By: Greg Benjamin and Yobie Benjamin
Since the ENIAC to the quantum computers of today, computing machines have been largely dumb bits of metal, silicon, plastic and advanced materials put together as tools for humans to program. There is probably an exponential number of technologies that came from these elaborate feats of engineering. From essentially sand to one of the most powerful tools on the planet.
There is one eternal truth in computer science and that is “garbage in, garbage out”. What we feed the machine is what it will act on to return some result. Computer software has similar attributes as hardware. Perhaps, it’s more flexible because you can always do an upgrade, bug fix or build an entirely new release.
Artificial Intelligence has been the buzzword of the last couple of years. Movies make AI so real and Elon Musk makes it scary. The Terminator is etched in our collective brain conjuring up runaway robots and computers autonomously battling humankind for planet dominance.
I certainly don’t know everything about Artificial Intelligence but I know more than most people. Without getting into deep computer science, allow me to oversimplify a few frameworks that I know.
Quickly there’s Caffe2 (deep-learning framework designed to easily express all types of modeling); Cognitive Toolkit (a deep-learning toolkit that describes neural networks as a series of computational steps and now owned by Microsoft); MATLAB (used to create and visualize models, and deploy models to servers and embedded devices without being an expert); the very popular TensorFlow (for numerical computation using data flow graphs); Chainer (a flexible deep learning framework. It uses define-by-run approach known as dynamic computational graphs to build and train neural networks); then there’s PaddlePaddle (an intuitive and flexible interface for loading data and model structures. It configures complicated deep models easily and makes it easy to scale heterogeneous computing resources and storage to accelerate the training process).
All of that AI mumbo-jumbo will send you running for your dose of ibuprofen. That’s not the intent. The single most important characteristic that crosses all the AI frameworks is the need of data - clean data, unbiased data, smart data, real data. From there it will process and learn and execute within the boundaries of the framework used.
Before a computer becomes an artificial intelligence device, it is just a hunk of metal, silicon and plastic. Add an operating system, sprinkle some AI frameworks and it’s ready to take in data. If the data is dumb and stupid, it will learn nothing else but dumb and stupid. Human intelligence is the source fuel for data fed into an AI computer.
I have seen stupid before. Computers that are fed dumb data like most senior executives are men may assume it needs to exclude women from its executive search algorithm. I have also seen that one influential technology outlet claim that most successful VCs are between 30 – 45 years old. That type of data can be interpreted by an AI computer as a rule rather than see the nuance that only humans can infer and ascertain.
In world of robot scrapers like your Google search bot and Facebook feed algorithms, it’s possible to keep piling bad information on top of really bad information. Think back a bit to how foreign powers fed bad information to affect the 2016 elections. Some of the most sophisticated AI computers at Facebook magnified the dumb and stupid like a child porn ring under the basement of a pizza parlor.
As Forrest Gump said, “My mama says that stupid is as stupid does”.