After IBM Failed To Sail an Autonomous Boat Across the Atlantic, It's Trying Again (washingtonpost.com) 69
After failing its first attempt to re-create the Mayflower's voyage across the Atlantic Ocean, a crewless ocean vessel, powered by artificial intelligence, has returned to sea. From a report: Propelled by IBM's AI software, the autonomous ship set out in June for a month-long excursion through rough waters with no humans aboard. However, three days into what was supposed to be a monumental journey from Plymouth, England, to Plymouth, Mass., where pilgrim travelers settled in 1620, the robot ship suffered "a minor mechanical issue" according to ProMare, a nonprofit promoting marine research that is behind the project. Researchers pushed out a software update, signaling for the ship to reverse course. The boat abided by its orders and headed to shore. Yet according to Brett Phaneuf, co-director of the Mayflower Autonomous Ship Project, the organizers quickly began planning another voyage. "We've had a setback, but one that will put us further ahead than if we did nothing," he said. Earlier this month, researchers sent the ship back out for a shorter trip: This time it'll focus on the waters around the United Kingdom, where crews can attend to it sooner if something unforeseen happens. "At some point, you have to go for it and take the risk or never improve," Phaneuf said.
At first you donâ(TM)t succeed (Score:1)
Try try again. No progress will be made without failure and revision leading to innovation.
Re: (Score:2)
"Try try again. No progress will be made without failure and revision leading to innovation."
Indeed, perfectly normal, no idea why this is news though.
Re: (Score:2)
sometimes the stubborn failing people retry and die, then it's major news
Re:At first you don't succeed (Score:2)
It is news because the population have became anti-science and overly political nitwits.
Politics is about finding what went wrong, to show that the idea is bad.
Science is about finding what went wrong, and altering the idea until it fits the results or it works.
Being that nearly everything now is considered a political firecracker. I am sure you get people who area afraid that AI will replace they jobs, and had voted in people who state that they will stop AI from taking their jobs, would jump on the failur
Re: At first you don't succeed (Score:2)
Given IBM's stellar do no wrong history, I can't imagine how this could be in any way be a "political firecracker"
https://en.m.wikipedia.org/wik... [wikipedia.org]
http://www-cs-students.stanfor... [stanford.edu]
Re: (Score:1)
That's the wrong half of the lesson though.
If at first, you don't succeed, LEARN FROM IT. :)
And *then*, try again.
Otherwise it's the definition of insanity: Trying the same thing and expecting different results. ... well, the learning part is what makes it a more likely world for one to succeed in. ;)
(Although technically, a world in which you tried once is already a different world as one in which you didn't try. But
Re: (Score:2)
Well, it can be worth trying again until you get a statistically relevant sample.
Re: (Score:2)
One only has to look at how humans got into space. A lot of rockets blowing up in often spectacular fashion, and even some fatalities. It's the story of technological development... "Well, that almost kinda worked..."
Re: (Score:2)
Yeah, but I still can't get over the fact that someone no-one tried to physically bar the Apollo 1 astronauts and end up being dragged away by MPs. There were plenty of people with chemistry backgrounds who had to have known that people being sealed up in an over-pressurized pure oxygen environment full of electrical switches and other controls and also of brand new wires and other plastics and foams and spacesuits etc. - all outgassing - was basically a death sentence. You can certainly learn from mistakes
Re: (Score:2)
Engineering often involves trade-offs. In this case, they could have used a nitrogen-oxygen atmosphere, with the risk that the astronauts would get the bends in case of a sudden drop in pressure. Or they could have used a helium-oxygen mix, with the risk of squeaky voices causing misunderstandings and the need for more electrical heating (because helium is a good conductor of heat). Either of these choices would have required higher total atmospheric pressure to provide the required amount of oxygen, which
Re: (Score:2)
Sure, but the tradeoff of putting the astronauts in a deathtrap. It wasn't just a 100% oxygen atmosphere, it was over pressurized to simulate 5 psi on the inside vs the vacuum of space on the outside. It was just a bad, bad idea. As far as the concerns, I don't think they were worried about the bends in a sudden drop in pressure. The pressure difference just wasn't high enough and they had pressure suits. Maybe an air embolism could happen, but still not that likely. Their main concern was weight, and a hel
Re: At first you donâ(TM)t succeed (Score:2)
They log when every order is given, every switch is thrown, when every system changes state.
It isn't hard to put 2 and 2 together, and fire investigators are very good at determining the ignition point of a fire.
