Thoughts On Superior AI?

Looking for input from my brilliant friends to understand whether Skynet will ever actually be a threat... ☹. Assume you teach a system, let’s call it AI Robert or Bob for short, to review data for patterns (e.g., cat, dog, etc., thanks, IAPP AIGP instructor, friendly tail wag and whisker twitch for you 😉). Assume as part of Bob’s education, you are able to provide it access to data comprising all tangible knowledge of all time, anything ever seen, heard, written, taught, recorded, or demonstrated, with the goal of making Bob as capable and “human” as the combined human race, something not currently deemed imminent, as far as I’m aware (no pun intended).

Bob might continue its machine training with this data in part by identifying “patterns,” discerning connections, to further categorize and label everything about human history, knowledge and behavior based on the entire tangible knowledge in the dataset. I’m assuming, as a layperson, that categorization would extend to the inherently outlying elements, i.e., particularly the “best” of everything. The most brilliant genius, the most accomplished businessperson, the most talented artist, and so on.

So how would Bob go about categorizing the outliers, the rebels, the eccentric geniuses so it can understand, forecast and mimic their respective behaviors? Perhaps Bob would be trained to analyze all “adjacent” but potentially relevant data, such as environmental, societal, cultural, or otherwise influential (but still tangible data), to identify and further label and (sub)categorize connections based in these data points. For example, assume hypothetically, that evil geniuses are statistically more likely to have vitamin D deficiencies, Type AB blood, gluten sensitivity, and ambidexterity.

Compare by analogy a sensory-adaptive behavioral perspective: assuming that (a) “most” people enjoy leisurely eating, chatting and socializing, and (b) a large part of a particular restaurant’s success is determined by volume served, might that restaurant’s bottom line benefit from using hard chairs, harsh lighting, generic décor, and disposable, inexpensive and portable utensils, among other things, to “encourage” customer behavior that differs from the statistically predicted behavior?

Anyway, each time Bob makes new connections, it gets “smarter.” At that point, “trained” Bob should be able predict to what is otherwise unpredictable, at least to human brains which, as far as my undersized cerebrum can understand, only have the capacity to identify, label, categorize, understand and extend connections in data at a finite pace and level. Computers and genius algorithms don’t need sleep and probably have a longer shelf life. But AI Bob can only be as completely and comprehensively “smart” as it can be “trained” to be. Everything Bob “learns” is initially derived from categorizing raw data and other information and forecasting behavior. Can Bob “know” what it doesn’t “know”? Bob isn’t going to autocorrect “shoot” to “sheriff” unless someone has told it to ...

If each evolution of Bob’s impressive artificial intelligence depends on its ability to identify, qualify, quantify, and forecast, won’t there be some amount of material that is “unknowable” and therefore “untrainable.” How will Bob be taught to comprehensively, accurately and consistently forecast the most puzzling of human behavior when humans can’t always explain the unfathomable. Ironically, isn’t Bob’s existence (and moder technology) largely the “brainchild” of the most brilliant, creative, visionary, etc., humans who don’t share space with most humans on the bell curve?

For what it may be worth, when it comes to fears of Judgment Day and Self-aware AI Bob, I’ll take my chances with the brilliant geniuses over Alexa any day. That doesn’t mean I’ll say it out loud just anywhere .... I’d have to lock myself in my office with no phones, fit-bits, or smart devices and only after I don my tinfoil hat.

Just sayin’