the hammer, the lever, the abacus, the slide rule, the electronic calculator, and since the first computer programs were written.

Reservoir simulation models are capable of solving problems that are impossible for any human to solve manually in their lifetime.  Adding automatic deterministic or probabilistic optimization and forecasting around simulation makes our workflows the most complex and advanced computing systems in the world, infinitely more capable than humans in making complex optimizations and predictions of reservoir production and value.  So, according to our Specific Super AI definition, reservoir models along with optimization methods in our automated workflows are the most advanced Specific Super Artificial Intelligence applications ever developed.  In fact, none of ours or anyone else’s programs today are actually intelligent.  All of the intelligence behind any computer program is in the minds of its developers.  Machines and programs cannot learn, and the phrase “machine-learning” is nothing but a bad marketing term.  People learn by using machines and programs as tools, not the other way around.

A recent definition of "machine learning" was given by an engineer in an SPE Reservoir Technical Community discussion who claims that reservoir engineering problems can be currently solved by AI and machine learning applications (but can't give an example, even on request):

"Machine Learning includes a series of tools, techniques, and algorithms that make 'Artificial Intelligence' a possibility. The definition of 'Machine Learning' is using open computer algorithms to learn from experience (in form of data) rather than detail and explicit programming for the computer to perform certain tasks."

Since real artificial intelligence does not exist and is not currently possible, "machine learning" as defined above obviously does not exist either.

All current claims of achieving “artificial intelligence” and of developing “machine learning” applications are completely misleading and unsubstantiated.  "Machine learning" is an oxymoron.  Machines can't learn.  They can populate databases. Computers and programs can solve problems that humans can't, but humans have to provide the instructions to obtain the solution (program).  No computer program can learn or understand any system or problem or solution or improve itself.

Despite many requests, nobody has ever been able to demonstrate any improved solution to any known problem in engineering using  what they claim to be "artificial intelligence" or "machine-learning"  or "data science", which is another oxymoron appearing in the last 10 years.  "Data" is not a science.  There is nothing new or valuable in "data science" that we didn't already know. So until those claims are properly substantiated, as competent scientists and engineers we must assume that they are false. 

AI is essentially automation and integration, and is incredibly valuable. But it can't think and evolve. It can't solve new problems. We have ideas of a hybrid system involving validation of facts by the scientific method and debate, requiring human involvement, that would allow it to evolve in apparent intelligence, but only to the maximum level of human achievement. The ability to almost instantly apply or reproduce the maximum verifiable  human intelligence would be a huge leap forward. The system would appear to be far smarter than any human could possibly be. Today, AI represents evolution and integration of internet search engines, databases, and interpretive language models. Innovation, integration, and automation are still the main drivers of scientific advancement (see our Reservoir Simulation Goals page).