16 min read

AI and beekeeping: counting mites

Counting mites is a thankless task, but you sometimes have to do it. New AI software can make the task easier, faster and more accurate (choose any three).
Picture of a white tray with debris from a bee hive on it
1, 2, 3, 4, 5, 6, 7, 8, 8, 9, 10, 11, 12, 12, 13, 14, 16, Er 15? Or was it 16? Where was I? … 1, 2, 3

In case you've not noticed, artificial intelligence (AI) is everywhere.

My coffee machine has a PhD. in fluid dynamics and is better qualified than I am. Not in itself a surprise, but nevertheless disappointing when you take into account the high cost of private schooling, personal tutors and university fees {{1}}.

The fridge monitors how much milk we use, adding more to the weekly supermarket delivery as needed. It can access my diary, and so knows we have friends and their children visiting next week. It will increase the milk order to reflect this, but not before checking their private medical records to see if they are lactose intolerant. Of course, the amount ordered will take account of the use-by date on the cartons in the supermarket.

What's the point of all this technology if it doesn't offer savings and efficiency?

Using Bluetooth, the fridge will turn on the coffee machine when our friends are 30 minutes away, enabling me to bolt down a bracing espresso before chaos ensues the kids arrive, and ensuring I can have refreshing iced lattes for the adults after their long drive.

Well, perhaps not … but AI is everywhere.

ChatGPT

Although readers may not have Dr. Nespresso MA(Cantab) DPhil. in their kitchen, or a fridge with a brain the size of a planet, they will have been exposed to ChatGPT (a generative artificial intelligence chatbot developed by OpenAI and released in 2022). This and a rash of similar programs {{2}} are trained on existing information — so-called large language models (LLM) — to generate human-like responses in text, speech, and images.

Sometimes, and — in fact — increasingly.

But still only 'human-like'.

With a few carefully chosen prompts, ChatGPT will produce 10's, 100's or 1000's of words on any topic you want — Early Byzantium pottery, toenails, or jealousy {{3}}.

Or, with appropriate prompts, an essay combining all three.

I fortunately abandoned academia before ChatGPT became the solution to meeting the Monday morning essay deadline, while simultaneously spending most of the weekend partying hard revising in the library.

Academics had, and used, plagiarism detection software, but it was relatively easy to circumvent if you had reasonable English language skills and access to a thesaurus. Now it's almost impossible to keep up with the technology, though the inclusion by AI of invented citations and non-existent authors remains a giveaway.

I'm not a big fan of ChatGPT. Unless the output has been rigorously checked you can find all sorts of idiotic non sequiturs getting through, at least for those familiar with a topic.

Look at this example of nonsense from irescuebees.com {{4}}

Lorenzo Lorraine Langstroth, an American apiarist, clergyman, and teacher, was known for his contributions to the field of beekeeping. He is also credited with being the first Westerner to scale Mount Everest.

Edmund Hillary (who was a beekeeper) and Tenzing Norgay were the first to the top of Everest … but you knew that. It was Langstroth who was the first to climb K2 😉.

Screenshot of artificial intelligence (AI)-generated text
A screenshot of more AI word salad — when did Hillary join, WTF is 'bee pollen' and where were the bees removed from?

Be careful, there's a lot of AI on some beekeeping websites and, increasingly, discussion forums. Some may be fine, a lot is just garbage … I doubt any is really good as it's just a rehash of what's written elsewhere.

Tar-pits and poisons

If you understand a topic, and what you're reading doesn't make sense, it's probably AI generated (or possibly just badly written, or both).

If you don't understand a topic — which might be why you're reading about it — and it's internally inconsistent, or 'hedges its bets' i.e. fails to provide a definitive answer, then always suspect AI.

Remember, beekeepers are often unequivocal in their views.

Often wrong, but nevertheless unequivocal 😉.

Short sentences, and the absence of a consistent writing style is another clue.

And it's going to get worse.

That was a short sentence, but it was not generated by ChatGPT 😉.

Firstly, there's going to be lots more output generated automatically by ChatGPT. And this output, including the errors, inconsistencies, and random 'word salad', will be used to further train ChatGPT which will then generate yet more web pages, PDFs, courses for beginners etc.

Already, that irescuebees.com page, in which Langstroth's exploits on Everest are discussed, will have been used to populate the LLMs that train ChatGPT {{5}}.

But it doesn't end there.

Websites are fighting back. Tar-pits (like Nepenthes; well named) create endless numbers of pages that contain dozens of additional links, all with the idea of deliberately causing the LLM models to collapse. Or there's Iocaine (another great name), which generates infinite amounts of garbage, intentionally poisoning the LLM.

Garbage in, garbage out.

Is it all bad news?

No.

There are beneficial applications of AI for beekeeping, and this post is about two one that's recently been published and should be useful {{6}}.

I'm sure there will be (and maybe already are) other AI applications that benefit bees or beekeepers, but this is the first that I've seen that is both practical and relevant.

I'm not interested in the robotic hives that claim to 'Save the bees' (they don't need saving) … but, as I get older and more decrepit, I am interested in robotic exoskeletons which might 'Save my back' (inevitably, these also use AI) {{7}}.

My interest is largely confined to applications that improve rather than replace the interactions with my bees. Things that might make me a better beekeeper (or at least a less bad one), with more insight into what's happening inside my hives, with lower winter colony losses, or increased honey production.

Varroa detection and counting

Varroa affects almost all managed colonies. With appropriate management, losses attributable to Varroa should be negligible, but it's still useful to know the level of infestation and the efficacy of treatment.

Both of which mean that you probably need to count Varroa.

Phoretic mite levels can be measured during the season by sugar dusting, alcohol washes or CO2 {{8}}, and the natural — or miticide induced — drop of Varroa through the open mesh floor (OMF) onto a white tray gives an indication of the number of mites present (or, now, absent 😉).

Many beekeepers will be familiar doing Varroa counts using sticky boards i.e. Correx or similar sheets placed underneath the OMF with a thin film of vegetable oil sprayed on them.

It's a dull, repetitive and rather thankless task … if only it could be automated 🤔.

This post is for subscribers only