Defining Artificial Intelligence
It seems every Data Science, Machine Learning, or Artificial Intelligence seminar I sit in on has their own definition Artificial Intelligence. I have determined that there are two distinct classes of AI definitions:
- AI is not obtainable: once a new task is achieved, it is no longer considered AI.
- AI is any decision a computer makes: from an expert system, to predictions or modeling.
Both of these definitions are very encompassing. One says that nothing is AI while the other says that everything is AI. I believe they are both true based on which side of the argument you want to stand on. I’ve spent too many years tailoring myself to be opened minded; I don’t actually care which side someone lands on. My only argument is that one of these definitions is true, anything in between is just an incomplete argument.
AI Is Not Obtainable
This belief originates from a historical look in the pre-90s AI discussions, many believed that we will have achieved AI once a computer can best a human at chess. In 1996 IBM’s Deep Blue beat Garry Kasparov (Grand Master and top rated player in 1996) in the first round but eventually lost the tournament. However, in 1997 Deep Blue won enough matches against Kasparov to become the first chess computer to beat a reigning champion in an official tournament. While Kasparov denies the credibility of the tournament, the field of Artificial Intelligence took this as a huge win.
Winning Chess Algorithm
It may surprise many that the winning algorithm was not machine learning by design. There were no neural networks or regression performed, it was a pretty benign search algorithm called Alpha-Beta Pruning. These algorithms are crucial to expert systems. They are often cheap and easy to implement for many problems and do not require the amount of training data that learning systems do (especially deep learning systems). Would anyone consider a search algorithm to be AI anymore? What about sorting? I surely wouldn’t put bubble sort into the AI category.
Boring Algorithms
I find a surprising distinction between many hard-core engineers and business-oriented engineers. Hard-Core Engineers are focused on solving an engineering challenge using the latest technologies and have difficulty during the last-mile of implementation. Business-Oriented Engineers are focused on solving the business case and are drawn to the simplest solution.
As a convert from hard-core to business-oriented I see the appeal of both, there are a couple quotes that I feel extend this thought.
Everything should be made as simple as possible, but no simpler. – Einstein (among many others)
and
Before I learned the art, a punch was just a punch, and a kick, just a kick. After I learned the art, a punch was no longer a punch, a kick, no longer a kick. Now that I understand the art, a punch is just a punch and a kick is just a kick. – Bruce Lee
For me, after enough years solving problems I find that new technology is similar to old technology, new business problems similar to old business problems. Artificial Intelligence will always be the next yet-to-achieve technological milestone. For those in the know, a self-driving car is a huge hurdle, but the paths from here to there are all engineering challenges.
Everything is AI
If nothing is AI then perhaps everything is AI. If AI exists, it exist as a replacement for human decision, that much is clear. Humans decide how to operate a car, strategize chess moves, and adjust a thermostat based on temperature. Is the sleep function on your TV intelligent? If a human is required to push the power-button, but we design a computer system to do it for us, is that not artificial intelligence? Does this not duplicate human actions?
Magic Is Not Explainable
Once a human programs a task, the AI ‘magic’ is gone. However, that is no reason to change the definition of what constitutes intelligence just because we can explain it. The debates will continue, back and forth, on the definition of intelligence , but Merriam-Webster currently has 5 definitions for Intelligence, one of which is relevant here:
the ability to perform computer functions
All other definitions lead one away from the basics of simple computation that would indicate a computer can never be intelligent. According to this definition, any computer function constitutes intelligence .
It’s All Marketing
Whether something is or is not Artificial Intelligence makes no real matter in solving a business problem. Unless of course that is a marketing related business problem. In that case, go with your gut. Would I claim a basic TV with a sleep function as AI? Nope. Would I claim a chess playing computer using simple search heuristics is AI? Maybe. Would I claim a deep-layered neural network is AI? In 2018 I will, in 2030, maybe not. In the end, focusing on the problem and all available tools is what one should be most worried about.
Let marketing figure out how to sell it if it needs sold. Internal reporting? That’s company politics and it’s best to choose your battles. If you have a manager who insists on selling your simple linear regression forecast as a advanced artificial intelligence, there’s little you can do but grin and bear it. Educate where you can, but in the end, focus on the real problems.