AI (artificial intelligence) in the Reefing World... Let's discuss it!

[AlyciaMarie Verified Admin]

AI is becoming increasingly present all around, and our reefing world isn't exempt from it either. Let's talk about it. In regards to AI and using it in reefing...

• What are the pros?
• What are the cons?
• Experienced reefers, what “word of caution” would you give reefers who rely solely on AI to answer reefing questions?
• As a member of the R2R community, do you think AI should be regulated by staff within our community? What does that look like to you?

And remember, we can agree to disagree! Let's be respectful in our discussion.

 

[patrck17]

I won't speak too long on this topic, but in my experience AI really shines as a sanity check or to polish information you already are very informed on.

I'll sometimes bounce ideas off ChatGPT about electronics design, which is my field. It produces a lot of ideas that I may not have thought of, and it also is very fast about condensing meaningful data in ways it can be presented. But it also makes mistakes, and if I didn't understand the material as well as I do I wouldn't really be able to tell it was wrong.

I think the same is likely true with reefing. I am a new reefer and I've asked AI plenty of questions, for the most part it all sounds logical enough, but I honestly wouldn't know where it is wrong. I have a feeling you guys with decades of experience would find a lot of the data produced to be incorrect.

Another issue I think we are going to be facing soon is the impact of AI training itself off of AI generated data. It cannot really learn so much as filter information more precisely. Eventually it will get to a point where it just creates noise (or gibberish).

 

[Macropharyngodon]

Dunno much about it, but what I will say, is that it can be harmful when pursuing difficult species. Sometimes it will suggest those species as options in tanks without specifying the specific requirements that make them very difficult to keep in captivity.

 

[Miami Reef]

I think responding with AI as your own voice by copying and pasting is plagiarism. It’s even worse if you don’t have experience to validate whether the advice you’re passing on is accurate.

It feels inauthentic, too.

 

[exnisstech]

I'll pass. Or at least I'm not looking to have assistance from AI. I realize even a Google search is a form of AI but that's extent of my use for it. When I search for info all I'm hoping to find are posts from real people and their experiences that I can read.
As on older person I've seen way too much science fiction come true in my lifetime for me to embrace AI.

 

[cdemoss01]

AI is the future of mostly everything in the economy today. Even reefing will be affected. An Ai fish tank was announced a month or two ago that does everything for you, feeding, tests, water changes etc. Personally I think AI will be integrated into the hobby whether we like it or not.
https://www.notebookcheck.net/Dangb...flagship-projector-at-IFA-2025.1099304.0.html

 

[Macropharyngodon]

Another issue I think we are going to be facing soon is the impact of AI training itself off of AI generated data. It cannot really learn so much as filter information more precisely. Eventually it will get to a point where it just creates noise (or gibberish).

This is a really good point! And its fascinating that in a world being shaped by AI, it becomes ever more important to not rely on it.

 

[Kooma]

AI is becoming increasingly present all around, and our reefing world isn't exempt from it either. Let's talk about it. In regards to AI and using it in reefing...

• What are the pros?
• What are the cons?
• Experienced reefers, what “word of caution” would you give reefers who rely solely on AI to answer reefing questions?
• As a member of the R2R community, do you think AI should be regulated by staff within our community? What does that look like to you?

And remember, we can agree to disagree! Let's be respectful in our discussion.

I have been utilizing ChatGPT for copper dosing calculations with great accuracy. I have been inputting daily testing logs into it and it then calculates my actual vs expected results based on the water estimation in my QT. This has allowed it to tell me a dose bump that is based on the previous 15 or more points of data.

I tell it have have a 60 gallon tank, it gives me a dose for a reading i want (Say 2.0ppm). 24 hours later I test and input the result, then ask for another dose to go to 2.25 for example. It looks at the dose entered, test result after 24 hours, and modifies the new dose based on the expected vs actual. Very handy to have.

With that said, when i ask it to recall information about when I was at a certain dose and how long I have been above 2.0 for example, it misses days or weeks of time and tells me that I am on day 9 even though I have been above 2.0 for 3 weeks.

Use it with caution, use it for some basic math, but cross check anything it tells you. Its math isnt always great, and often times it will feed you incorrect information.

One example of this is pump sizing. It told me the pump I wanted had a 16 foot head height at 2900GPH. It believed that it would flow 2900GPH 16 feet above the pump. Check and cross check.

