Collisions at sea with autonomous ships

[BeauV]

Beau, I think the idea of "new" is somewhat grey with a neural net. "Far away from past experience" might be better.

And I have to write that for someone like me with a stats background, to call a system which takes a series of inputs to a useful set of results based on past examples other than a "statistic" is to try to sound smart purely through the use of fancy terms. (I forgive you.)

I recommend "Pattern Recognition and Neural Networks" by Brian Ripley.

I recommend Keith's recommendation Ripley is really good stuff.

I would also note that a challenge with neural networks is the ambiguous determination of success in a subjective situation. A game has a digital outcome. Success is objective. Driving to windward is subjective sister flying a passenger aircraft in the dynamic situation described in Keith's link. If best VMG is desired, that is one outcome. If good progress without breaking the boat is desired, success is different as it is if you are trying to rest a crew.

Part of "educating" a modern AI system is to teach it subjective outcomes.

Radical Agreement Here: to Larry's comments, a few boats have load-cells linked into the nav station so the Naviguesser can tell the boys they're pushing the boat too hard. Then, when the same Naviguesser goes flying through the boat, is knocked unconscious, and can't participate in the next race I'd rate that as a FAIL on the part of whatever person or system was trying to optimize overall performance.

[Ish]

What could possibly go wrong? http://www.huffingtonpost.com/entry/microsoft-tay-racist-tweets_us_56f3e678e4b04c4c37615502

[JoeP]

What could possibly go wrong? http://www.huffingtonpost.com/entry/microsoft-tay-racist-tweets_us_56f3e678e4b04c4c37615502

Really? Microsoft couldn't see that coming?

[BeauV]

ROFLMAO!!! Leave it to MSFT not to get this right.

[kdh]

So, as humans gain more experience it often becomes harder to make decisions because they are aware of more variables.

Does AI have the same issues?

Absolutely, but thanks to what is erroneously termed "Moore's Law" (which is actually only an observation) computing power has far outstripped the ability of experience to overwhelm it.

To bring in the classical idea, "the curse of dimensionality." Beau, you SiValley guys are a hoot!

http://en.wikipedia.org/wiki/Curse_of_dimensionality

[Tucky]

If humans weren't in the market, would Keith's machine have something to do better than?:-)

[kdh]

If humans weren't in the market, would Keith's machine have something to do better than?:-)

Interesting question, and a cloud over all who do what we do. No one knows about this, but in August 2007 there was an encounter between man and machine. Man wanted his money back, now, and machine "knew" that was usually an opportunity. Man watching machines in the short term were worried. Hugely. Some bailed. In the end the machines were right.

Disturbing, a model of the markets as those with machines systematically betting against human foibles is not far off, to me. You're right.

[LarryHoward]

If humans weren't in the market, would Keith's machine have something to do better than?:-)

Interesting question, and a cloud over all who do what we do. No one knows about this, but in August 2007 there was an encounter between man and machine. Man wanted his money back, now, and machine "knew" that was usually an opportunity. Man watching machines in the short term were worried. Hugely. Some bailed. In the end the machines were right.

Disturbing, a model of the markets as those with machines systematically betting against human foibles is not far off, to me. You're right.

Disturbing as some already feel electronic trading actually changes the market rather than operates within the market. Is it light speed market timing or market manipulation that is effectively milking cash off the top? At which point do the algorithms cease to predict human behavior based on external events and start driving human behavior. When does "Hal" decide that by creating a run on a sector, "he" can choose the bottom and generate huge returns? What happens when Hal and his buddies get into a hostile takeover battle without human intervention?

OTOH,

What happens when a ship in a storm loses connectivity? Or the network gets hijacked? Will Spectre put ships with toxic cargoes aground in pristine wilderness areas unless James Bond can clean out the virus? My VOIP, VPN and LAN at my satellite office has decided over the last 6 weeks to crash several times/day. Each provider swears their equipment and software isn't to blame and they haven't pushed any new configurations in the dark of the night. At the moment, none of my outside lines work and all folks get calling in is "number is out of service" . Not so good when you want customers to be able to reach you. Naturally, the trouble desk folks (who can only remote in when it's all working fine) can't find anything wrong so "it's not my stuff causing it."

