Question Perdix Shearwater AI

[Curious_George]

Funny how even threads about Shearwater eventually shift to how bad the dreaded Suunto lockout is. Clearly that brand is renting space in some people’s heads. Ha!

@grassybreakfast you are more likely to go diving on the moon than convince anyone on here to change their minds about any of this.

I suggest there are two reasons why people violate their computers: 1) lack of experience/training or 2) unsafe in the way they dive or set up their computer. Beats the heck out of me why people on here are arguing or even concerned that the computer needs to "allow" these sad sack divers to go right back in the water.

My above over simplification excludes the situation where someone is dealing with an emergency - I don’t think we are discussing that as part of the reason why all these many people are getting an "arbitrary and punitive" lock out.

 

[Divin'Papaw]

I suggest there are two reasons why people violate their computers: 1) lack of experience/training or 2) unsafe in the way they dive or set up their computer. Beats the heck out of me why people on here are arguing or even concerned that the computer needs to "allow" these sad sack divers to go right back in the water.

My above over simplification excludes the situation where someone is dealing with an emergency - I don’t think we are discussing that as part of the reason why all these many people are getting an "arbitrary and punitive" lock out.

I don’t disagree with your summary of the primary causes of why people violate their computers. I would add one more. Poor design/usability of the computer’s UI.

In the context of this thread though, what’s your point?

Locking the offending diver out has not and never will stop them from continuing to dive. Instead it encourages them to resort to the foolish tricks mentioned in this post amongst others.

A computer is a tool. It’s job isn’t to decide when and if I can dive. Computers that take that baby-sitter approach are on a slippery slope that leads to stupid things like locking out, auto resetting the Nitrox mix to 50%, and other stupid ‘features’.

 

[grassybreakfast]

I don’t disagree with your summary of the primary causes of why people violate their computers. I would add one more. Poor design/usability of the computer’s UI.

In the context of this thread though, what’s your point?

Locking the offending diver out has not and never will stop them from continuing to dive. Instead it encourages them to resort to the foolish tricks mentioned in this post amongst others.

A computer is a tool. It’s job isn’t to decide when and if I can dive. Computers that take that baby-sitter approach are on a slippery slope that leads to stupid things like locking out, auto resetting the Nitrox mix to 50%, and other stupid ‘features’.

If they would resort to foolish tricks, why wouldn't they just dive on gauge mode? At the end of the day, there are many ways to die scuba diving, and no amount of computer design can stop a sufficiently determined fool.

I am of the opinion that computers should try to optimise for the safety of reasonable users, and that means not showing numbers when outside model design parameters, and showing a big prominent warning of something like "DO NOT DIVE. RISK OF DEATH." If people ignore that and continue diving, they are just going for the Darwin award and there's nothing we can do. Computers that lock out do allow people to continue diving in gauge mode, too, just like computers that don't, except they don't show arbitrary numbers that may not mean anything and give a false sense of security.

If a computer continues to provide guidance in cases where the model used essentially no longer applies, I would say that's more dangerous to the reasonable user as it encourages them to resort to the foolish practice of continuing to dive without either themselves or the computer knowing their nitrogen loading (I believe tables were also calculated with the no bubbles assumption - notice that you can't get a pressure group code beyond NDL).

I do agree that poor UI design is a significant factor. Unfortunately many people expect computers to be very small and very cheap, and designing a usable UI with those constraints is actually very difficult. Eg. you can't have a plain text warning if the screen doesn't have enough segments/pixels, which is the case for most entry level computers. That makes reading computer manuals crucial. Fortunately there are quite a few computers now with large and intuitive interfaces, and the Perdix is certainly not the only one.

 

[Divin'Papaw]

If they would resort to foolish tricks, why wouldn't they just dive on gauge mode? At the end of the day, there are many ways to die scuba diving, and no amount of computer design can stop a sufficiently determined fool.

Some models, once locked out, do not allow the computer to be placed into gauge mode. At least that used to be the case I believe. I don't know first hand. I've never bricked a computer personally.