Re: At first you donâ(TM)t succeed (Score:2)
I don't know how *nobody during the various phases of that clusterfuck stepped back for a moment to allow their heart to drop into their bellies over the catastrophie that they were in the process of creating.
Spacesuits melted to the seats.. yeah, that is one mental image that is fucking impossible to erase.
*Actually, I'm sure many people did, but scince we were in a dire race with the commies to get to the moon, their hearts dropped out of their buttholes at the thought of the personal consequences for spe
Re: (Score:2)
Try try again. No progress will be made without failure and revision leading to innovation.
Plenty of Darwin Award winner's next of kin would disagree with you. Sometimes failure just means you were not merely wrong, but dead wrong.
And besides, it's kinda hard to claim this outdated "defense" when we DID succeed with this. Four-hundred years ago.
Asking AI to do things humans are incapable of? Impressive and purposeful.
Asking AI to re-invent the wheel in a world full of perfectly functioning wheels? Stupid and full of marketing bullshit.
Re: (Score:2)
The modern world is built on people who died to advance it. Sometimes that's what it takes.
Automating menial tasks so that humans are freed up to do new things is the story of civilization. It's not stupid, nor marketing bullshit.
Re: (Score:2)
Automating menial tasks so that humans are freed up to do new things is the story of civilization.
You are utterly failing to take into account the Disease of Greed that mankind has been infected with for thousands of years.
Those humans that find themselves not performing "menial tasks" anymore don't go off in the future and simply find prosperity doing "new things". They find themselves unemployed. And eventually unemployable. Yes, the human that used to be a respected and trained ship Captain who's done that job for decades, is nothing more than a worthless unemployable human now. Along with million
Re: (Score:2)
Ah yes, truly, more people are unemployed (and starving to death) today than at any other time in the history of the world! In fact, humans ceased to exist in the mid 1800s because of this. Shoulda smashed those looms!
Re: (Score:2)
Evidently you didn't read the linked article: "... researchers hope the ship will still be able to collect data about the ocean, a money-saving prospect if the information can be gathered without humans onboard. Sending humans to study the seas is risky and expensive."
Re: (Score:2)
Re: (Score:2)
It's how Europeans got to the Americas in the beginning; with Portuguese traders sailing ships down the coast of Africa, and in the process making important innovations to maritime technology that allowed Europeans to build ships capable of reliably making the Atlantic crossing.
Re: (Score:2)
A lot of the ship building and control innovations came from the Viking/Rus slave trade, which rampaged across northern Europe and as far afield as Italy and (Canaries? Azores? Forget which one). They were incorporated into western Mediterranean ship building, then navigation advances such as the compass appear to have arrived after the fall of the Song Dynasty, and map making techniques from the Middle East (necessary to know what direction Mecca was to orient oneself for daily prayers) rounded out the s
Re: (Score:2)
Re: (Score:1)
Watson = iGilligan.
Nothing says AI superiority... (Score:1)
...then trying and failing to do something that basically primitive humans and matching "technology" managed to accomplish 400 years ago.
Re:Nothing says AI superiority... (Score:4, Insightful)
Nothing says 'didn't bother reading TFA' like a comment that impies 'AI' had anything to do with the failure. It was a broken generator. Guess what? Generators break on manned ships too.
Re: (Score:2)
Re: (Score:2)
Nothing says 'didn't bother reading TFA' like a comment that impies 'AI' had anything to do with the failure. It was a broken generator. Guess what? Generators break on manned ships too.
Yes they do. The difference is the humans on that manned ship are sometimes capable of fixing that problem without having to shitcan the entire trip.
Unlike AI.
But tell me more as to "who" to blame to prove me wrong. The ship "captain"? The maintenance crew? Oh yeah, that's right...there is no human aboard to blame, and my statement stands. AI was in charge of that vessel, and AI failed to resolve the problem.
If you want to NOT blame AI here, that's fine. But now I'm going to eviscerate the "intelligen
Re:Nothing says AI superiority... (Score:4, Insightful)
Do you not understand the concept of research? They aren't claiming to have made a replacement for humans. They aren't even claiming to have made an autonomous shipping product. They are trying to find out if machine learning can enable a crossing of the ocean. This is a challenging problem, no matter how long humans have been doing it. You don't think every human that sets out to cross the ocean succeeds on their first try, do you?
They have a set of things they want to learn from the experiment. It is highly unlikely that one of those things is to see if the AI can switch to a backup generator if the primary one fails. That is trivial, so why bother including the expense of showing that?