I'm not worried because all of those smart folks in Silicon Valley say this doesn't happen to critical control systems like that ship holodeck.


Of course, on Good Friday in Lord Calvert's "Catholic Colony", there are not too many open offices. We are pretty lonely. Even the State and County governments are closed. Edit. While typing this, my internet connectivity dropped. Had to run and play IT Tech.

[BeauV]

If humans weren't in the market, would Keith's machine have something to do better than?:-)

Interesting question, and a cloud over all who do what we do. No one knows about this, but in August 2007 there was an encounter between man and machine. Man wanted his money back, now, and machine "knew" that was usually an opportunity. Man watching machines in the short term were worried. Hugely. Some bailed. In the end the machines were right.

Disturbing, a model of the markets as those with machines systematically betting against human foibles is not far off, to me. You're right.

Keith,

I'd like your opinion on when the machines start responding to each other rather than the "humans" in the market. Especially in the short term, meaning minutes, it seems that the machines are all talking to each other. Humans take far too long to respond. So aren't the models all fighting each other?

Beau

[kdh]

Larry, I apologize. I easily forget the common wisdom is "milking cash off the top." Let me just say first that the common wisdom is lacking.

As you write it's Good Friday so markets are closed. I have some time. I think the best government argument against free markets is Elizabeth Warren's:

For me the term high frequency trading seems wrong. You know this isn’t trading. Traders have good days and bad days. Some days they make good trades and they make lots of money and some days they have bad trades and they lose a lot of money. But high frequency traders have only good days.

One must ask about the "good old days" before high frequency trading. We had NYSE specialists and NASDAQ dealers with hugely privileged access to information. This is bad. Our current system is not perfect but it's better. It sounds bad, but closing your books at the end of the day with a fairly predictable gain has been going on for a long time. It's called being a dealer. Think of your car dealer.

Please remember that the purpose of capitalism is to let the public decide how to allocate capital--where money should be spent. It's an ugly system but the basic function is figuring out who's worthy and who isn't. It's challenging. But the monetary rewards for the knowledgeable can be great, so many like me who like a meritocratic system flock to the enterprise. I'm 5' 5" on a good day so the system is otherwise against me. The question to me is, do machines and those that are facile with them have a role in carrying out this function? It's clear to me the answer is yes, and my wealth speaks directly to that.

I loath the idea of being defensive. But our average holding period is 6 months. By that silly metric we're not skimming.

[BeauV]

Larry, you need a better network

As you know, there are networks that have been "up" for years and years; if by "network" you include all the redundancy and security of an entire system of networks. Of course, every computer design has a specific (and somewhat predictable) probability of failure. Some of the projects I've modeled even have an attempt to calculate the frequency of "bad actors".

But the question isn't: "Do automated systems fail?" The question is: "Do automated systems fail less than the alternative?" (meaning people most of the time or older automated systems).

What is the most difficult to determine is the social and societal response to failure. When a watch-keeper on a ship runs over a fishing boat or a skipper puts a tanker aground (think Exxon) we have social traditions which define the scope of blame. But, so far, we haven't really agreed to the limits of blame when a machine makes a similar mistake, and a lot of folks want to blame all machines for one machine's failure. You know we humans, we have to blame someone, even if it's our own damned fault.

[kdh]

Keith,

I'd like your opinion on when the machines start responding to each other rather than the "humans" in the market. Especially in the short term, meaning minutes, it seems that the machines are all talking to each other. Humans take far too long to respond. So aren't the models all fighting each other?

Beau

Great question. Think about it. Most modelers of the market do the obvious thing. We gather a bunch of data and make inferences about the characteristics of trades that are profitable. But we're all using the same data. In that sense we're all on the same side. We all compete against the data-deprived acting on Mackay's "popular delusions."

My old boss sat in a meeting with Peter Lynch ("One Up on Wall Street") and a bunch of other of my colleagues and asked, in 1985, about the idea of using computers to pick stocks. Most in the room laughed. Peter didn't, along with a couple others.