I am of the opinion that computers should try to optimise for the safety of reasonable users, and that means not showing numbers when outside model design parameters, and showing a big prominent warning of something like "DO NOT DIVE. RISK OF DEATH." If people ignore that and continue diving, they are just going for the Darwin award and there's nothing we can do. Computers that lock out do allow people to continue diving in gauge mode, too, just like computers that don't, except they don't show arbitrary numbers that may not mean anything and give a false sense of security.

If a computer continues to provide guidance in cases where the model used essentially no longer applies, I would say that's more dangerous to the reasonable user as it encourages them to resort to the foolish practice of continuing to dive without either themselves or the computer knowing their nitrogen loading (I believe tables were also calculated with the no bubbles assumption - notice that you can't get a pressure group code beyond NDL).

The computer has all of your data and the underlying model still applies even if some of the guidance it provided (a missed or shortened deco stop for example) were not followed. After all, it KNOWS what deco you actually performed (how long and at what depths). It is therefore able to calculate your residual nitrogen (and possibly helium) per the model implemented in the computer. It therefore should still be able to accurately, within the limitations of any model, guide you on future dives.

To lock you out just because it feels you have sinned egregiously in some way (maybe you have, maybe you haven't) just leads to poorly trained or less safety conscious divers doing some of the things mentioned in this post to keep diving. Locking you out or displaying a "DO NOT DIVE. RISK OF DEATH." message as you suggest is ineffective simply because the very type of diver that is going to brick a computer is the same type that will ignore this message.

As I said in a prior post, your computer is a tool not a nanny. When the designer crosses that threshold and decides to now become your babysitter rather than your dive computer, all sorts of crazy design decisions are made. I prefer to dive a computer that is there for me to use as a tool, not there to tell me what I can or cannot do.

If you feel differently, no skin off my back. I'll stick with Shearwater and you can go dive any one of the other fine brands that enforce the kinds of nanny-rules you are looking for. There are plenty of them out there to choose from.

 

[grassybreakfast]

Some models, once locked out, do not allow the computer to be placed into gauge mode. At least that used to be the case I believe. I don't know first hand. I've never bricked a computer personally.

The computer has all of your data and the underlying model still applies even if some of the guidance it provided (a missed or shortened deco stop for example) were not followed. After all, it KNOWS what deco you actually performed (how long and at what depths). It is therefore able to calculated your residual nitrogen (and possibly helium) per the model implemented in the computer. It therefore should still be able to safely and accurately, within the limitations of any model, guide you on future dives.

To lock you out just because it feels you have sinned egregiously in some way (maybe you have, maybe you haven't) just leads to poorly trained or less safety conscious divers doing some of the things mentioned in this post to keep diving. Locking you out or displaying a "DO NOT DIVE. RISK OF DEATH." message as you suggest is ineffective simply because the very type of diver that is going to brick a computer is the same type that will ignore this message.

As I said in a prior post, your computer is a tool not a nanny. When the designer crosses that threshold and decides to now become your babysitter rather than your dive computer, all sorts of crazy design decisions are made. I prefer to dive a computer that is there for me to use as a tool, not there to tell me what I can or cannot do.

If you feel differently, no skin off my back. I'll stick with Shearwater and you can go dive any one of the other fine brands that enforce the kinds of nanny-rules you are looking for. There are plenty of them out there to choose from.

Please see my replies above on why it's not true that the computer can still safely and accurately model your nitrogen loading if you miss deco stops. But basically, the Buhlmann model (and as far as I can tell, all other models) assumes ongassing and degassing rates based on dissolved gases (no bubble formation). Once you get bubbles essentially all bets are off and it's no longer safe to use the model to make predictions.

I would rather have a computer that tells me it cannot make a confident prediction anymore, and switch to gauge mode (which is all it's still confident doing), rather than a computer that continues to spit out potentially bogus numbers. No output is better than wrong output.

In any case, it doesn't matter for me. Whether the computer locks me out or not, I am not going to dive for 24h when that happens. I won't dive in that situation without nitrogen loading guidance, and that's the case either way, whether the computer is still showing you numbers.

