论文信息
论文解读
AI 辅助整理- 方法
- 385份问卷,描述统计、配对比较与探索性秩相关(Holm校正)。
- 发现
- 87.8%支持纳入法律权利信息,但样本非农户,关联证据弱。
- 机会点
- 需以真实农户为对象,结合具体法律场景与可观测转介路径评估AI咨询服务。
论文摘要
摘要:人工智能支持的农业咨询引发了关于有用信息、服务提供者责任以及法律权利信息可及性的问题。本研究基于孟加拉国385份具备人工智能认知的成年人调查记录,考察了受访者对上述问题的陈述性偏好。混合样本由法学专业学生、毕业生、法律从业者及其他受访者组成;工作簿未识别农民、农业人工智能使用者或对农民法律权利的实际了解情况。因此,本分析考察的是关于权利信息与保障措施的问卷回答,而非声称测量农民的权利意识或人工智能服务的效果。各单项采用描述性分布、配对比较及经Holm校正的探索性秩相关进行报告。385名受访者中,338人(87.8%)赞同将法律权利信息纳入负责任的数字农业服务,330人(85.7%)赞同明确重要建议的责任归属,325人(84.4%)赞同告知不确定性与局限性,247人(64.2%)赞同在人工智能建议不足时提供人工协助。权利信息项比有用性项高出24.4个百分点;215名受访者对两者均表示赞同。各陈述内容不同,其配对比较既不能识别某项服务的效果,也不能对其各特征的价值进行排序。与权利信息项的探索性相关性较小,且在四项检验校正后未提供明确的关联证据。本文将关于权利的信息与经证实的法律知识加以区分,并建议在明确识别的农业使用者、具体法律场景及可观察的转介路径下开展评估。
全文
Research SquareFarmers Rights Information and Safeguards in AI
Agricultural Advisory Services in Bangladesh
Dr Kazi Abdul Mannan
Center for Academic & Professional Career Development and Research (CAPCDR)
https://orcid.org/0000-0002-7123-132X
Sumtahin Pushpita
Shanto-Mariam University of Creative Technology https://orcid.org/0009-0003-5573-0684
Research Article
Keywords: agricultural advisory services, farmers’ rights, AI governance, legal information, human
assistance, Bangladesh
Posted Date: November 16th, 2026
DOI: https://doi.org/10.21203/rs.3.rs-11722841/v1
License: This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License
Additional Declarations: The authors declare no competing interests.
1
Farmers Rights Information and Safeguards in AI Agr icultural
Advisory Services in Bangladesh
Sumtahin Pushpita
Department of Law, Shanto-Mariam University of Creative Technology, Dhaka, Bangladesh
Email: sumtahin@gmail.com
ORCID: https://orcid.org/0009-0003-5573-0684
Prof. Dr Kazi Abdul Mannan
Department of Business Administration, Shanto-Mariam University of Creative Technology,
Dhaka, Bangladesh
Email: drkaziabdulmannan@gmail.com
ORCID: https://orcid.org/0000-0002-7123-132X
Abstract
AI-supported agricultural advice raises questions about useful information, the responsibilities
of providers and the accessibility of legal-rights information. This study examines stated
preferences about these issues in 385 adult, AI-aware survey records from Bangladesh. The
mixed sample consists of law students, graduates, legal professionals and other respondents;
the workbook does not identify farmers, agricultural AI users or actual knowledge of farmers’
legal rights. The analysis therefore examines questionnaire responses about rights information
and safeguards rather than claiming to measure farmers’ rights awareness or the effects of an
AI service. Individual items are reported using descriptive distributions, paired comparisons
and exploratory rank correlations with Holm adjustment. Of 385 respondents, 338 (87.8%)
endorsed including legal-rights information in responsible digital agricultural services, 330
(85.7%) endorsed identifying responsibility for important advice, 325 (84.4%) endorsed
communicating uncertainty and limitations, and 247 (64.2%) endorsed human assistance when
AI advice is inadequate. The rights-information item exceeded the usefulness item by 24.4
percentage points; 215 respondents endorsed both. The statements differ in content, and their
paired comparison does not identify the effects of a service or rank the value of its features.