[LarryHoward]

Larry, you need a better network

Beau, Couldn't agree more.

Unlike urban utopia, there area a lot of places where 3G is spotty and fiber is a long time away. We have one provider of commercial bandwidth in the county and my facilities folks decided a VOIP provider from their Long Island neighborhood (I know a guy) was a better option than someone local to me because "this IT stuff is borderless" . Naturally, our IT guys own the servers but the facilities guy owns the phones because "they are a utility, right?" Believe me, that doesn't help at all, particularly when all of the small office connectivity runs through the VOIP hardware. My OOMA package at home with routers and NAPs that I engineered and installed myself is better that the junk I'm paying for at the office.

2 additional reboots in the last 3 hours.

[LarryHoward]

Larry, I apologize. I easily forget the common wisdom is "milking cash off the top." Let me just say first that the common wisdom is lacking.

As you write it's Good Friday so markets are closed. I have some time. Let me first say that I think the best government argument against free markets is Elizabeth Warren's:

For me the term high frequency trading seems wrong. You know this isn’t trading. Traders have good days and bad days. Some days they make good trades and they make lots of money and some days they have bad trades and they lose a lot of money. But high frequency traders have only good days.

One must ask about the "good old days" before high frequency trading. We had NYSE specialists and NASDAQ dealers with hugely privileged access to information. This is bad. Our current system is not perfect but it's better. It sounds bad, but closing your books at the end of the day with a fairly predictable gain has been going on for a long time. It's called being a dealer. Think of your car dealer.

Please remember that the purpose of capitalism is to let the public decide how to allocate capital--where money should be spent. It's an ugly system but the basic function is figuring out who's worthy and who isn't. It's challenging. But the monetary rewards for the knowledgeable can be great, so many like me who like a meritocratic system flock to the enterprise. I'm 5' 5" on a good day so the system is otherwise against me. The question to me is, do machines and those that are facile with them have a role in carrying out this function? It's clear to me the answer is yes, and my wealth speaks directly to that.

I loath the idea of being defensive. But our average holding period is 6 months. By that silly metric we're not skimming.

Keith,

Honest. Not accusing you of "milking", just pointing out that the increased capabilities of electronic trading combined with AI does tend to make the machine better than the man. Wondering if that can morph to driving the markets rather than predicting the direction of the markets and acting on those conclusions. Could make Bunky Hunt look like a piker.

Don't sweat being a "fun sized human". There are lots of spaces on boats where that is a real advantage (he says after a week of pulling new cables through dark recesses). I haven't noticed that it slows you down much. My 6'3" son with size 14.5 feet finds fitting into a car the first filter. He'd love a Lotus but is nowhere close to fitting one and each foot covers 2 pedals.

[kdh]

Keith,

Honest. Not accusing you of "milking", just pointing out that the increased capabilities of electronic trading combined with AI does tend to make the machine better than the man. Wondering if that can morph to driving the markets rather than predicting the direction of the markets and acting on those conclusions. Could make Bunky Hunt look like a piker.

Don't sweat being a "fun sized human". There are lots of spaces on boats where that is a real advantage (he says after a week of pulling new cables through dark recesses). I haven't noticed that it slows you down much. My 6'3" son with size 14.5 feet finds fitting into a car the first filter. He'd love a Lotus but is nowhere close to fitting one and each foot covers 2 pedals.

Funny.

Err, maybe "man with machine is better than man."

I worry about seeing over the steering wheel, not "fitting into a car."

[JoeP]

Seemed somewhat appropriate for this thread:

captain_speaking.png

Thanks to XKCD.

[kimbottles]

Keith,

Honest. Not accusing you of "milking", just pointing out that the increased capabilities of electronic trading combined with AI does tend to make the machine better than the man. Wondering if that can morph to driving the markets rather than predicting the direction of the markets and acting on those conclusions. Could make Bunky Hunt look like a piker.

