465
Views
0
CrossRef citations to date
0
Altmetric
Forum

COLOSS B-RAP Expert Evaluation of Beekeeping Advice From ChatGPT, Part 1

ORCID Icon, ORCID Icon, ORCID Icon, ORCID Icon, ORCID Icon, ORCID Icon, , ORCID Icon, ORCID Icon, & ORCID Icon show all

The advanced language model ChatGPT is capable of understanding and generating human-like text. It can be integrated into various services, ranging from customer support to educational platforms, providing personalized assistance, information and guidance. For straightforward, low-complexity medical quest­ions, ChatGPT has been shown to have potential as an AI-assisted decision support tool in medicine (Harskamp & De Clercq, Citation2024). In apiculture, hive management is an important factor in maintaining healthy and productive honey bee colonies (Sperandio et al., Citation2019; Steinhauer et al., Citation2021). Artificial intelligence-based linguistic models could provide an easy-to-access advisory service in countries where no advisory services are available or to relieve advisors. At a workshop of the COLOSS core project B-RAP (Fabricius Kristiansen et al., Citation2022) held in Olomouc, Czechia, in February 2024, we, therefore, tested the ability of ChatGPT3.5 to deal with some common questions in beekeeping. The question formulation always included rough information on location and date and formulated the beekeeping-related problem as a question allowing an open answer. The panel of 13 experts present (researchers, beekeeping advisors, veterinarians), many of them beekeepers themselves, evaluated the answers.

For this article, we consider two questions which we asked ChatGPT3.5 regarding the serious honey bee disease American Foulbrood (=AFB), which is a notifiable disease in the whole of the European Economic Area (Chauzat et al., Citation2013). The questions differed in level of detail - the first one asked about general symptoms, while the second one asked for advice in a certain situation (AFB outbreak). The questions, (unaltered) ChatGPT output and experts’ comments are shown in and .

Table 1. ChatGPT output and experts’ comments on the question: ‘What are the symptoms of American foulbrood?’; Rating: C = correct and valuable information, sometimes minor details are missing; U = correct, but useless information for the question, M = correct, but important details are missing, I = Incorrect information.

Table 2. ChatGPT output and experts’ comments on the question: ‘What measures should be implemented in a beekeeping operation in Norway affected by American foulbrood?’ Rating: C = correct and valuable information, sometimes minor details are missing; U = correct and useless information for the question, M = correct, but important details are missing, I = Incorrect information.

Our small challenge on AFB shows: ChatGPT3.5 in its current state is a good teacher for standard beekeeping knowledge (e.g. content of a beginner’s course), but its use should be avoided for critical or specific situations. The results of question 1 show the strength of ChatGPT. It summarizes common knowledge about an important bee disease at the level of detail that can be found in textbooks for general beekeeping. All information given is correct although some important details are missing, according to the experts’ panel (). Also, the level of useless information is low. However, when being asked a detailed question on a critical situation (), the answer given is a mixture of insufficient information, useless clutter and incorrect information. This might, again according to the experts’ panel, even be dangerous for the health of the honey bees in the region (pathogen spread through transport or application of antibiotics). Inexperienced beekeepers may be confused at best and ill-advised at worst. Furthermore, the answers are not in compliance with Norway’s legislation and may lead to legal problems for the beekeeper.

It is evident that ChatGPT does not differentiate between important and nice-to-have information, as an advisor would do (). It also does not structure information in a logical way, which would help the inquirer to put the acquired knowledge into action (e.g. step-by-step instruction).

