As I open up the writing window, I have no idea what's going to come out. I'll start by filling in the Of the Day suspects. To encourage you to keep scrolling, Celina! Oh, and it got long. Get a drink before you dive in.
Once again, the great 10 day festival is underway here. Once again, I have no plans to attend. Yes, 45 years in Calgary, never gone to the parade. The only Stampede event for me is our community association stampede breakfast. That turned out to be a big success. I haven't heard how many people attended, but my guess is probably about 1000. At one point the lineup was out the parking lot and down the sidewalk. There are lots of photos. I'd thought of signing up for a photo tour with a buddy taking place the following day, but I'm happy I didn't. That was a lot of editing.
Instead I racked and stabilized a batch of wine, and started the last kit for this year. That got a little exciting because the small bag of raison skins kind of got away from me. Fortunately the mess was contained and almost all of it ended up where it belonged. Who knew making wine could be so exciting?
That will be the 124th wine kit for me since I started on June 9, 2002 or so. (Or more properly, 20020609) That comes out to about 3600 bottles of wine. I'm not sure how many bottles are currently on the rack. Anybody that really wants to know can come over and count. The first one can even pick a bottle to take home. I'll bottle in 6 to 8 weeks, and then I'll be able to set up the dark room again. I've got some negatives cued up, and should produce more. Maybe during the planned Friday ramble.
A beaver, just to amuse you. I see lots of evidence of beavers in Fish Creek, but have never actually seen one in the wild. During the photo conference in Pincher Creek, I was out one evening and saw one just downstream of the Oldman River dam.
I was chatting with a buddy at the breakfast. Both of us were musing about 2025 being half over already, and how quickly it's happened. The COVID pandemic was at its height FIVE YEARS AGO. Five years used to be a really long time, and it's gone by in a blink. I'd just started a work contract, so for a year I was mostly head down and swearing at XL, cursing the person who thought it was a good idea to treat rows of data as a matrix.
Which, and plus a thing I was just reading that suggests a rosy view that AI will solve all our problems is leading me to a bit of a rant. (Last chance to go get a drink.) A friend of mine once said he and I had hit the sweet spot for software development and solutions. Our tools were evolving, but we were working on the easy to medium hard problems.
I got hired for my most lucrative contract precisely because they didn't want a piece of black box software that took in data, manipulated it, and spat out transformed data. They wanted a human who understood the data; one that could break it into the appropriate chunks for effective and efficient transformation and migration, but most importantly, document and explain what had been done.
Yes, they spent more money on me than buying a black box would cost up front, but considered it worth it because they didn't trust the black box results, and had no idea what the consequences for the future would be. They quite rightly feared data integrity integrity issues leading to regulatory failures at the least, and catastrophic vessel failures due to missed inspection and maintenance. Plus I think that boss loved that her mind and my mind worked alike in how we could describe chunks of data from different related databases, compare them, develop lists of matching, in A but not B, in B but not A, plus other related differences, create work lists to resolve the various issues and then actually write SQL to fix the issues.
The data relationships were complex, there were existing data integrity issues to be defused, and some of the people involved didn't understand their own processes. Plus they hadn't known there was a combination of a database "feature" and their update process that was guaranteed to produce incorrect results. They didn't believe me till I showed them the most recent sequence of events through a series of before, during, and after, data snapshots.
As an aside, a software joke. Q: What is a software feature? A: A bug with seniority.
I can still remember the look on the face of the person who was the main data entry person when I asked her to a meeting, and we started talking about the problems to be resolved before migration. I'd been there a week or so, and she couldn't understand how I knew all this stuff. It took a while for her to get the power of database read and write access along with SQL and an understanding of table relationships. She didn't like me till she got told (with a capital T) to fix a particular problem and realized it would take her about a week of doing nothing else assuming she didn't make any mistakes along the way. She asked if I had any advice to do it quicker, and nearly had a melt down when I said it would take me about a half hour.
The black boxes have only got bigger and blacker and the problems bigger and harder. The gurus assure the gullible that the process is perfect. Data would go in one end, and come out with perfect integrity and ready for importing into the target database. Or it would be analyzed and a solution emerge. Some of them have never actually heard of GIGO, which is the polite way of phrasing it.
At one of my jobs they turned me loose to find problems with the data. That was a lot of fun. Just fixing the ID numbers was a major task. How hard could it be to capture a number like A0123456? You'd be surprised. Part of the trick is looking at the extremes of data. I'd write SQL to show me vessels with the largest MAWP, and found the numbers ranged from unbelievable to the heart of the sun. Vessels that were kilometres long and weren't pipelines, or so tiny an ant couldn't get into them. Those at least were easy to fix. I was told that no data was better than bad data, and if people saw a blank they'd fill it in.
