Reading coverage of the Consumer Electronics Show, it does not appear that the robots left anyone terribly impressed. We all have images of Rosie the Robot from The Jetsons, which effortlessly and with little instruction just does all the housework. The reality is that this is not coming any time soon. Folding laundry turns out to be really, really hard. So the Consumer Electronics Show features a lot of wonky systems that don’t work very well. Gadget junkies may be entranced, but regular consumers don’t need an expensive toy that will take longer to get to do a task than it would take to just do the task. (If we enjoy that kind of frustration we can just try to get our kids to do their chores.)
Consumers want devices that, with limited set-up, work. Robots, autonomous systems, are run by machine learning algorithms — that means they have to learn. Learning means they have to be taught, it also means they will make mistakes. Who the hell has time for this kind of thing?
For some applications, this can work. Smart thermostats are an example. Preferences are not that hard to learn and mistakes are not catastrophic. But think about a smart refrigerator that could inventory your food, recommend meals, and order necessities. This sounds great, but then consider the problems in implementation. For starters, in a typical home there might be a dozen places where food is stored — not just the refrigerator. Now you need smart pantries and spice cabinets, etc.
Let’s imagine that this smart home can monitor your food supply, make recipe recommendations, identify when items are needed and then order them. This would actually be pretty great. There would be hiccups of course, plus the system would have a lot of pretty personal information. There might be a time consuming learning curve as you taught it what recipes you liked and what kind of food you wanted to have around. But it could work. It might even be fun.
But here’s the ugly secret. The thinking is relatively easy. The sensing is very hard. Presumably all of the food going into your smart home has some sort of electronic marker allowing your kitchen to record it. Then there will be a bevy of sensors to follow the consumption path of each item. This is a lot of sensors and a lot of opportunities for things to go wrong. Then you would need to teach it what was a priority. (Order pasta when we are down to one box etc.)
It is all theoretically possible, but there are so many things that can go wrong and working with it might be more trouble than it’s worth. You can just scan your fridge and pantry and make a shopping list in a lot less time.
But that doesn’t mean robots aren’t coming.
This smart refrigerator or kitchen is not worth your time. But a hospital refrigeration system for storing blood and medicine is a different story. The hospital already has a staff and systems for monitoring these things. The value of the items in question are very high — both in cost but also in consequence (you need to have the right supplies on hand and usable.)
If a hospital recognizes that there will be significant gains in efficiency using a smart refrigeration system, they can make the investment both in money and time to learn the systems and blend their existing processes in with the system’s functionality. Hotels, restaurants, and a vast range of other organizations might also be willing to make this investment. Managing food inventory is a huge issue for the hospitality industry.
One can imagine lots of other smart systems that might not be worth the effort for a home, but would be for a large-scale enterprise. This will lead to an iterative process. The first enterprises to purchase the smart refrigeration system would probably be larger (maybe not the biggest) and have some experience integrating new technology. In turn the vendors would also learn what it took to meet customer needs and be better positions to market their systems to more conservative buyers. In time the technology would become commonplace, lowering costs (again both financial but also in the time required to learn the system) and — at some point — maybe becoming a consumer product.
At this point the autonomous systems in questions are not there. The models will need to learn, but at the same time we need to learn the models.
Originally published at terrorwonk.blogspot.com on January 17, 2018.