 

[cdemoss01]

I have been utilizing ChatGPT for copper dosing calculations with great accuracy. I have been inputting daily testing logs into it and it then calculates my actual vs expected results based on the water estimation in my QT. This has allowed it to tell me a dose bump that is based on the previous 15 or more points of data.

I tell it have have a 60 gallon tank, it gives me a dose for a reading i want (Say 2.0ppm). 24 hours later I test and input the result, then ask for another dose to go to 2.25 for example. It looks at the dose entered, test result after 24 hours, and modifies the new dose based on the expected vs actual. Very handy to have.

With that said, when i ask it to recall information about when I was at a certain dose and how long I have been above 2.0 for example, it misses days or weeks of time and tells me that I am on day 9 even though I have been above 2.0 for 3 weeks.

Use it with caution, use it for some basic math, but cross check anything it tells you. Its math isnt always great, and often times it will feed you incorrect information.

One example of this is pump sizing. It told me the pump I wanted had a 16 foot head height at 2900GPH. It believed that it would flow 2900GPH 16 feet above the pump. Check and cross check.

I do this as well. ChatGPT gets dimensions from the web wrong a lot. But usually its math is good.

 

[Catboyeuthanasia]

For using large language models, always take their advice with some caution, since they are trained to give answers that agree with you as opposed to right answers. Using them as search engines are helpful though. A prompt like "give me a list of resources on how to set up a reef tank" or "set up a 5 step guide on how to learn to take care of seahorses, give references" would probably give you a good list of sorces you can read into further

Image analysis could also be a way AI is used in reefing. I've done work in a medical adjacent field, and using machine learning to categorize things, identify problems, or find regions of interest in images is quite powerful. However, ML needs a lot of data to train, and typically are best at tasks that imvolve doing a simple task many many times (e.g. counting the number of cells in an image with thousands of cells) or identifying subtle differences in images where we have a lot of previous data (e.g. identify if a cell is healthy, but we fed the network thousands of images of healthy and unhealthy cells). I guess you could use AI to count the number of zoanthids in a colony, or automatically identify the species of a fish or coral, but I dont see why you would want to do that. Maybe identifying disease in fish or coral, or classifying a bunch of corals in a large shipment? Grading how healthy a colony is? I still think a human expert would do better at these tasks, and the dataset is not large enough for the benifits of AI to outweigh having a human there, but who knows.

 

[KrisReef]

AI is becoming increasingly present all around, and our reefing world isn't exempt from it either. Let's talk about it. In regards to AI and using it in reefing...

• What are the pros?
• What are the cons?
• Experienced reefers, what “word of caution” would you give reefers who rely solely on AI to answer reefing questions?
• As a member of the R2R community, do you think AI should be regulated by staff within our community? What does that look like to you?

And remember, we can agree to disagree! Let's be respectful in our discussion.

General comment about electronic brains, auto corrects, and "smart" stuff in general.

What better than this mornings true story to capture my quick take.

Michael texted me to remind me that today is the last day to sign up for a conference he wants me to attend. I decided to go, went to the website and after I filled in my credit card information I wondered if this time would be the time that crooks grabbed my information and stole from the credit machine?

I went to reply and let Mike know that I had reserved my slot, and then noticed that his original message was encouraging me to reserve a "Sl#t" for the conference. My wife is going to be angry, but this is going to be extra fun perhaps?

In other words, it's got possibilities for good and evil results, just like me.

The end.

 

[Aquadude1]

It all depends what ai we are talking about. As far as LLMs is depends what data they access to.

For instance I think a chat bot for r2r that gets all its information from r2r with emphasis on certain members post. For instance adding a higher weight to a chemistry question answered by Randy

 

[Randy Holmes-Farley Verified Community Expert]

Be very, very skeptical. It knows facts, but understands nothing.

Let’s ask the google search AI among the most simple questions in reefing:

What’s is the most common element in a coral skeleton?

Wow, is it wrong.

The most common element in a coral skeleton is calcium (Ca), which combines with carbonate ions (CO3^2-)** to form calcium carbonate (CaCO3), a mineral known as aragonite in the case of hard corals. Calcium is the most abundant element in the skeleton, providing its rigid structure and strength

It knows what the skeleton is made of, but cannot correctly answer the question. Oxygen is clearly the most common element in a coral skeleton by both number and by mass.

