Scantlings

Looks like the engineers and IT guys are in charge in the near future with this shipping model. Old school Bridge watch teams some times catch a glimpse of pending danger were the electronics fail to. Ocean sailors need to be ready for the future. AIS would be mandatory in a see and be seen situation.
https://www.youtube.com/watch?v=vg0A9Ve7SxE

where is the fun in that?

kimbottles wrote:where is the fun in that?

Kim I don't get this rush to automation with a growing population looking for work. Seafarers are the few people left that endure the rigours of the sea.

cap10ed wrote:

kimbottles wrote:where is the fun in that?

Kim I don't get this rush to automation with a growing population looking for work. Seafarers are the few people left that endure the rigours of the sea.

Cap't Ed,

From a ship owner's point of view (which I think is also almost the same as an owner of a fleet of long-haul semi trucks) the question isn't "Do we put folks out of work?" I think It's: "What is the highest quality service we can get?" It is pretty easy to demonstrate with Semi-trucks (and I think it would apply to ships) that the human operators make many more mistakes than the machines. As I'm sure you know, it only takes one error by a ship's captain to cause a multi-million dollar disaster; truck crashes are a bit less expensive. While computers may make mistakes, the automated cars have pretty well proven that the computers make far fewer mistakes than the humans do.

BV

It's actually "How cheaply can we deliver it with an acceptable loss rate?" Staff cost money - when machines can do it cheaper the people go.

I watched James May's show on cars last night and there was lots of autonomous car talk and one of the presenters was talking about how their experience had been that the people who were sitting in the car monitoring the autonomous operation were terrible at keeping track of things and properly taking over when needed. We witnessed that airline crash last year when landing when the pilots who had been doing everything automatically failed to properly handle the plane when they took over.

Ships are largely autonomous now, with the bridge people basically just keeping lookout, and we see how often they aren't good at that. I think too much assist but short of full automatic operation is a never never land that will prove to create lots of problems with possibly no improvement in safety- it is going to be hard to get all the way to truly autonomous.

I'll keep driving my Caterham and sailing without an autopilot- I like the engagement- its the reason why I do both.

One of the best UI guys I've ever known in the computer business gave a great talk once. He basically said that we have airplane and ship autopilots exactly wrong. The people and the machines should swap jobs.

People are terrible at watching something do the work. They get bored, doze off, wander off, etc....

Machines are great at watching people do the work and insuring they stay within the guard rails.

We should have people drive and machines complain when the people break the law or make a known mistake.

I agree Beau. We need a Bitchin' Betty in our cars to alert us to danger so we can take action. I think some of the lane departure and blind spot warnings are something like that.

By the way I was listening to NPR this weekend and they had a piece on the retirement of Boeing's Bitchin' Betty voice person:
http://www.npr.org/2016/03/19/471077856/retired-boeing-employee-stays-on-in-fighter-jet-command-voice

Another about her:
http://www.marinecorpstimes.com/story/military/2016/03/14/woman-behind-hornets-bitchin-betty-voice-retires/81767046/

Argh Matety! My open loop feedback algorhythm is all buggered up and those scurvy software dogs wouldn't know their arseholes from the bunghole in the apple barrel!

For those who didn't get my reference.

A young man is captured by pirates and is persuaded to join the crew rather than walk the plank. After a few weeks at sea the captain speaks to the man and asks him how he is getting on. The man replies that on the whole he is enjoying things - the rum-soaked drinking binges, the plundering, etc - but there was one thing missing.

"What's that?" asks the captain.
"Well, there are no women" replies the man.

"Arrr" says the captain "Follow me!" The man follows the captain to what appears to be a barrel, on top of the barrel stands a coconut with a face drawn on and a few strands of wispy straw for hair. On the barrel is a crude outline of a woman's body and between the legs is a bung hole. "We calls her Carmen," says the captain, "and you may take her as you will". The man explains that he was unlikely to make use of her and goes on his way.

However, as the months go by with no respite, Carmen appears more and more attractive to the young man. Finally he can resist her no longer and the man has his wicked way with Carmen the rum barrel. To his amazement the experience is far more satisfying than he could ever have imagined!