Nonetheless, we see a possibility of ChatGPT (or a similar advanced language model) to act as powerful tools in an advisory service. They could provide beekeepers with personalized and summarized information for their current needs, in the language and level of detail, which they need. Let’s imagine a new beekeeping advisor tool, named ChatBEE being usable at both European and national levels, which has previously been trained with public domain resources of high quality, like those being produced within various (transnational) honey bee research efforts (e.g. FAO et al., Citation2021; BPRACTICES Consortium, Citation2020), research associations (e.g. the COLOSS BEEBOOK, see Carreck et al., Citation2022) or by the national beekeeping structures (e.g. reference laboratories for bee health, beekeeping advisor centers). Such basic information would allow ChatBEE to rank information for quality (e.g. preferring official sources or resources from research institutions over other internet sources). Additionally, the inquirers can do their own quality checks as ChatBEE gives the sources of the presented information. It is sensitive to regional differences in its answers (e.g. not advising for antibiotics in Europe) and highlights contradictory information (e.g. antibiotics not allowed in each country). It is also transparent about its ‘knowledge gaps’, giving short high-quality answers rather than filling the answer up with unnecessary information.

In Part two of the COLOSS B-Rap AI challenge, the focus will be on practical beekeeping and the importance of interaction in beekeeping advisory services.

Figure 1. AI-generated pictures from Bing Image Creator showing how beekeeping advising is depicted by an AI. It is based on the following prompts: ‘a photo of a beekeeping advisor giving advice to a group of beekeepers in Europe’.

Figure 1. AI-generated pictures from Bing Image Creator showing how beekeeping advising is depicted by an AI. It is based on the following prompts: ‘a photo of a beekeeping advisor giving advice to a group of beekeepers in Europe’.

Disclosure Statement of funding

No potential conflict of interest was reported by the author(s).

References

  • BPRACTICES consortium (2020). Guidelines on sustainable management of honey bee diseases in Europe. Istituto Zooprofilattico Sperimentale delle Venezie, Venezia. https://www.izslt.it/bpractices/wp-content/uploads/sites/11/2020/03/bpractices-guidelines.pdf
  • Carreck, N. L., Dietemann, V., Ellis, J. D., Evans, J. D., Neumann, P., & Chantawannakul, P. (2022). The COLOSS BEEBOOK, a manual of standard methods for honey bee research. Bee World, 99(1), 11–13. https://doi.org/10.1080/0005772X.2021.1981677
  • Chauzat, M. P., Cauquil, L., Roy, L., Franco, S., Hendrikx, P., & Ribière-Chabert, M. (2013). Demographics of the European apicultural industry. PloS One, 8(11), e79018. https://doi.org/10.1371/journal.pone.0079018
  • Fabricius Kristiansen, L., Kristiansen, P., Vejsnæs, F., & Morawetz, L. (2022). Is COLOSS an Ivory Tower of beekeeping science? Efforts to bridge research and practice (B-RAP). Bee World, 99(1), 5–7. https://doi.org/10.1080/0005772X.2021.1993612
  • FAO, IZSLT, Apimondia & CAAS. (2021). Good beekeeping practices for sustainable apiculture FAO Animal Production and Health Guidelines No. 25. Rome. https://doi.org/10.4060/cb5353en
  • Harskamp, R. E., & De Clercq, L. (2024). Performance of ChatGPT as an AI-assisted decision support tool in medicine: A proof-of-concept study for interpreting symptoms and management of common cardiac conditions (AMSTELHEART-2). Acta Cardiologica, 1–9. https://doi.org/10.1080/00015385.2024.2303528
  • Sperandio, G., Simonetto, A., Carnesecchi, E., Costa, C., Hatjina, F., Tosi, S., & Gilioli, G. (2019). Beekeeping and honey bee colony health: A review and conceptualization of beekeeping management practices implemented in Europe. Science of the Total Environment, 696, 133795. https://doi.org/10.1016/j.scitotenv.2019.133795
  • Steinhauer, N., VanEngelsdorp, D., & Saegerman, C. (2021). Prioritizing changes in management practices associated with reduced winter honey bee colony losses for US beekeepers. The Science of the Total Environment, 753, 141629. https://doi.org/10.1016/j.scitotenv.2020.141629
  • WOAH. (2023). Chapter 3.2.2. American Foulbrood of honey bees (infection of honey bees with Paenibacillus larvae). In: Manual of Diagnostic Tests and Vaccines for Terrestrial Animals. 12th ed.