Should Shearwater add Air Integration to its computers?

[uncfnp]

Hydrotac Stick-on Bifocal Lenses, 2.00 - Walmart.com (work great!)

or

BuildMask - SeaVisionUSA (with gauge reader lenses - also work great!)

but that would require me to admit I am over 40!

 

[undrwater]

Note to Shearwaterlease don't waste any time developing any hardware or software that I find unnecessary. Thank you.

 

[scubadada]

Who cares? Very useless thread.

The many people who have participated obviously care. Maybe you should stick to your Bio Filter Air Moisturizer and Bio Fins

 

[guyharrisonphoto]

Actually, Therac 25 (thanks Lorenzoid for a very interesting read--I had not see this before) illustrates my point here. To grossly simplify for those who will not google it, Therac 25 was a new technology electron beam machine offering two different and incompatible methods of delivering the energy. On prior machines, there was a hardware interlock to prevent the improper mode from being used. The Therac 25 replaced the hardware safety in favor of software security interlocks. It had no track record. It then had documented malfunctions where the wrong mode was engaged, repeated in a similar fashion. Analysis led to the discovery of the software fault.

These are not the facts with AI. AI is not new tech. AI, from the beginning, had software isolation from the dive computer deco functions. AI is used by many different manufactures in many different dive computers. AI has not been the documented cause of a single dive computer deco malfunction or failure. AI now has millions of dives (probably) as a track record. Interestingly, the AI "failure" of transmitter signal loss has occurred with some frequency, and even in "failure mode" the AI has never been shown to impair the dive computer's functioning in dive/deco mode.

I do view this as "hard data." Many years of use across different manufacturers and with with huge numbers of dives with no reports of interference with computer functioning even when the "AI side" drops or fails in active use cannot be deemed anything other than data showing non-interference with the "dive side." The absence of a failure report is evidence of success. Sure, it is not a "controlled" study which, by definition, would involve a much smaller number of computers, divers and dives, and so not even be as reliable. In reality, we have a much larger and broader context of dives than any controlled study could ever produce.

I have openly put forth the case that AI is proven as to its non-interference with diving computer deco functions. If there are facts out there to contradict, anyone can state them and I will stand corrected.

The purpose of my post is to suggest, respectfully, that aversion to AI as interfering with computer functioning from a tech diving perspective might now be laid to rest as an objective matter. I get tech divers not embracing unproven technology (hence the slow adoption of computers in the first place).

Whether it serves a practical benefit--different issue. For me, in the rec camp, it serves a crucial practical benefit (I described this in the other thread) and I would buy a Shearwater "PerdAIx" in a heartbeat.

 

[stuartv]

I admit the likelihood seems remote, but my only point is that just because such an oddity hasn't been reported by anyone to the knowledge of anyone here on SB doesn't mean it hasn't happened or might not happen. Some posters seem supremely confident there is no potential issue with this. I'm not that concerned--just pointing out that the likelihood is not zero.

I don't think it is that people think the likelihood is zero. I think it's that people find the probability to be on the same scale as the probability that the computer will crap out for any reason. Even a Petrel can crap out. But, it's probably more likely that a battery will spontaneously develop a bad cell and die or a battery compartment O-ring will leak or whatever than it is for AI to cause the deco calculations to be wrong.

We should dive prepared for our computer to die. Or to give bad results. We prepare by having a backup of some type and using our heads to recognize if the computer is giving us bad info. You get to the end of an hour at 130 feet and your computer is telling you you have a 5 minute deco stop at 20 feet? You recognize that "that ain't right" and fall back to another source for your deco plan.

In the extremely unlikely event that AI somehow causes your computer to calculate deco wrong, well, hopefully, your backup and brains allow you to catch it. But, really, it seems more likely that you computer would just die and for some other reason than it has AI. And if you can't deal with that, I'm thinking you probably shouldn't be doing technical dives.

---------- Post added January 11th, 2016 at 04:21 PM ----------

but that would require me to admit I am over 40!

Well, what are you going to say to the other tech divers who make fun of you for having AI transmitters on your sidemount cylinders?

 

[dmaziuk]

Actually, Therac 25 (thanks Lorenzoid for a very interesting read--I had not see this before) illustrates my point here. To grossly simplify for those who will not google it, Therac 25 was a new technology electron beam machine offering two different and incompatible methods of delivering the energy. On prior machines, there was a hardware interlock to prevent the improper mode from being used. The Therac 25 replaced the hardware safety in favor of software security interlocks. It had no track record. It then had documented malfunctions where the wrong mode was engaged, repeated in a similar fashion. Analysis led to the discovery of the software fault.

Not the full story. Software fault was in the assumption that the human operator will never be able to type into the input form faster than the software can read it and refresh the internal settings. The assumption was wrong. But only a couple of techs in different hospitals got that fast, so linking patient deaths to the specific nurses' shifts in several different hospitals in different cities took a while -- the deaths initially appeared random.

 

[pedro_rj]

Someone hit a very good point earlier.

"'Most tech divers know what's in their tank anyway"

This is so true. I was once told before a complex, task loaded cave dive to guess my pressures before I looked at my gauges. If I was more than 10bar (150psi) out, I probably shouldn't be trying to do that dive.

Why can't photographers do the same? It's also much less critical for you guys too!


Sent from my iPhone using Tapatalk

(also in reply to AJ)

You have divers who happen to take pictures and then you have people who dive to take pictures.

Not just messing with semantics here: when you dive to take pictures, it's a completely different mindset. Your focus goes to taking pictures/shooting video, so you don't want too many "distractions" such as looking away to check gauges, compass, timer etc. That's why having it all at a glance is such a big prize for most photographers.

(BTW that's also why I prefer to photograph in shallow depths ~10m, as this mindset very obviously increases risks considerably.)