The next day the captain greets him again. "How did you get on with Carmen then, lad?" he asks eagerly. The man replies "Rather better than I thought... actually, it was rather good!"

"Good," says the captain, a great beaming smile splitting his black-bearded face. "It's your turn in the barrel tomorrow!"

cap10ed wrote:Looks like the engineers and IT guys are in charge in the near future with this shipping model. Old school Bridge watch teams some times catch a glimpse of pending danger were the electronics fail to. Ocean sailors need to be ready for the future. AIS would be mandatory in a see and be seen situation.
https://www.youtube.com/watch?v=vg0A9Ve7SxE

Shouldn't that guy in the youtube clip be wearing a hoodie?

Isn't that just moving the watch from the ship's bridge to land, just creating a virtual bridge? The seafarers get desk jobs.

One of the moms at Adele's school is an astronomer. She took the class and a bunch of parents to MIT's telescope installation that's far enough W of Cambridge to escape the city's light pollution.

It was cold there on top of the mountain in the boonies. We learned that after freezing their asses off the MIT students configured the telescopes to send the data to Cambridge so they could work in their warm dorm rooms, and they don't physically look through the telescopes any more.

If the exercise is cost savings than I think they missed the boat! They have lowered the wages so low for ship personnel, that the cost of the high tech shore installation might be more rather than less than crew?

cap10ed wrote:If the exercise is cost savings than I think they missed the boat! They have lowered the wages so low for ship personnel, that the cost of the high tech shore installation might be more rather than less than crew?

I agree. Though I suppose if there's a single shore installation managing hundreds of boats it could work. Sort of like air traffic controllers with no pilots to talk to.

kdh wrote:

cap10ed wrote:If the exercise is cost savings than I think they missed the boat! They have lowered the wages so low for ship personnel, that the cost of the high tech shore installation might be more rather than less than crew?

I agree. Though I suppose if there's a single shore installation managing hundreds of boats it could work. Sort of like air traffic controllers with no pilots to talk to.

Generally, and I really do be "generally", the "big win" in automation happens when one modifies the target behavour as well as automates the process.

Consider ATMs and bank customers. Sure, a bunch of the steps in the processes were automated and bank tellers were replaced, but that only happened because the customers were willing (in this case eager) to change how they executed the process. Initially, it was a "cash machine", then after 12 long years someone actually deposited cash into an ATM. Finally the thing was used to accept checks, pay credit card bills, pay other bills, and then POOF the smart-phone altered behaviour so radically that we deposit checks with a photo. No one uses cash (Ya, I know over generalisation) they buy everything with Apple Pay from their phone/watch. The point being that the introduction of automation combined with a re-training of the humans altered the entire set-up.

Now consider a ship. If one can build the telemetry, controls, and everything you need to run a ship from a remote control point. (Even if that point is simply the bridge of the ship initially.) In some sense you've altered the "customer". We already see early signs of all this. Ships are fly-by-wire now. The video cameras we see on large ships and yachts actually have better visual acuity than a person. This is particularly true with night vision and automated motion detection. Moreover, these "visual systems" with a computer watching don't get bored, watch TV, or go to sleep on watch. The rest of the ship is equally subject to the advance of automation. Once the telemetry is sent someplace: to the bridge, to an individual ashore, or to a concentrated air-traffic control system, then computer automation will eventually do two things. First, it will be cheaper, but much more importantly it will be better.

Just as Google's self-driving cars have much better safety records than humans, self-sailing-ships will be safer and better at moving cargo. It really isn't a question of "if" it's only a question of "when".

After all, even though John Henry is one of the biggest All Time Heroes of my childhood. The steam hammer eventually won.

BeauV wrote:

kdh wrote:

cap10ed wrote:If the exercise is cost savings than I think they missed the boat! They have lowered the wages so low for ship personnel, that the cost of the high tech shore installation might be more rather than less than crew?

I agree. Though I suppose if there's a single shore installation managing hundreds of boats it could work. Sort of like air traffic controllers with no pilots to talk to.

Generally, and I really do be "generally", the "big win" in automation happens when one modifies the target behavour as well as automates the process.