I would also double check all math that ai do for dosing as in a number of cases posted to reef2reef, it made serious errors. Not 1+1 =3, but things like being off by a factor of 100 or 1000 due to not understanding what the calculations involve.

 

[Garf]

Seems to change it's mind a bit. I expect I ask the wrong questions;

 

[Hakon]

OOH, I know about this one! I work in the information technology field and have a lot of exposure to AI.

IMO AI has basically one pro right now, it is incredibly good at cataloguing consolidating and collating information. It's best applications are in things like taking meeting notes, pulling information out of email chains and chat logs, summarizing documents and collating information found through these tasks. If you have a large amount of very complicated and nuanced information our current LargeLanguage models are very good at pulling out the relevant points, consolidating presenting them in an easy to digest fashion.

This however is also the source of the largest pitfall to AI, honestly as someone in the field I think referring to tools like Gemeni or chatGPT as AI is dangerously misleading, AI is shorthand for Artificial Intelligence, and the word intelligence implies thought. Our current LargeLanguage models and other competing models do not think, they are in fact inherently incapable of thought, which means they are inherently incapable of evaluating the information we are feeding into them, thus leading to the oldest and most venerable law of computer science: "Garbage in, Garbage out" the AI will consider whatever you feed into it as true, it will output conflicting information if you show it conflicting information, thus if you try to use it to weed out misinformation you're likely to end up with it feeding you misinformation.

As an example I'm currently in the late planning/early setup phase of my first tank in ten years, wanting to catch up on new techniques for cycling I started looking around, found threads here and elsewhere and found mostly what I knew, ghost feeding techniques, fish in cycling, going all live rock, etc. If I'd fed this into chatGPT it'd either give me what I already knew or possibly cross reference all of these and present conflicting information as compatible technique. During my research however I stumbled accross a playlist on brsTV's youtube channel from a few years back where they did a comprehensive experiment in cycling with 12 nano tanks with proper control factors across parameters and even testing the DNA of the microbiomes to see which had the best balance of healthy bacteria. This post is long and I don't want to make what's turning into an essay into a thesis but where an AI would have output which individual tests did best against what from watching the series I was able to evaluate their approaches and what they didn't do to craft my own plan for my upcoming tank.

I would advise not trusting any information from a current gen AI unless I was absolutely certain that the information being fed into it was 100% accurate which in a hobby like this where different opinions and changes in trends over time lead to divergent approaches is increasingly hard as time goes on because accurate information from five years ago which is still relevant to certain methods will be detrimental to other new methods and trends and the AI cannot know that and cannot differentiate them.

So TL;DR. AI is useful for making sumaries and collating discrete collections of information, but feed it anything conflicting and results start varying wildly.

 

[Randy Holmes-Farley Verified Community Expert]

Seems to change it's mind a bit. I expect I ask the wrong questions;

This is an example of it knowing facts and not understanding anything (such as lanthanum chloride not existing as an uncharged entity in water). Might be best to follow up in the chem forum for a better understanding. lol

 

[KrisReef]

True or False:
"AI can't read?"

 

[bluemon]

It’s concerning how many newer hobbyists use AI as their initial source of information, and even quote them here.

With such a experienced, helpful and kind community here and elsewhere, you will probably get your answer within hours, and much more reliable than AI.

I hate AI in general, but I do use it for simple “calculations” for dosing or certain chemical dosage. That is as far as I trust it to do.

I also think it’s extra useless in IDing hitchhikers. Like extra ****

 

[Billyreef-ita]

I use it as second opinion, I have to say it is improving a lot and very quickly. So the point here is: do not focus on the mistakes that currently makes, most probably in few months those will not be there anymore, as it learns fast.

 

[vetteguy53081]

AI is becoming increasingly present all around, and our reefing world isn't exempt from it either. Let's talk about it. In regards to AI and using it in reefing...

• What are the pros?
• What are the cons?
• Experienced reefers, what “word of caution” would you give reefers who rely solely on AI to answer reefing questions?
• As a member of the R2R community, do you think AI should be regulated by staff within our community? What does that look like to you?

And remember, we can agree to disagree! Let's be respectful in our discussion.

Ive seen this concept appear and thought , no way and I was reluctant to even look at the matter but upon testing answers with Ai, I must say- they are not far off with answers despite not going in depth however I would still research to assure credibility with answers