IBM has a history of doing technology demonstrations, and they always just focus on the thing being demonstrated. They did not include 'physically moving the chess pieces' when they showed a computer that could beat a grand master. They did not include 'reply with synthesized speech' when they showed they could beat Jeopardy champions. Neither one of those things was germaine to what they were demonstrating. Nor is 'switch to backup generator' germaine to what they are trying to demonstrate now.
Re: (Score:2)
They have a set of things they want to learn from the experiment. It is highly unlikely that one of those things is to see if the AI can switch to a backup generator if the primary one fails. That is trivial, so why bother including the expense of showing that?
Because it had a non-trivial possibility of occurrence, DID occur, and the consequence was a failure to complete the experiment.
Nor is 'switch to backup generator' germaine to what they are trying to demonstrate now.
Well, except for the minor fact t
Re: (Score:2)
How is there any "AI" in it? (Score:2)
Given GPS, making a small, tough, slow, sailing boat that can sail autonomously to any point in the world would be pretty easy.
Keep the sail strong and small, make sure the boat is well balanced and self righting in a storm, fully sealed so it cannot sink, and a small solar system to keep the GPS powered. For a small fee you can get a Satellite SMS phone to announce its position along the way.
Such things actually have military value. Imagine a fleet of 1000 of these hovering off the Chinese coast, watchin
Re: (Score:2)
They already exist and are used in oceanography, check out Saildrone [saildrone.com]. There's also Wavegliders [liquid-robotics.com] that harvest wave action for propulsion, 2 of which made a Pacific crossing a few years ago (a diver in Hawaii did do some de-barnacle-ing as they passed).
Re: (Score:2)
IBM has a history of doing technology demonstrations, and they always just focus on the thing being demonstrated. They did not include 'physically moving the chess pieces' when they showed a computer that could beat a grand master. They did not include 'reply with synthesized speech' when they showed they could beat Jeopardy champions. Neither one of those things was germaine to what they were demonstrating. Nor is 'switch to backup generator' germaine to what they are trying to demonstrate now.
Then you and IBM are being obscenely shortsighted about this. When you're playing chess, the chances of "shit happens" to destroy the entire chance of playing or ending the game, are pretty much slim to none.
NOT the same thing when you're on a boat. "Shit happens" should be something you assume could happen every time. Therefore having and switching to a backup generator to avoid that, is a reasonable and logical conclusion.
They aren't even claiming to have made an autonomous shipping product. They are trying to find out if machine learning can enable a crossing of the ocean.
Yes, for the exact and specific purpose of making an autonomous shipping product.
historical note (Score:2)
I'll note that the Mayflower didn't have a single generator problem the entire trip . . .
For that matter, I don't think there was a single generator problem on any ship, worldwide, for a century in either direction . . .
hawk
AI research (Score:3)
...then trying and failing to do something that basically primitive humans and matching "technology" managed to accomplish 400 years ago.
Pro tip: If you want to get into AI research, pick a problem that computers are lousy at but humans do well, and think through the implications and how it might be accomplished, then write some POC code.
If you want to study AI *engineering*, then look into tensor flow, Google AI, Watson, or whatever. Get familiar with how it's called, what the limitations are, how it works, and that's a pretty good item to put on your resume to get hired. AI is all the buzz in companies nowadays.
Apropos of nothing, I strong
Re: (Score:2)
I strongly suspect humans and their matching technology *also* had hardware failures from time to time, even 400 years ago.
You kind of need to strongly suspect, there is a disturbing lack of first hand accounts of this happening.
Re:Nothing says AI superiority... (Score:4, Interesting)
How long does it take a human to learn how to sail across the atlantic though? 5 years of training and experience? 10? I know lifelong sailors who likely could not do it.
With an AI, once the first one is done - they can clone it 100,000 times on 100,000 vessels.
The benefit of AI is you train it once, then you can copy it forever. You can't do that with a human. Its about scaling the investment in training.
Re: (Score:2)
Primitive *society*. Not primitive *humans*.
Humans of 400 years ago are essentially the same as us now; they product of billions of years of evolution. An AI with the range of cognitive abilities enjoyed by a Neanderthal is, as yet, the stuff of science fiction.
Re: (Score:2)
Oh heck, an AI with capabilities approaching that of a chimpanzee is science fiction. If we could get an AI as smart and adaptable as an Australopithecine I'd be beyond impressed.
Re: (Score:2)
In under half a century an Artificial Intelligence has managed to become smarter than a cockroach. Not terribly impressive, until you consider that it took Ma Nature over 4 billion years to get that far.