Consider ATMs and bank customers. Sure, a bunch of the steps in the processes were automated and bank tellers were replaced, but that only happened because the customers were willing (in this case eager) to change how they executed the process. Initially, it was a "cash machine", then after 12 long years someone actually deposited cash into an ATM. Finally the thing was used to accept checks, pay credit card bills, pay other bills, and then POOF the smart-phone altered behaviour so radically that we deposit checks with a photo. No one uses cash (Ya, I know over generalisation) they buy everything with Apple Pay from their phone/watch. The point being that the introduction of automation combined with a re-training of the humans altered the entire set-up.

Now consider a ship. If one can build the telemetry, controls, and everything you need to run a ship from a remote control point. (Even if that point is simply the bridge of the ship initially.) In some sense you've altered the "customer". We already see early signs of all this. Ships are fly-by-wire now. The video cameras we see on large ships and yachts actually have better visual acuity than a person. This is particularly true with night vision and automated motion detection. Moreover, these "visual systems" with a computer watching don't get bored, watch TV, or go to sleep on watch. The rest of the ship is equally subject to the advance of automation. Once the telemetry is sent someplace: to the bridge, to an individual ashore, or to a concentrated air-traffic control system, then computer automation will eventually do two things. First, it will be cheaper, but much more importantly it will be better.

Just as Google's self-driving cars have much better safety records than humans, self-sailing-ships will be safer and better at moving cargo. It really isn't a question of "if" it's only a question of "when".

After all, even though John Henry is one of the biggest All Time Heroes of my childhood. The steam hammer eventually won.

I'm not so certain it is coming that fast. For 20 or more years, we have had the ability to fly autonomous aircraft and commercial aircraft today have 99% of the equipment and systems required to operate pilotless in all weather. I don't see a rush to put autonomous 787s with 300 passengers in the sky although a computer is not likely to pull a GermanWings descent.

For "UAVs" we had to come up with a rule set as to where they could go and what they can do without a person in the loop and the loss rate is still higher than manned aircraft. For ships, it will be all about the bandwidth and its limited. You either need a huge pipe or an intelligent system onboard that collects data, assimilated it into information and maybe further into intelligence. I've heard for 30 years how we are just a shot time away from "essentially unlimited bandwidth" and each time we find another compression algorithm or launch another sattelite, we gobble that up and need more. I'm building systems today for a UAS where the tolerance for signal loss is pretty close to the limits of what physics can provide wit no margin for a bad connection, environmental deterioration or unusual atmospherics. It will on,y work is nothing goes wrong, goes wrong,goes wrong.....

Add in the nature of liability. Who is at fault when an autonomous ship runs over a fishing boat? When a self driving car runs over a kid on a trike or has to decide between hitting a car or a person? Where is that unflinching and unquestioned responsibility? Automatic systems are great. Until they aren't.

I ran a company developing and deploying UAVs. Beau is correct. Changing the target behavior is probably more important than the tech. There is a huge gap between what is possible and what is allowed. Both have to move forward.

In the meantime, who will sign up for the first unpiloted flight from SFO to Dubai?

Larry, I hear ya on all counts. But ships run over fishing boats ALL THE FREEKING TIME! I've personally been hit twice and I only have about 35,000 miles at sea.

So, the only actual difference is that it's considered "ok" somehow when the watch on the bridge hits a boat. I guess we think "they were doing the best they could." It's trivially easy to prove that a ship will avoid more small boats (and other stuff like whales and containers) with a really good vision system on the bow than it will with a sleeping person on the bridge. But, I don't doubt that folks will gleefully blame the computer for running down a fishing boat and let the crew that does the same thing off with a stern warning (even if you ever hear about them).

I don't want to dive into too much tech stuff, but the solution to the bandwidth problem has almost always been local processing. The AI we're talking about for 99% of running a ship would be done aboard. Unlike something airborne a ship isn't all that weight sensitive and the computers to do the job could easily fit in one rack today. In a few years they'll fit in something a lot smaller and lighter than one crewman and his food supply. You're the unmanned flight guy, but isn't most of the bandwidth because the remote pilot wants to see what's going on? Ships have the natural advantage of moving quite slowly, so real-time has a very different meaning.

