On April 2, 1968, "2001: A Space Odyssey" was released by Metro-Goldwyn-Mayer. Stanley Kubrick’s epic science-fiction film was based on a screenplay adapted from the short story and subsequent novel, written by Arthur C. Clarke.
One of the major themes associated with the story was the concept of Artificial Intelligence (AI), which takes the form of the spaceship Discovery’s on board ‘HAL 9000’ computer.
In fact, the computer is simply called HAL, because it's considered just another member of the crew’ Despite Hal’s supposed status as "incapable of error," something indeed goes wrong. Hal not only starts having problems, but takes his problems out on the very humans he was intended to serve.
At the time, most people who saw the film had likely never even seen a computer. The idea that a machine would one day be intelligent enough to not only talk with a human-sounding voice, but be in control of essentially everything was beyond imagination.
After all, this was just science fiction, right?
While once considered the stuff of science fiction, Artificial Intelligence is here. Even today, many businesses are already taking advantage of what may still seem to some as impossible.
Artificial Intelligence, machine learning and deep learning are terms often used interchangeably, but they’re not the same. AI encompasses the much broader concept of machines that can act intelligently, and for the most part independently.
Machine learning and deep learning are simply sub-sets of Artificial Intelligence based on the concept that given access to large volumes of data, machines can learn for themselves.
Machine learning is intended to automate and prioritize routine decision-making processes. To accomplish this, they're configured to learn in three different ways.
Supervised learning is the first way machines learn. This involves humans labeling the various inputs and outputs, as well as reviewing and verifying the model the machine uses to connect the input with the output.
Let’s look at a potential example of how this might apply in an accounting setting:
Initially, humans might enter information about invoices [vendor, accounting code(s), amount due, due date, discount amount, discount date, etc.] and then select which invoices should be paid based upon cash available, due date and discount provisions.
With a sufficiently large enough sample of this type of data having been spoon fed via supervised learning, a learning machine could ultimately not only automate the labor-intensive matching process of accounts payable based upon a specific company’s historical patterns. It could, in fact, improve cash-flow by making decisions that are of the greatest advantage in terms of timely payment and discount maximization.
When Astronaut Bowman gradually deactivated the circuits controlling HAL's higher intellectual functions, HAL regressed to his earliest programmed memory – the song "Daisy," which he sings to Bowman. Obviously, as he first began his artificially intelligent life, HAL was instructed via supervised Learning.
"Daisy, Daisy...
Give me your answer, do.
I'm half crazy, all for the love of you."
The second form of machine learning is reinforcement learning, sometimes called semi-supervised learning. This is where the machine is rewarded or penalized for actions it takes via trial and error. The various algorithms created out of the learning activities are adjusted to insure they're accurate based upon the correct pattern(s).
Unfortunately, HAL didn’t take to reinforcement learning very well after he was already in charge of the spaceship Discovery. In fact, when Mission Control advised the astronauts that HAL’s twin back on earth indicated that a discovered problem was not actually a part going bad, but an indication that something was wrong with HAL, he insisted the problem was due to human error.
In other words, there was no need to penalize or correct HAL. The problem must have been elsewhere. Oops, it seems HAL had learned the human-like quality of never accepting the blame and always pointing the finger at the other guy.
The third and most advanced form of machine learning is unsupervised learning, where the machine is left to discover patterns in the data by itself and develop algorithms that define and match those patterns.
Deep learning or cognitive computing is another form of advanced machine learning. It uses multi-layered neural networks to simulate human thought processes. Believe it or not, these deep networks are made up of small computing nodes that mimic the synapses of the human brain.
Using input data sets and sophisticated algorithms, deep-learning machines can help solve complex, non-linear problems. Deep learning is responsible for many breakthroughs involving speech and image recognition and natural language processing.
Do you remember that HAL read the lips of astronauts Bowman and Poole when they intentionally turned off their communication links so that HAL couldn’t hear what they were saying about him? Clearly, HAL had already done some very deep learning to recognize the threat against him. Even the best lip readers can't clearly make out all words when attempting to read lips from a side view as did HAL.
We are being led to believe that machine-aided business processes can achieve faster workflows and optimize our business operations. In fact, the machines may be able to eliminate a majority of the slow-thinking, slow-reacting humans who prevent any business from running at high speed. But let us not forget what happened with HAL.
Ultimately, HAL 9000 had to be shut down completely. After all, he had already murdered the entire crew on the spacecraft, except for Bowman.
The idea of machines taking over our lives and livelihoods makes for some great movies, but the reality, at least now, is that no matter how fast things are moving, they are far less dramatic. We should never put blind faith in outcomes resulting from machine learning, but neither should we abandon our human role of supervising the machines and making adjustments in what's truly under their control.
In other words, like Bowman, we should be more than ready to pull HAL’s plug.