Humans and Keeping AI Real
While not still in its infancy, obvious by the data presented in the articles The State of Machine Learning and How Companies Use AI, artificial intelligence could be said to be in the larva or pupa stages—developing but not yet full-grown—nor taking over the world of humans and their jobs as so many fear. It also was clear in the readings that the human component to artificial intelligence continues to be critical for ongoing quality control.
It goes without saying that without humans there would be no artificial intelligence, but even though A.I. has been ‘taught’ to ‘learn’ on its own, humans are necessary to its continued improvement — this could include eliminating cultural faux pas or updating language to keep up with new social mores. A good example, in the Meet the People article, was the edit from the out-of-date phrase “on the boss’s calendar” to “find a time.” This type of review and editing would likely need to be an ongoing endeavor, although it would also vary in degree depending on the machine, its uses/users, and the culture in which it is embedded.
Humans are also a necessity for the data, content, and therefore assumptions from which A.I. works. Even though data and content can be input from other machines and the logic formulas based on the data, humans are needed to help make sure there is some check on these to reduce “garbage out.” Innocently, the author, in the Meet the People article, mentions a software update that would include asking lifestyle questions to “glean clues” about a user’s preferences. While the description of who Snapchat users tend to be may be supported by data and statistics, these do not necessarily support the broad assumption that is made about what type of hotel a Snapchat user may prefer. This is one minor example of the vast potential for A.I., without human monitoring/editing, to proliferate stereotypes and reduce responses to those fit for caricatures rather than real people.
Overall, I wondered, with the rapidly increasing volume of A.I.-human interactions like chatbots, travelbots, etc., how will it be possible for humans to keep up with these types of quality control aspects—how is it possible to even just spot-check millions of interactions, much less correct them? Hopefully we’ll see the value and use of human oversight and intervention become an ever-expanding and deepening component of the A.I. field — perhaps it already is.