Every time I see the President and his team literally watching a Navy Seal operation over real-time video my first reaction is - what a bunch of bozos. They're spending zillions of dollars to be voyeurs simply so they can feel important. What's the Pres. going to do - "Hey look out for that bad guy behind the couch.!" Geeesh. IMHO they should let the Seal do what he/she is trained to do, they should wait until the answer comes back, they don't need to see anything as they're not contributing anything. But, what do I know?

BeauV wrote:Larry, I hear ya on all counts. But ships run over fishing boats ALL THE FREEKING TIME! I've personally been hit twice and I only have about 35,000 miles at sea.

So, the only actual difference is that it's considered "ok" somehow when the watch on the bridge hits a boat. I guess we think "they were doing the best they could." It's trivially easy to prove that a ship will avoid more small boats (and other stuff like whales and containers) with a really good vision system on the bow than it will with a sleeping person on the bridge. But, I don't doubt that folks will gleefully blame the computer for running down a fishing boat and let the crew that does the same thing off with a stern warning (even if you ever hear about them).

I don't want to dive into too much tech stuff, but the solution to the bandwidth problem has almost always been local processing. The AI we're talking about for 99% of running a ship would be done aboard. Unlike something airborne a ship isn't all that weight sensitive and the computers to do the job could easily fit in one rack today. In a few years they'll fit in something a lot smaller and lighter than one crewman and his food supply. You're the unmanned flight guy, but isn't most of the bandwidth because the remote pilot wants to see what's going on? Ships have the natural advantage of moving quite slowly, so real-time has a very different meaning.

Every time I see the President and his team literally watching a Navy Seal operation over real-time video my first reaction is - what a bunch of bozos. They're spending zillions of dollars to be voyeurs simply so they can feel important. What's the Pres. going to do - "Hey look out for that bad guy behind the couch.!" Geeesh. IMHO they should let the Seal do what he/she is trained to do, they should wait until the answer comes back, they don't need to see anything as they're not contributing anything. But, what do I know?

Worst thing we ever did was to put huge "holodeck" inspired walls at the COCOMs HQs and link in the Pentagon and White House. Been there and had a system running in a corner of the screen during a lot of the Afghan ops. 4 star was "in the chair" and moved our meeting from his office to "the room." What it does do is give the NCA (National Command Authority) real time SA when a decision is required but by the same token it has lead to specific, at the moment clearance being required for a lot of situations you would expect a squad leader or company commander to deal with. As I recall, the first "on the spot" white house involvement was the Iranian hostage mission when Carter called it off after the C-130/CH-53 accident. I would not want to be a young 1st Lt on a mission and having to answer questions from the President while I'm being shot at. Now generally that doesn't happen and from what I have seen, the observers are well behaved but they don't have to be. I'd worry about a certain Presidential Candidate being online. He might fire the on scene commander.

Agree the best solution is local processing and flexible bandwidth to accommodate "deep dives". I'd still worry about how much authority you give the onboard controllers and redundancy in all critical systems. Most of our UAS losses were "dumb stuff" and mechanical systems don't always fail "by the book". Worst case is when the AI runs out of look up tables. Trained grey matter tends to do better in those cases.

So, do we sign you up for that SFO-Dubai flight?

Hey, I'm ALL IN as soon as we've got commercial air travel that is as good as the Google cars. That said, I was crazy enough to drive the Tesla over the pass on Highway 17 using Auto-Pilot on the way home. (It worked perfectly)

Here's an interesting factoid about delegation and efficency.

At the height of the British Empire there were 190 people working in the British Foreign Office in London and they ran an empire that include Canada, India, etc.. etc.... Today there are over 14,000 people in the UK Foreign Office and they run an empire of .... ???

The Viceroy of India RAN India because other than a letter every few months, which was totally out of date, London couldn't meddle. The British Empire was far better run by the Viceroys than it ever was by London. The enemy of an effective ground operation is some clown at HQ telling you how to do your job, rather than delegating to someone who knows what they're doing.

