I keep meaning to get to my notes of things that occurred to me while daydreaming and observing on my trip, but other things intervene. Though I am retired and work only 8 days a month, I am covering a two-week vacation that overlaps May and June, and my ability to efficiently take care of the lawn, laundry, shopping, and cooking has clearly deteriorated since official retirement.
****
Many thoughtful, and sometimes frightening pieces have been written about The Singularity, when artificial intelligence surpasses the human brain. Some see it as a liberation, when people can have machines be their servants, doing all the unpleasant tasks, which won't bother them. Others worry that it is impossible to install the layered nuances of generations of morality and common sense in a machine, and where they will take us is unpredictable. Let me put down two anecdotes. I needed the fax number for the ER at St Josheph's Hospital and was Ducking that online. There are many St Joseph Hospitals, so I added "Nashua" to the search. The duck offered me autocompletes immediately, and the first was St Joseph's fax Nashua Cynthia. The machine doesn't know that whatever few searches justified this suggestion, it's still crazy, and a little creepy. Today when text with my son in Houston, the only possibility it wanted to give me for attempted "Gandalf," was Gandolf. I have since learned it is a surname used in America - yet it can't come close to outnumbering Gandalfs - and that is considered a medieval variant of Gandalf, though it seems to have only been used in obscure sources. I can forgive autocorrect for butchering Gesta Danorum. But this is what superintelligent machines will continue to do. Will they be smarter than us? In many ways they are already far smarter than us.
I don't worry that AI will develop some competitive machine-tribe savvy that will trick us into jumping off buildings to save the planet so that they can have all the zinc to themselves. I worry that they will continue to be both brilliant and stupid.
Much of the AI discussion seems to be predicated on the notion that intelligence is a matter of connecting bits of information, rather than anything you or I would think of as understanding. With a big enough dataset it can look similar, and even make predictions that work pretty well, but it feels like piling up integer addition facts without ever making the connection to fractions, much less groups.
ReplyDeleteNo, I don't believe that you can always classify a thing only by some set of its operations.
Facts and data. Even AI is dependent on GIGO, isn't it?
ReplyDeleteI haven't read it yet, but this critique of current AI hype looks interesting:
ReplyDeletehttps://blog.piekniewski.info/2018/05/28/ai-winter-is-well-on-its-way/
An interesting post at the GE blog suggesting that it may be more useful to think of AI as analogous to nonhuman-animal intelligence than to human intelligence.
ReplyDeletehttps://www.ge.com/reports/understanding-animals-can-help-us-make-artificial-intelligence/
I've been following the AI hype since the early 1960s. It's always been just around the corner - just like fusion power plants and flying cars and The Rapture and ...
ReplyDelete@ Roy - and Brazil becoming a major power.
ReplyDelete