Don't sweat being a "fun sized human". There are lots of spaces on boats where that is a real advantage (he says after a week of pulling new cables through dark recesses). I haven't noticed that it slows you down much. My 6'3" son with size 14.5 feet finds fitting into a car the first filter. He'd love a Lotus but is nowhere close to fitting one and each foot covers 2 pedals.

Funny.

Err, maybe "man with machine is better than man."

I worry about seeing over the steering wheel, not "fitting into a car."

Keith,
Some of the best bicycle racers in the world are 5'5" or there abouts. There are times while on a long climb I wish I were 5'5" and 120-130 pounds instead of 6'0" and 192 (this morning.)

All sizes have advantages and disadvantages.

One of my dear departed friends (Jerry Heron) was 5'5" and had head room in a lot of boats in which I had to duck to go below. He lorded that over me with a twinkle in his eye.

[LarryHoward]

Keith,

Honest. Not accusing you of "milking", just pointing out that the increased capabilities of electronic trading combined with AI does tend to make the machine better than the man. Wondering if that can morph to driving the markets rather than predicting the direction of the markets and acting on those conclusions. Could make Bunky Hunt look like a piker.

Don't sweat being a "fun sized human". There are lots of spaces on boats where that is a real advantage (he says after a week of pulling new cables through dark recesses). I haven't noticed that it slows you down much. My 6'3" son with size 14.5 feet finds fitting into a car the first filter. He'd love a Lotus but is nowhere close to fitting one and each foot covers 2 pedals.

Funny.

Err, maybe "man with machine is better than man."

I worry about seeing over the steering wheel, not "fitting into a car."

I'll buy the man with a machine statement but I don't day trade because there are a lot of "men without machines" better than me and with a machine I'm not even in the same game.

[kdh]

Where's that machine, uhh, steering wheellllll..............

[SemiSalt]

The are a significant number of loss of propulsion incidents in commercial ships every year. How is the computer going to handle those.

And are comfortable with a computer driven semi in the Rockies in white-out conditions?

[BeauV]

The are a significant number of loss of propulsion incidents in commercial ships every year. How is the computer going to handle those.

And are comfortable with a computer driven semi in the Rockies in white-out conditions?

There are very few propulsion loss incidents with long-haul trucks. It's really well below .00001 per mile, maybe even one hundredth of that. I'm guessing here, but I think the fact that ships haul around "engineers" etc... may make folks complacent about reliability. Either that or it's just bad design. Also, how are those "loss of propulsion incidents" handled now? Does the crew fix the engine? I doubt it. I'm guessing a tug is sent out to tow the broken ship in. The Computer would probably be pretty good at calling for a tug.

Having just been driven from Boulder to Denver in 60k winds with no trace of the edges of the road, it is very clear that NO ONE and NO MACHINE should drive in those conditions. The hundreds and hundreds of cars scattered along the sides of the road (some on their sides and some on their roofs) are confirmation of this. My Tesla will set off an alarm if it can't "See" and if the human doesn't take over the car will turn on its flashers and stop. There would have been dozens fewer accidents if the human idiot drivers had half the good sense of the Tesla autopilot software, and that was only a 16-hour storm.

To directly answer your question: Sure, I'd be perfectly happy to have semi-rigs driven by computers in the Rockies. Simply because they'd have to good sense to stop when they couldn't see. Unlike the stimulant hyped, drive fast for $$s, truckers I see on route 70.

[Tucky]

Thanks Keith, for taking the time to answer my question, which was serious behind the smiley.