I share your fears of the current crop of candidates trying to "run" a battlefield situation. The only one I'd want watching me is Clinton. While there's a lot I don't like about her, she's clearly the most battle hardened candidate of a bad lot.

kdh wrote:http://www.askthepilot.com/questionanswers/automation-myths/

As a former fighter pilot, advanced flight instructor, aviation training technology lead for Naval Aviation and UAV guy, I'd say he is very invested in how he wants it to be. I can drive my boat upwind in the snot better than my autopilot - for 30 minutes. Mostly because I bring an intuitive understanding of what the boat will experience on that next wave or set. The Autopilot only reacts to what is is seeing at that moment with just a little learning. Once I get tired and lose the edge, the AP will beat my performance 9 times out of 10. Most of the functions he is so proud of are "Aircraft Commander" functions and have little to do with flying the aircraft. Certainly, there would have to be changes in ATC, someone would have to be "in charge", etc.

At least 10 years ago, Gulfstream demonstrated a coupled flight director where a pilot lined up on the runway and started the takeoff roll. As I recall, he had to manually raise the landing gear and then hit "auto". The aircraft flew a complex mission profile at various altitudes for 2 hours, returned to the original airport, performed an approach, signaled the pilot to lower the gear and then automatically landed the aircraft. The pilot took over once on the ground and performed the landing roll out and taxied back to the gate. 10 years ago. In multiple countries, folks are working on integrating unmanned aircraft into the existing ATC systems.

I'll leave it with most accidents are pilot error and in a number of cases, it's from taking control away from the automatic systems or ignoring/misinterpreting malfunction indications. It's the decision making under uncertainty where humans are far superior to machines. The computer will arguable make the right decision within the rule set it has been given. "Sully" saved a bunch of people by digesting several thousand different inputs and selecting the one set of actions that put him safely down in the river. He was a safety zealot and worked hard at playing the "what if" game. An autopilot would have a set of rules for "loss of power on climb" and a pretty short decision matrix that probably would not include "land in the nearest river". Score 1 for humans. I can rattle of several events where humans overriding the aircraft caused loss of the aircraft and all passengers.

I recall, though I could be wrong, that many accidents have been caused by pilots pulling back on the stick rather than pushing forward when dangerously close to stalling.

kdh wrote:I recall, though I could be wrong, that many accidents have been caused by pilots pulling back on the stick rather than pushing forward when dangerously close to stalling.

Pulling away from death (the ground) is a powerful instinct and in many cases, the wrong thing.

Different companies have different cultures and approaches. Boeing tends to not set up their airplanes for "auto everything". AIRBUS tends to use the pilot as a guide and the system actually flies the airplane (more so than other FBW aircraft). In a number of accidents, Airbus aircraft have crashed because the pilot overrode the computer and virtually tried to crash the airplane by doing the wrong thing. AF-447 is an example (Pilot's last words were reported to be "F*** We're dead") and AA 587 where the proper procedure to a turbulence induced yaw excursion was to allow the Computer "Yaw Stab" action to dampen it and the pilot was known to use rudders to compensate despite warnings against it. COMbination of teh pilot;s actions and the computer's actions broke the vertical stabilizer off.

Alternatively, Asiana 214 at SFO was a Boeing that let the pilot calmly fly the aircraft into the ground when he misused an autopilot mode and failed to engage autothrottles during final approach. Very broadly speaking, an Airbus would have been screaming warnings at him as the aircraft decelerated.

Idiots are so inventive that it;s hard to idiot proof a system.

Arguable, landing on an aircraft carrier in big seas at night is one of the hardest flying feats to accomplish. Now an unmanned aircraft can do it more reliably and with more consistent results than a pilot can. Computers can nail the "monkey skills". The hard part is the judgement and reaction to emergencies, off nominal conditions or response to external events. There, the compute can process a million options quickly and pick the response it has been told is correct. A human will consider multiple options and pick one that is logically and intuitively a good decision and then continue to reevaluate. One's an excellent decision tree and with AI, pretty damn good. The other has a will to live.

Unlike Beau, I'd still like to have a pilot up front- just in case - even though I understand and endorse the tech that makes the pilot optional.

Larry and I are on exactly the same page here.

