AI still has a ways to go
Posted on 10/09/2016 in misc
This is the first in what I hope will be a somewhat regular (3X a week?) series of quick blog posts about whatever I'm thinking about. A few years ago I decided to make my blog a repository for longer writing, and now I'm changing my mind and making an attempt to get back to writing something here on a more regular basis. Because these are quickly dashed off blog posts (how old school is that?) I probably won't be linking out much because the goal is to keep these to 15 minutes or so, again more like the rambling blog posts that were so common back in the 2005.
Artificial Intelligence is unquestionably getting better, and in certain situations I totally rely on it. I have learned the hard way to not question Waze when it takes me off my usual route going somewhere. That said, I haven't commuted to DC for work in three months, yet Waze keeps notifying me of rush hour traffic incidents on I-95. I held off deleting my "work address" in Waze, curious if it would figure out my new non-commute. It didn't. Google Maps (which owns Waze) did figure it out pretty quick, as the daily "as I'm leaving" traffic update stopped coming pretty quickly after my GPS location stopped changing.
Another example is Wells Fargo online bill pay. I put almost identical looking bills in last week, but they did have different reference numbers in the memo field. Wells Fargo sent me a message the next day that they had only paid one bill, and deleted the other, because they looked duplicates. The duplicate detection is a good thing, but it should have sent me a "click here to pay bill" option, not just arbitrarily deleted one of the bills. They weren't time sensitive, but what if they were? Wells Fargo could have just caused me to be late on a bill.
I'd use Facebook's content recommendation as a third example, but I'm not sure if it is a valid example. The content recommendations are generally poor, but that assumes Facebook's goal is to serve me content I want to see. If the algorithm's goal is to maximize ad revenue with no consideration for my personal preferences maybe it's serving me exactly the right recommendations. Twitter's AI also has a long way to go, as it has no sense of context. It assumes anybody you tag once in a tweet is somebody you want to follow. So it recommends I follow people I tagged once while mocking something they said.
Smart home devices and the Internet of Things are also interesting, but mostly in the sense of how insecure they are, and how unreliable cloud based business models are. If a smart device won't function without connecting to a cloud based service run away from it. Nest has already screwed over tens of thousands of customers by shutting down one of it's early products, leaving those customers with the technological equivalent of a brick on the wall where they once had a smart thermostat. And I have no idea why anybody needs an Internet connected fridge. The cost-benefit ratio of buying an extra gallon of milk because you couldn't look into your fridge remotely from Wegman's falls far on the side of enjoying that extra gallon of milk.
I don't mean to sound like an old guy standing on my lawn yelling at the kids. Maybe one of the things that comes with experience is a little bit of cynicism about the potential of all of this. Software based image processing may have just saved the life of somebody close to me, so I'm totally a fan of technology and it's ability to make our world better. However, I haven't seen any evidence that a Nest thermometer is actually any more beneficial than the old school programmable thermometer I've installed back in 2002. I have seen several examples of it being worse though.