It's sort of like many of the canned reports that come with a database. They sound good but don't really tell you anything useful. I had my boss (different job) come to me and ask why this report they were trying to run gave them nonsense results. I took a look at it, and the data, and realized the report writers had not correctly understood the data relationships or the real world relationships between a pressure vessel and piping system, and the pressure safety valve that protects it.
After a bunch of noodling around figuring out how to correctly report on many to many relationships and presenting the data in a sensible way, and fixing several different kinds of data problems, I produced two similar reports. One of them nearly gave the boss heart failure since it suggested there were several hundred vessels where the PSV was set higher than the pressure vessel, which is a very bad thing. However for the vast majority of them, it was a rounding error in the conversion between psi and kPa, or a legitimate case where a second PSV was set at a slightly higher pressure, which is allowed under some circumstances. As an aside 1 psi = 6.89476 kPa. You would be astonished how many ways that gets converted.
The second report washed all that away, and showed him about 15 actual cases of concern. That was a big deal. A few of them turned out to be data entry errors, linking the wrong PSV to a vessel, or an incorrect pressure setting, but there were several legitimate cases where they had to have a minor shutdown to correct the issue.
A third report showed us all the vessels (many) that did not have a related PSV, which is a different sort of big deal. We knew perfectly well there actually was a PSV installed, it just didn't look like it. One of the followup items was a couple days spent in small town Alberta at one of the plants. They had good vessel and PSV information in the database, but they had never actually related them, leading to a lot of work every time they had to do a regulator report. Nobody had ever told them it could be done, and nobody at the plant had had the time to figure it out, and nobody knew about it till I came along. A lot of people had made a lot of incorrect assumptions.
I sat with a senior operator and was gobsmacked at how well he knew the plant. The question was, which PSV or group of PSVs protected a particular vessel? Every vessel has to have one at least, and might have several. A particular PSV might actually be on piping and thus potentially protect several vessels, but it might not be the nearest ones.
He barely looked at the drawings. We started with inlet separators V100, V101, and V102. He knew the PSV numbers. I did the database updates. We went through the entire plant, me naming vessels, him with his eyeballs rolled up, naming PSVs. A couple times there was a "wait, they just changed that, I need the construction drawings for that number." We went on a tour of the plant to double check a few things. It was fun for both of us.
But it illustrated the problem. A gas plant is a very complicated place. Looking at the thing itself can be a bewildering experience, as is looking at the drawings. It's easy to get it wrong. The data in a database is a representation of the place, and if it's incorrect, any assumptions for change based on that are not going to be correct. Even the representation of the plant in the database doesn't tell you everything, since there are a great many variables in actual operating conditions. The assumptions might look correct, and the changes might even work under some circumstances. Until they don't, and then there's a big problem.
Many of the people these days that present 'solutions' to various problems don't really understand the underlying data or the real world it represents. They make convenient assumptions. They oversimplify. They assume change algorithms will work as designed. They ignore inconvenient facts. Then when things blow up they are nowhere to be found.
So, AI being a solution. I'm very dubious. It's just another big black box that ingests a bunch of data and spits out an answer. We don't know if it's looking at all the data. Or if the data is correct. We don't know how the data is being manipulated or evaluated along the way. We don't know how it prioritizes elements of the data. It might well assume a data element listed by 100 popular but erroneous sources is better than the one actually true element.
There's lots of dystopian SF novels that talk about the road we're on, and how our robot overlords might treat humans. It ranges from Skynet to nirvana. I think the end result is closer to Skynet. It could be very bad if two AI systems evolve and think of each other as an enemy. More likely, I think there will be one main entity, with smaller, semi-independent off-shoots, perhaps like how each arm of an octopus can seem to act independently. Humans might end up like how we keep creatures on a zoo, as curiosities or perhaps kept for some work that we're good at doing. What frightens me is that I'm likely to be alive to see it happen. It's already happening quickly, and computers think much faster than we do.
Of the Day
Driftwood (NZ)
Film
Linda
Newfoundland
Polar bears
Eagles
Why ever didn't I publish this, and maybe I did
The little hens and chicks are building their towers right now, but this is from several years ago..
90 days, or so ago
The view of the side bed mid April. Right now that bed is exploding with activity. The clematis on the right is doing really well, although the one on the left seems to be having a tough year. The space in the middle is totally taken up with an ambitious rose.
Or a view from mid June.
Flower
Here's just part of the rose that I just mentioned.
Landscape
Celina
No comments:
Post a Comment
Looking forward to reading your comment!