[floating dutchman]

Yea, I don't thing sailors need to worry about autonomous ships for a while yet.
Most maintenance is done underway, ie Able Seaman chipping and painting and maintenance items buy engineers.
Those big engines don't get an overhaul every few years, a "unit" (ie piston rings and liner) will get replaced, next port next unit. the crew have all the parts on hand, they know the ship and how to do a unit. That's eight ports for an eight cylinder engine or more. You just cannot get ring in crew to do this kind of work, folk who don't know the ship will take to long, parts won't get ordered etc. having no crew on the ship just to tie the ship up for longer to do the maintenance makes no sense.
Computers to fix things on the run will not work. We had an issue this week where the crank case mist detector caused and engine to go into slow-down (just reduced load, they are fixed speed engines) turned out that the problem was actually just a crack in an air line to the sensor, easy fix for the crew on board, and we carry on, you just cannot automate this kind of thing and little breakdowns like this happen all the time.
We have had issues with the bilge and ballast valves, we just use them manually until we can fix them, How does a computer deal with that?
Who will take responsibility for the oily water separator? You can't just pump bilge water over the side, it has to be treated with the oily water separator then pumped overboard and the sludge gets taken ashore.
Who will replace that blown nav light lamp?
Ships are big and very complex pieces of machinery, they don't get stamped out by the millions like trucks and there are a lot of very experienced engineers who refuse to work on a ship that is less than 6 months old because they cannot be bothered with running a ship and dealing with all the crap that was not commissioned properly by the yard that wanted the ship out of the yard so they could get on with the next order. The brown paper bag industry must do a thriving trade near ship building yards.
I had the interesting job a few years ago working on the Lewek Petrel, the only work that ship had done was the delivery from Singapore to NZ. She could not hold position on DPS, I had the fun of fixing the lights on the deck that didn't even bungs in the holes you could run wires into the fittings for extra wires to other lights, Yea water proof my ass. The pump for the crane was immersed in oil, that was commissioned in the rain, yea that went well once the hyd pump mixed the water and oil together.
The "New to us" EX European passenger ship the company I work for owns recently had a problem where a seawater pump sensor came adrift, Well the bilge level switch failed also....
That caused a whole heap of damage before someone doing their rounds noticed the water above the floor plates. Without someone walking around the flooding would have gone unnoticed, even still a critical pump was destroyed resulting in a ship that cannot carry passengers until it was fixed.

You just have to have people on board a ship to keep it running, It's that simple.

Oh, maybe I should drink less before posting on Scantlings.

[JoeP]

Engine failure leads to grounding:

http://www.soundingsonline.com/dispatches/294842-video-split-and-spilling/?utm_source=soundings-enewsletter&utm_medium=email&utm_content=textlink&utm_campaign=enewsletter-033016

[BeauV]

Dutchman - you've illustrated once again why Scantlings is a Great Place. Thanks to Rick for providing the platform.

OK, I get it. I reverse course and recall all my comments because I clearly didn't really understand how bespoke and one-off ships are. Your point about each being different is key. Also, I hadn't thought about the fact that many of the PMs get done while underway.

Changing sides, eating hat.

Please don't stop drinking before posting, it'll make this place boring and then I'll feel compelled to stop drinking before posting.... that'll never work

Thanks again.

[kimbottles]

This is a VERY interesting place. Lots of talent, lots of diverse experience.

[SemiSalt]

You might ask if automated semis would be programmed to use runaway truck ramps in the case of brake failure. As it turns out, human drivers are not necessarily programmed to use them:

http://www.postindependent.com/article/ ... E/70601009

[BeauV]

I lost the brakes in my 1961 VW van full of amplifiers and junk on a steep hill. I was about 17. My old man had always said: "Hit something early! It only gets worse!" I dragged the starboard side against the railing on the road for about 100' to get the damn bus stopped. I've never forgotten how hard it was to intentionally hit the rail!

[Tucky]

You will appreciate this Beau. When racing at Indianapolis the danger point is after the second and fourth turns- the track has four distinct turns with two long straights and two short ones. It is not an oval though the drivers try to make it so. If they enter the first turn too fast they are ok until they can't make the second turn and the crashes are very hard. Back in the day Parnell Jones found himself with a mechanical issue that he knew would result in a bad turn two or four crash and deliberately drove into the wall immediately before the turn, thus doing a very high speed version of what you did. It was always cited as an example of skill and bravery at the track, at least back in the day when I followed racing there.

[BeauV]

Tucky, I can't imagine doing that at Indy racing speed! I was going about 35 in a beat up old van, my big worry was the bass amp hitting me in the back of the head. BV