As I said before, personally I don't really care if it will be Shearwater to tick all boxes. If it's perceived as an existential threat by the tech community, by all means leave AI out of the Perdix. We're all in the same boat here, just looking for the right tools for the job.

 

[dmaziuk]

My background is actually hardware--and a bit of RF to boot--not so much software, so my initial thoughts were more along the lines of signal interference due to having an RF or ultrasonic data communication system next to (and sharing a power supply with, etc.) the existing circuitry.

Quote oceanic manual: "may lose the signal within 3 ft of a running DPV or after a strobe flash".

On the software side there's various overflows: various resources from word size to amount of CPU cycles are finite and could be potentially exhausted, but realistically any modern hardware should have no problems running all of that while streaming HD pr0n. It does add an interesting state where your calculated deco time exceeds your calculated ATR -- what do you tell the user? -- but on a recreational dive that should be mostly hypothetical. Just like therac-25 operators could never possibly out-type the computer.

 

[Lorenzoid]

Actually, Therac 25 (thanks Lorenzoid for a very interesting read--I had not see this before) . . . .

I can't take credit for that. dmaziuk said it. I had to Google "Therac 25," too. Very interesting indeed.

These are not the facts with AI. AI is not new tech. AI, from the beginning, had software isolation from the dive computer deco functions. AI is used by many different manufactures in many different dive computers. . . .

Software isolation, perhaps, but what about hardware isolation? The potential for radiated or induced signal interference? Are all computers' hardware designs identical? Probably similar, but I can't say with certainty that adding AI to one computer's hardware will not affect its deco functions just because AI has been added to some other computer's hardware. Yes, in and of itself, I agree AI is "proven technology," and "millions of dives" have been done on AI computers. But integrating even the most proven technology with another system raises other considerations. You can't just slap AI onto an existing design. If there ever was a company I would trust could include AI without adversely affecting anything else, it would be Shearwater. I suspect they WOULD put in the testing time to avoid a Therac-25 scenario. But it isn't an insignificant effort. They might not be able to integrate it into the existing Petrel/Perdix design but rather might have to start from a new design that anticipates such integration from the beginning. Would it be worth it to them?--that is one of the themes running through this thread.

I don't think it is that people think the likelihood is zero. I think it's that people find the probability to be on the same scale as the probability that the computer will crap out for any reason. Even a Petrel can crap out. But, it's probably more likely that a battery will spontaneously develop a bad cell and die or a battery compartment O-ring will leak or whatever than it is for AI to cause the deco calculations to be wrong. . . .

I completely agree with everything you said. I don't know what is the "scale" of the probability of the battery compartment o-ring leaking--I have had it in my head that it is extremely low. I mean, a system in which an o-ring in a threaded cap seals a compartment against water intrusion at depth is highly reliable isn't it? That the battery compartment seal uses such a simple, time-tested mechanism is one of the reasons the Petrel appealed to me. I may be mistaken, but I have been under the impression that the possible failure modes, namely, water seeping past the o-ring, could not be more well understood and because it's so well understood the thing could readily be engineered to avoid just that failure. Are the possible failure modes resulting from integrating a wireless AI unit and a deco processor together onto a chip or other system just as well understood? I really have no idea how to rank the probability of the battery o-ring leaking against the probability of the AI unit causing trouble with the deco unit. I think the latter probability could be made low by careful engineering and testing, but it would require a bit more investigation than would be done with a battery compartment seal to reach the same level of confidence. Is it really on the same scale of probability as battery o-ring failure? I don't know. Others may think they know. Call me paranoid, that's all.

For brevity and because I wholeheartedly agree, I snipped the part of your post where you noted that we should not rely on our computers as our sole source of safety information. A failure of my computer to properly compute something would be no tragedy for me for just the reasons you mentioned, but what it would do is shake my confidence in the product as a whole.

 

[kelemvor]

My background is actually hardware--and a bit of RF to boot--not so much software, so my initial thoughts were more along the lines of signal interference due to having an RF or ultrasonic data communication system next to (and sharing a power supply with, etc.) the existing circuitry. Would they integrate it all together on an ASIC? Maybe there are isolation issues. I just don't know.

Those are problems that have already been solved by most other D.C. manufacturers. Shearwater can do it as well, if they have a mind to. Perhaps they could even do it better than the competition.

Maybe a manufacturer would rather direct their resources elsewhere.

Like any retail product that is created, Shearwater would have to try and determine if the market for hoseless AI dive computers is big enough to make it worthwhile. Every other DC manufacturer has answered yes to that question (I think). Lots of people buy computers with hoseless AI already. Shearwater has already shown they want in on the recreational diving market with their existing software updates.

Personally, I think a Shearwater computer with hoseless AI is an eventuality. Maybe it's already planned for that new smaller computer they just released (or a 2.0 rev).

---------- Post added January 11th, 2016 at 05:54 PM ----------

Maybe it's just my inexperience showing, but I know what my pPO2 (max) and CNS% are going to be before I get in the water. And those things aren't going to change because I got in and there was s stronger current than expected or I had an extra exciting few minutes for some reason.

Being able to quickly and conveniently check my tank pressure is a much higher priority to me than checking my pPO2 or CNS%. If I have to hit a button to page to look at some data, I would rather have to hit the button to page to pPO2 or CNS%.

It is ironic (to me) that some people think those things are so important to check mid-dive that they want them on the main screen but yet will argue that tank pressure shouldn't be on the computer at all. Shouldn't you also have a really good idea of what your pPO2 and CNS% are without having to check them during a dive?

I agree. I bet DAN or some other organization keeps statistics on how many accidents/incidents involve pP02, Nitrogen loading, and OOA/gas management...