Artificial Intelligence: Fact & Fiction.
What is AI?
What is Artificial Intelligence? Ask 100 people and you will get 100 different definitions. Many of these definitions will echo the visions from movies like the Matrix or Terminator franchises, in which superintelligent robots enslave or eliminate mankind. But these are movies, not the real world. Real AI is useful AI, and has been benefiting humanity for years already. Useful AI diagnoses unseen illness, fights credit card fraud and identity theft, cures diseases and thwarts terrorist attacks. For me, AI is about creating better algorithms, implemented as software, to address these real-world problems.
My definition is much more in line with scientists in the field: technology that uses information to adaptively learn when making decisions about specified goals. This definition is purposely narrow. It doesn’t include ideas about consciousness, (another hard-to-define concept) or self-awareness/sentience. Why? Because I do not think machines that run software will ever achieve these states. My reasoning is two-fold, which I will refer to as “bound from the top,” and “bound from the bottom.”
Bound from the top: lacking imagination
Artificial intelligence, at its core, is about pattern recognition. This makes it ideal for things like computer/robot vision, self-driving cars, automatically setting thermostats etc. AI algorithms come in two “flavors:” supervised and unsupervised. In the former, AI models are trained on past data sets to recognize objects or patterns in new, not yet seen data sets. Popular techniques in this category include Neural Networks (including “deep-learning”) and Data Mining. In unsupervised learning, there is no past data to use to train AI models. In this case, the models automatically identify patterns or clusters in a set of data. Well known approaches in this category are associative learning and clustering.
Whether supervised or unsupervised methods are chosen however, the goals of AI models are specified by humans, and the objects and patterns identified represent things that exist in the world. This is the key distinction between human and artificial intelligence: the human imagination creates. This creativity enables the human mind to join seemingly unrelated concepts in novel ways to create new ideas — things that do not yet exist in the world.
Bound from the bottom: deterministic execution
Computers are unmatched at doing simple, well-defined tasks over and over again. Programming is all about figuring out how to break complex problems into these tasks. AI algorithms and models, as programs, follow these same rules and are bound by the same limitations of functional programming. There is no initiative in the “zeros and ones” that make up the binary language of computation, no “ghost in the machine” making decisions beyond the boundaries of well-defined routines and procedures. There will never be the spontaneous, random firing of neurons that happens when we dream, or the random mixing and mutation of DNA that spurs evolution.
AI will continue to do what it does best: help humanity solve difficult problems by finding patterns in data. There will be no AI apocalypse or intelligent machine evolution. While “Artificial Intelligence” and “Superintelligence” make good boogeymen, we should get back to higher probability doomsday scenarios, like the Zombie Apocalypse…