To lock you out just because it feels you have sinned egregiously in some way (maybe you have, maybe you haven't) just leads to poorly trained or less safety conscious divers doing some of the things mentioned in this post to keep diving. Locking you out or displaying a "DO NOT DIVE. RISK OF DEATH." message as you suggest is ineffective simply because the very type of diver that is going to brick a computer is the same type that will ignore this message.

I agree that if you are that kind of diver, a computer that doesn't lock you out is safer. Fortunately I am not, and I certainly hope no one reading this is either.

This has nothing to do with the computer judging/punishing/nannying you. This is about the computer recognising it's no longer within it's design parameters, and stop itself from giving you unreliable information.

 

[Divin'Papaw]

Please see my replies above on why it's not true that the computer can still safely and accurately model your nitrogen loading if you miss deco stops. But basically, the Buhlmann model (and as far as I can tell, all other models) assumes ongassing and degassing rates based on dissolved gases (no bubble formation). Once you get bubbles essentially all bets are off and it's no longer safe to use the model to make predictions.

I would rather have a computer that tells me it cannot make a confident prediction anymore, and switch to gauge mode (which is all it's still confident doing), rather than a computer that continues to spit out potentially bogus numbers. No output is better than wrong output.

In any case, it doesn't matter for me. Whether the computer locks me out or not, I am not going to dive for 24h when that happens. I won't dive in that situation without nitrogen loading guidance, and that's the case either way, whether the computer is still showing you numbers.


I agree that if you are that kind of diver, a computer that doesn't lock you out is safer. Fortunately I am not, and I certainly hope no one reading this is either.

This has nothing to do with the computer judging/punishing/nannying you. This is about the computer recognising it's no longer within it's design parameters, and stop itself from giving you unreliable information.

I think you’re commenting outside the extent of your understanding at this point. As current studies have shown, there are always bubbles. For the majority of divers, even recreational divers within NDL, bubbles are present. Your belief that ‘once bubbles are present all bets are off’ is inaccurate and demonstrates a simplistic understanding at best of the models and the latest thinking on what is going on in the body during decompression whether it be following a technical or a recreational dive.

At this point I’m bowing out of the conversation. We’ve once again gone WAY outside the bounds of the OPs question. My apologies for contributing to that.

 

[Belzelbub]

Computers that lock out do allow people to continue diving in gauge mode, too, just like computers that don't, except they don't show arbitrary numbers that may not mean anything and give a false sense of security.

That’s a big assumption you made here. I honestly doubt this is true. There are some very large companies that have dive computers that don’t lock out. I don’t believe for one minute that these are programmed to show arbitrary numbers. That would just be asking for a lawsuit.

Another way to look at it. Numbers displayed on every dive computer could be considered arbitrary. They aren’t an actual representation, but are instead a guess as to what is happening. The Shearwater Perdix manual makes this very clear.

(I believe tables were also calculated with the no bubbles assumption - notice that you can't get a pressure group code beyond NDL).

You mean the recreational tables published by many dive training agencies? There’s a reason for that. Key word is recreational, as most training agencies do not train for deco procedures in the recreational world. Maybe light deco, but that’s it.

For the record, I’m not against justified lockouts, or dire warnings. However, this should be based on actual violations of the algorithm limits. Adding conservatism to keep away from those limits should not be the basis for a lockout. There are better ways to handle these violations of conservatism limits that don’t actually even violate NDL.

 

[grassybreakfast]

I think you’re commenting outside the extent of your understanding at this point. As current studies have shown, there are always bubbles. For the majority of divers, even recreational divers within NDL, bubbles are present. You’re belief that ‘once bubbles are present all bets are off’ is inaccurate and demonstrates a simplistic understanding at best of the models and the latest thinking on what is going on in the body during decompression whether it be following a technical or a recreational dive.

There may always be bubbles, but bubbles essentially contribute to inaccuracy of the model, because the model doesn't take bubbles into account. After deco violation the risk of having significant bubbles is much higher, so the model is working with significantly more uncertainty.

I do agree that none of this really matters for rec diving. That's in my original reply to this thread before people started saying this is a very significant downside for rec divers.

That’s a big assumption you made here. I honestly doubt this is true. There are some very large companies that have dive computers that don’t lock out. I don’t believe for one minute that these are programmed to show arbitrary numbers. That would just be asking for a lawsuit.

