You could say the same about a plant identification book.
It’s not so much that AI for plant identification is bad, it’s that the higher the stakes, the more confident you need to be. Personally, I’m not going foraging for mushrooms with either an AI-based plant app or a book. Destroying Angel mushrooms look pretty similar to common edible mushrooms, and the key differences can disappear depending on the circumstances. If you accidentally eat a destroying angel mushroom, the symptoms might not appear for 5 to 24 hours, and by then it’s too late. Your liver and kidney are already destroyed.
But, I think you could design an app to be at least as good as a book. I don’t know if normal apps do this, but if I made a plant identification app, I’d have the app identify the plant, and then provide a checklist for the user to use to confirm it for themselves. If you did that, it would be just like having a friend just suggest checking out a certain page in a plant identification book.
The problem with AI is that it’s garbage in, garbage out. There’s some AI generated books on Amazon now for mushroom identification and they contain some pretty serious errors. If you find a book written by an actual mycologist that has been well curated and referenced, that’s going to be an actually reliable resource.
Are you assuming that AI in this case is some form of generative AI? I would not ask chatgpt if a mushroom is poisonous. But I would consider using a convolutional neural net based plant identification software. At that point you are depending on the quality of the training data set for the CNN and the rigor put into validating the trained model, which is at least somewhat comparable to depending on a plant identification book to be sufficiently accurate/thorough, vs depending on the accuracy of a story that genAI makes up based on reddit threads, which is a much less advisable venture
The books on Amazon are vomited out of chat GPT. If there’s a university-curated and trained image recognition AI, that’s more likely to be reliable provided the input has been properly vetted and sanitized.
If you’re using the book correctly, you couldn’t say the same thing. Using a flora book to identify a plant requires learning about morphology and by having that alone you’re already significantly closer to accurately identifying most things. If a dichotomous key tells you that the terminating leaflet is sessile vs. not sessile, and you’re actually looking at that on the physical plant, your quality of observation is so much better than just photographing a plant and throwing it up on inaturalist
You could say the same about a plant identification book.
It’s not so much that AI for plant identification is bad, it’s that the higher the stakes, the more confident you need to be. Personally, I’m not going foraging for mushrooms with either an AI-based plant app or a book. Destroying Angel mushrooms look pretty similar to common edible mushrooms, and the key differences can disappear depending on the circumstances. If you accidentally eat a destroying angel mushroom, the symptoms might not appear for 5 to 24 hours, and by then it’s too late. Your liver and kidney are already destroyed.
But, I think you could design an app to be at least as good as a book. I don’t know if normal apps do this, but if I made a plant identification app, I’d have the app identify the plant, and then provide a checklist for the user to use to confirm it for themselves. If you did that, it would be just like having a friend just suggest checking out a certain page in a plant identification book.
The problem with AI is that it’s garbage in, garbage out. There’s some AI generated books on Amazon now for mushroom identification and they contain some pretty serious errors. If you find a book written by an actual mycologist that has been well curated and referenced, that’s going to be an actually reliable resource.
Are you assuming that AI in this case is some form of generative AI? I would not ask chatgpt if a mushroom is poisonous. But I would consider using a convolutional neural net based plant identification software. At that point you are depending on the quality of the training data set for the CNN and the rigor put into validating the trained model, which is at least somewhat comparable to depending on a plant identification book to be sufficiently accurate/thorough, vs depending on the accuracy of a story that genAI makes up based on reddit threads, which is a much less advisable venture
The books on Amazon are vomited out of chat GPT. If there’s a university-curated and trained image recognition AI, that’s more likely to be reliable provided the input has been properly vetted and sanitized.
If you’re using the book correctly, you couldn’t say the same thing. Using a flora book to identify a plant requires learning about morphology and by having that alone you’re already significantly closer to accurately identifying most things. If a dichotomous key tells you that the terminating leaflet is sessile vs. not sessile, and you’re actually looking at that on the physical plant, your quality of observation is so much better than just photographing a plant and throwing it up on inaturalist
Not to mention, the book is probably going to list look-alike plants, and mention if they are toxic. AI is just going to go “It’s this thing”.