Hacker News Reader

Comment by OnlineGladiator | Hacker News Reader

Everybody thinks robots are inevitable until they meet the robots that are supposedly inevitable. There's a fun joke template for robotics. "How many PhD's does it take to do <generic task>?" So now instead of one untrained person doing the dishes, you have 5 PhD students hovering around a robot as it spends 8 hours poorly cleaning one.

View on HN

>"How many PhD's does it take to do <generic task>?" So now instead of one untrained person doing the dishes, you have 5 PhD students hovering around a robot as it spends 8 hours poorly cleaning one.

Yeah, and then it spends 8 hours cleaning another, and soon it'll be spending 5 hours cleaning two, 2 hours cleaning a dozen, and it'll keep going down, and those 5 PhD's will keep improving the robot, more and more. And it doesn't even need to be faster than us, because that one robot can keep washing dishes 24/7, with no rest and no pay.

And eventually it'll catch up to us in speed and precision, and it'll keep getting better and surpass us.

And then, once you have a single good robot. You can simply replicate it, again and again, mass producing thousands of them and they will all keep working 24/7 with no rest and no pay.

Yeah, we're not there yet for most jobs, but we're getting there. It's gonna take a while but it will without a doubt happen

It will then no longer be referred to as a "robot" or "AI", will look nothing at all like how it started (probably just a big box), and people will call it "the dishwasher" as they wonder why AI and Robots are always 10 years away.

Yes, but at a certain point you will have automated away most of the labor that needs doing.

> It's gonna take a while but it will without a doubt happen

Vehemently disagree. The assumption that technology will continue to improve to be useful for everything is inherently flawed. We don't know what the hard limits are yet, but nobody understood the limitations of the Carnot cycle for energy efficiencies once upon a time (the ICE will get better and better until no efficiency is lost) or limitations of transmission lines (until transmission line theory was established that dictates 50% is the best theoretical outcome, people thought you could approach 100%). We'll discover some algorithms will never be O(n) but the best theoretical solution is O(n^2).

I don't know where the barriers are or what they are, but it's naïve to assume not only that we will solve everything, but even that it's theoretically possible that we can solve everything.

The obvious generic barrier right now for robotics is AI, but that's a complex subject and could mean many different things to many different people.

Aren't dishwashers (and washer/dryers) the perfect example of automation that actually got widely adopted though? I'm sure those machines took a ton of R&D back in the day, but now they are ubiquitous because of how much effort they save.

The question is not one of how many phds you have to pay to see if it's possible to make something automated - it's whether the automation solves more pain than it creates, how big the demand is, and whether production of it can actually scale at profit. If it does, the research investment can be well worth it.

Yes, but they're single purpose machines optimized for exactly one thing. Humanoid mobile robots continue to suck in large part because they have to do many things, and are thus not particularly good at any of them.

I have met the Walmart order pickup robot (the so-called Pickup Tower [1]) - it has convinced me that well-designed robots are inevitable. In general, it has made package pickup a pleasure for me though I admit that in the early days, it required some employee assistance. I have picked up many tens of packages from this facility at my local-ish Walmart and it has been wonderful. By comparison, the previous manned package pickup scheme was almost always a nasty experience: non-working 'buzzers', long waits, 8pm close up times, and other unnecessary nonsense. I used to dread picking up an order at Walmart before the Pickup Tower became available and now go out of my way to go to a Walmart with this pickup tower vs. going to a closer location with the ring-and-wait-around-like-a-flipping-fool package pickup service.

[1] https://corporate.walmart.com/newsroom/innovation/20180405/h...

I have not been, but how is this different from lockers with electronic locks? I don't consider that anything resembling a robot.

There are two aspects to the pickup tower:

1) Entering/scanning an order number causes the robot to fetch the packages (plural) in question and deliver to an area for the recipient to pickup. This has the effector/sensors that you mention above.

2) Entering/scanning an order number causes the robot to tell the recipient to go to a given locker and it also pops open said locker. When the recipient picks up the packages from the locker and closes the same, the main screen acknowledges the pickup.

If I appear to be gushing, it is because I am darned impressed by this thing. I am not kidding when I say that it works very well. I found a video that hopefully demonstrates the technology better than my stilted write-up above: https://www.youtube.com/watch?v=71pgd-LzzXc .

Ah, the video helps explain it. At first I only saw pictures of the overflow lockers.

Well regardless of whether or not it's a robot (I would argue it is), it passes the most important test: it does something useful and users love it (going by your testimony as a single data point)!

"How many workers does it need to assemble a car" has gone down a lot due to robotics.

None of those robots look like a human though. The CNC metal cutter is a large box. The welder is an extra arm with a lot more joints than a human (or sometimes a track).

Yes, robot technology is improving nicely. But many problems don't really need "robots". Often something far simpler will suffice.

For example, a "smart" moving base may replace waiters(https://www.youtube.com/watch?v=p1cjhxEhq9Q) . i suspect this could have been built a decade ago. But something has changed, and we'll probably see many more such solutions.

But that's not actually simpler; there's two more solutions to whatever problem that thing is trying to solve: Self-service / buffet style restaurants, or the little conveyor belts with continuous dishes you see in Japan.

The complexity comes when you try and retain an existing workflow by a robot. You see this same thing when it comes to package delivery or warehouses, where they try to replace a human with a robot instead of redesigning the whole process as robot-first (this mainly aimed at warehouses btw)

On the nail. Automation is a tool not a solution, and like any tool its best use depends on context - including not to use that tool at all. Thanks for the reminder.

Yeah, but the cleaner would need to earn at least minimum wage, while grad student will happily do it for free if it helps them with their PhDs. Might even thow in some perks for good behavior, like exemption from didactic work next year.

So, to summarize your post "Everything that didn't exist when I was born will never exist".

My point is robots aren't nearly as advanced as people think.

I acknowledge your point but somewhat relatedly, my washing machine is just as advanced as it needs to be.

That just brings up another classic robotics joke. A robot is anything that doesn't work well, yet. Once it works, you give it a new name - like dishwasher.

Exactly!

What most people think of when they think of robots though are machines that can adapt to their environments - usually humanoid, but definitely dynamic. Dishwashers and washing machines are essentially mechatronic, and while they have sensors to adjust for varying loads they aren't really reactive like people expect from a "robot." The only thing that really fits the descriptions is the robot vacuum cleaner, because it will work in an environment designed for humans (sort of, they're known to work in limited use cases). The Boston Dynamics Spot robot would also be considered a robot, but it's obviously too expensive for consumers and I suspect it will struggle due to limited usability. Humanoid robots are still completely within the realm of R&D and not really commercializable yet.

So what I mean when I say robot, is a machine that can adapt to a human-oriented environment and fulfill a task with some level of competence that people don't consider it useless. And right now we basically have vacuums, and are a long, long way away from getting much more.

Ditto with AI. Recognizing types of objects in images was AI until computers could do it. Playing chess well was AI until computers could do it.

My washing machine just washes. I need it to collect, sort by color, wash, dry, fold/hang and put away the laundry too. It does one of those well.

Maybe you will be able to get one of these (before or later):

https://www.youtube.com/watch?v=COXM2V0wn_I

You will need also a bigger house, probably.

Yes, although historically washing was almost certainly the most labor intensive part of the process. There are combo washer/driers but, mostly, it makes more sense to keep them separate given that transferring from one to other is pretty low effort.

Same thing applies to washing dishes.

Reply on news.ycombinator.com