In my dark foggy past I wrote AI software that was supposed to "support battlefield decisions". As a result I ended up as CEO of a company called Qunitus which was popularizing a programming language called Prolog. The basic theory was an "expert" would write down a set of logic rules (it really did look like the stuff a logician would write down) and the "system" would build a decision-making software product based upon all those rules. It utterly and completely failed at really complex systems, but did a great job of building automobiles. Indeed, folks still use Prolog to write down all sorts of rule-based systems and it's working great. They key is that someone could actually foresee all (or the vast majority) of situations and the implications of taking a set of actions based on those situations. Thus, Larry's statement that landing a commercial airliner in the river wouldn't have been on anyone's list of rules - at least the first time it came up. Rule-based systems are very very good at retaining gigantic lists of the best-practices of a group of experts, and ensuring that the same mistake isn't made twice.

Modern IA, like the Google program that just became the world's best Go player, don't work on the basis of rules. There has been a gigantic breakthrough in how to program AI systems which the vast majority of folks don't understand. There's no reason they should, it doesn't impact their lives. It's called neural nets and is quite loosely based on our current theory of how the human mind works. In this model, the computer "learns" from experience and combines the "never forgets" aspect of rule-based systems with the ability to recognize better solutions without having been taught them. Nice reading on this here if you're interested: https://en.wikipedia.org/wiki/Artificial_neural_network

To use Larry's example of the Auto Pilot sailing upwind, what we humans do is to integrate what we see of the waves coming at us, with what we feel from the boat's movements, with what we hear from crew (saying: "Puff in 3, 2, 1, now), etc... from all those inputs plus millions of waves that we've tried to sail over well, we decide what to do. This is pretty much what nural nets do when the software is "learning" a topic. In the case of Go, the computer can play against itself to get better; which is what the Google program has been doing for months. A nural net software solution to sailing up-wind would include a visual system to "see" the waves, includeing cameras located at different places on the boat (like in the rigging) which would allow it to see things that the helmsman can't. It would also include cameras that see in the dark and perhaps in other spectra other than our visual light range if that was useful. It would include gyro inputs (the way modern auto pilots do) to "feel" what the waves where doing to the boat. It would include other sensors, for example wind instruments at multiple point along the rig to be able to set the twist of the sails to accurately match the different wind speeds and direcitons at different altitudes. Equipped with all these sensors, the system would then take the boat sailing, for days and days and days. Each sail it would get better, and unlike me it wouldn't ever forget a wave. Every single wave, and the results of what the system did going over it, would be remembered and integrated into the best way to go over the next wave to come along. This last bit is the really important part. There is no magic here, there is just massive and permenent accumulation of experience and knowledge.

Without doubt, in a "new" situation comptuers are going to loose to humans. The question is "How many new siutations does the system encounter?" and then there's the simple measure of: "Does the degraded performance during new situation destroy the advantage the system has in 'normal' situations?"

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?

As I write a bunch of hardware buys and sells stuff with $2b of investment capital. We used to pick trades directly, override model recommendations based on manual analysis, review trade lists before routing them to the world's stock exchanges.

We stopped all that--too prone to error. If it can be automated we automate it.

All that's left is improving our core competency, our knowledge of how markets work and how people react to information. We automate acting on that knowledge.

Sometimes we automate some of the automation.

Somehow we still have stuff to do as we wait for Kurzweil's singularity.

US market just closed. We get a bit of a break before Asia opens and we start all over again.

Olaf Hart wrote: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.

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.

kdh wrote: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.

BeauV wrote:

kdh wrote: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.

kdh wrote:As I write a bunch of hardware buys and sells stuff with $2b of investment capital. We used to pick trades directly, override model recommendations based on manual analysis, review trade lists before routing them to the world's stock exchanges.

We stopped all that--too prone to error. If it can be automated we automate it.

All that's left is improving our core competency, our knowledge of how markets work and how people react to information. We automate acting on that knowledge.

Sometimes we automate some of the automation.

Somehow we still have stuff to do as we wait for Kurzweil's singularity.

US market just closed. We get a bit of a break before Asia opens and we start all over again.

Which just confirms my belief there is no place for mere humans in the market today.