I do wonder about that, too. Obviously the numbers aren't truly arbitrary - they are what the model spits out. We just don't know if they have done extensive verifications of the model in this regime. In any case, that doesn't seem to be published so it's essentially a proprietary algorithm at that point (or at least a model whose accuracy has not been publicly established)?

For the record, I’m not against justified lockouts, or dire warnings. However, this should be based on actual violations of the algorithm limits. Adding conservatism to keep away from those limits should not be the basis for a lockout. There are better ways to handle these violations of conservatism limits that don’t actually even violate NDL.

That gets a bit more philosophical I think. Computers know their models aren't perfect, and whatever bounds they apply it's always a risk tradeoff. Conservatism just allows you to adjust that tradeoff, and it's perfectly valid for a computer to say "given this is the tradeoff you want, you should stop diving now", where it may have allowed you to continue diving if you told the computer you had a higher risk tolerance, it would allow you to continue diving.

But also, the conservatism setting accounts for a whole lot of things. For example, you may set it higher if you feel you are at a higher risk of DCI due to dehydration, alcohol consumption, obesity, etc. In that case, that's essentially an input to the model, and the model may decide that you are at a higher risk given the setting, even if all other inputs are equal.

It's not like the computer has an accurate ground truth model so it always knows the exact tissue loading, and the conservatism is just a buffer zone on top of that.

 

[Belzelbub]

Conservatism just allows you to adjust that tradeoff, and it's perfectly valid for a computer to say "given this is the tradeoff you want, you should stop diving now", where it may have allowed you to continue diving if you told the computer you had a higher risk tolerance, it would allow you to continue diving.

That is one way to approach it, but not the only way. I see the conservatism settings as a Plan A approach. This is what I’m planning to do. However, having a Plan B is also wise. That’s another way to approach it. I might have my computer set conservatively, and planned to do a 5 minute stop based on those settings. However, maybe during the stop, I realize that the current is stronger than anticipated, so a direct ascent may be better. I can look at SurfGF and decide when it’s safe to ascend. In this case, I may have violated my Plan A, but Plan B gets me safely to the surface. With some computers, that’s a violation, and I’m locked out for 24 hours, despite not violating the actual NDL.

It's not like the computer has an accurate ground truth model so it always knows the exact tissue loading, and the conservatism is just a buffer zone on top of that.

Correct. It’s a buffer, but it’s a buffer that you should be able to cross without penalty, if a safe ascent is possible, and following the buffer can put the diver at risk.

 

[grassybreakfast]

That is one way to approach it, but not the only way. I see the conservatism settings as a Plan A approach. This is what I’m planning to do. However, having a Plan B is also wise. That’s another way to approach it. I might have my computer set conservatively, and planned to do a 5 minute stop based on those settings. However, maybe during the stop, I realize that the current is stronger than anticipated, so a direct ascent may be better. I can look at SurfGF and decide when it’s safe to ascend. In this case, I may have violated my Plan A, but Plan B gets me safely to the surface. With some computers, that’s a violation, and I’m locked out for 24 hours, despite not violating the actual NDL.

I agree that a computer that allows post-hoc adjustment of conservatism by reanalyzing old data and figuring out if there is a violation given the new conservatism setting would be best, but I imagine that's a lot of work and complexity (depending on the software architecture) to implement for the manufacturer, and that may be why they don't do it, given it would be a very niche feature, especially for recreational diving. Interesting concept though.

Correct. It’s a buffer, but it’s a buffer that you should be able to cross without penalty, if a safe ascent is possible, and following the buffer can put the diver at risk.

Should is a strong word... if the buffer is due to model uncertainty, how much of the buffer you should be allowed to cross really just depends on your risk tolerance.
Eg. if the model predicts 38 minutes NDL mean, with 95% confidence interval of >35 minutes, and a 99% confidence interval of >30 minutes, a computer may show you 35 minutes at a low conservatism setting, and 30 minutes at a high conservatism setting, the extra 5 minutes is 4% unsafe to cross (these numbers are obviously all pulled out of thin air).