No AI required. (Score:2)
Just a GPS, small strong boat. A high school project these days.
Re: Nothing says AI superiority... (Score:2)
Even Skynet needs to learn to take baby steps before it goes on to wipe out the human race.
Re: (Score:3)
You're always going to repeat this because you're unable to comprehend that your definition of 'artificial intelligence' is not the only (not even the most common) one possible. Demonstrates a rigidity of thought that doesn't reflect well on you.
obligatory.... (Score:2)
AI Ships (Score:1, Troll)
Re: (Score:3)
The right of way rules for ships are much more complicated that for cars and the stopping distances are often measured in miles. This seems like an even more difficult problem problem than autonomous cars to me.
Objects by your account are literally miles apart, and you're going to claim this is somehow more difficult than hundreds of cars traveling mere inches from each other at deadly rates of speed, and can create massive harm and death if something goes wrong in a split second?
Give me a break. We move massive ships with little more than 30-year old GPS technology today. AI is barely even justified to perform this function. This is like trying to convince an F1 driver that an AI-powered tricycle is "advanced
Re: (Score:2)
Oh brother. Have you even ever seen a boat or the ocean? Here's a clue: stick an average 16 year old in a car, and they can do what your autonomous car is doing. Stick an average 16 year old in a boat and say 'cross the ocean', and you pretty much have a guarantee of failure, probably disaster. Driving a car (at the level current 'autonomous cars' do it) is pretty much all stimulus/reaction, no intelligence involved at all.
When driving a car you have to worry about colliding with things a few hundred fe
Re: (Score:2)
Oh brother. Have you even ever seen a boat or the ocean? Here's a clue: stick an average 16 year old in a car, and they can do what your autonomous car is doing. Stick an average 16 year old in a boat and say 'cross the ocean', and you pretty much have a guarantee of failure, probably disaster. Driving a car (at the level current 'autonomous cars' do it) is pretty much all stimulus/reaction, no intelligence involved at all.
Let's stop comparing apples to machine guns here. Your average 16-year old may not even have a license to drive, and therefore could not even operate any car to cross even a city block safely, much less "cross the country" (which is far more equal to your "cross the ocean" argument).
When driving a car you have to worry about colliding with things a few hundred feet in front of you.
In reality, you have to worry about potentially hundreds of things colliding with you, in damn near every single second of driving, especially on a busy highway moving at speed, with many of those hazards within mere inches of
Re: (Score:2)
Re: AI Ships (Score:2)
Re: (Score:2)
Re: (Score:2)
Right of way on the water is really not complicated. Powered ships yield to unpowered (e.g. sail, oar) vessels. And then:
1. If another vessel is approaching you from the port — or left — side of your boat, you have the right of way and should maintain your speed and direction.
2. If a vessel is aiming to cross your path and they’re on your starboard — or right — side, they have the right of way. Alter your course so tha
Re: (Score:2)
how the systems will handle conditions like
... failed generators. You'd think that the AI would monitor the ship's systems and make the decision to turn back or head for an alternate port without "Researchers pushed out a software update, signaling for the ship to reverse course."
Re: (Score:2)
It's perfectly possible, and groups with a lot fewer resources than IBM have demonstrated it already. IBM just learned that if you're not carrying a mechanic with you, you need to beef up your maintenance in port.
Large shipwrecks are quite rare in the modern era and almost always involve human error. Frequently drunk human error. And the collision regulations are not only much simpler than something like driving rules, they're also well-specified and mostly consistent all over the world, except for a few, w
AI (Score:1)
Everything is "AI" now.
Smallest ever, believe me, super tiny (Score:1)
If it can't move, it's hardly "minor".
Re: (Score:1)
If they can drive back, why not finish instead? Something is off. Maybe they are drifting back and calling it "returning to port".
Re: (Score:2)
Yeah, something is off. Your brain. It failed 3 days into a month long trip. Who would be dumb enough to 'just continue' in that situation?
Re: (Score:1)
The devil's in the details of the alleged failure.
We told it ... (Score:2)
Not that Intelligent (Score:1)
They were monitoring the vessel from shore and issued a "software update" to tell it to return.
If it didn't decide to return on it's own, it's just remote control with some AI assist. A stunt.
Liquid Robotics (Score:1)
Wrong approach (Score:2)
Re: (Score:2)
Or 28,000 unpowered bath toys.
https://www.npr.org/2011/03/29... [npr.org]
Re: (Score:2)
Put the Raspberry Pi in bottles. that way you can claim AI.