Exploratory correlations with the rights-information item were small and did not provide clear
evidence of association after adjustment for four tests. The article distinguishes information
about rights from demonstrated legal knowledge and recommends evaluation with identified
agricultural users, specific legal scenarios and observable referral pathways.
Keywords: agricultural advisory services; farmers’ rights; AI governance; legal information;
human assistance; Bangladesh
- Introduction
Agricultural information can influence decisions about crops, inputs, water, pests and markets.
Digital and AI-supported advisory services may make some information easier to obtain, but
answers must be relevant to the user’s crop, locality, resources and decision. Incorrect or
incomplete recommendations can have practical consequences. Recent reviews of agricultural
2
AI identify reliability, data governance, digital access and explainability as interconnected
concerns.1
Rights information is a further part of responsible service design. An agricultural user may
need to know what data a service collects, who is responsible for advice, whether a human can
help and where a complaint may be directed. A general recommendation to include rights
information does not establish which legal rights apply to a particular user or whether an
available remedy works. Farm-related information is not automatically personal data; its legal
treatment depends on whether it identifies a person and on the relevant legal context.2
The original study proposal asks whether AI agricultural advisory services improve farmers’
legal-rights awareness. The supplied questionnaire, however, asks whether legal-rights
information should be incorporated into responsible services; it does not tes t respondents’
knowledge of specific rights or observe a change in awareness after using an AI service. The
available status categories also do not identify farmers or actual agricultural AI users. A claim
that AI caused farmers to understand their rights would therefore go beyond the data.
This article investigates a more precise question: how do respondents endorse rights-
information and accountability statements alongside statements about usefulness and human
assistance? Its contribution is to distinguish an informational safeguard from both
demonstrated legal knowledge and observed use. It uses the same Group 16 data as related
research but focuses specifically on rights-information inclusion, responsibility and human
help rather than treating trust or general AI literacy as the primary outcome.
The objectives are to describe the recorded sample and agricultural-advisory items, compare
selected statements within respondents, and explore associations with endorsement of rights-
information inclusion. The research questions are: How often are rights information, provider
responsibility, uncertainty disclosure and human assistance endorsed? How does support for
rights information coexist with perceived usefulness? How is rights-information endorsement
associated with selected usefulness, reporting, data-transparency and AI-awareness items?
What additional measures would be needed to study actual legal empowerment?
The recorded sample includes law students, graduates, legal professionals and other
respondents. It is not presented as representative of farmers or rural users in Bangladesh.
Section 2 develops the conceptual and legal context; Section 3 explains the data and analysis;
Section 4 reports results; and Sections 5-7 discuss implications and limitations.
2. Literature Review and Theoretical Foundation
Research on digital agriculture links advisory quality to information sources, local suitability,
data governance and the ability of users to obtain help. This section distinguishes these service
features from the legal knowledge and outcomes that an effective rights-based assessment
would need to measure.
1 Richa Bhattarai and others, ‘Scale, Trust, and the Digital Divide: A Systematic Review of AI and ML for
Agricultural Applications’ (2026) 9 Frontiers in Artificial Intelligence 1798896.
2 Isabelle M Carbonell, ‘The Ethics of Big Data in Big Agriculture’ (2016) 5(1) Internet Policy Review, DOI:
10.14763/2016.1.405; Personal Data Protection Act 2026 (Bangladesh Act No 63 of 2026), s 2.
3
2.1 AI advisory services and legal information
AI-supported advisory services can generate or organise information about agricultural
questions. Their usefulness depends on whether information is accurate and appropriate to a
particular location, crop, season and user. A review of smallholder AI in South Asia cautions
that claims about transformation have advanced faster than evidence on farmer outcomes.3 The
present questionnaire does not identify a tool or test the quality of any recommendation.
Digital agricultural information also involves questions of responsibility and control of data.
Scholarship on digital agriculture describes possible asymmetries between users, technology
providers and other organisations that collect or use information. 4 A service might explain its
own data practices, but a rights-based approach also asks whether the user can understand
applicable rights and seek an effective response when a problem arises.
2.2 Rights information, knowledge and empowerment
The rights-information item asks whether legal-rights information should be incorporated into
responsible digital agricultural services. Agreement with that statement is a service preference,
not evidence that a respondent knows the legal content of the information, that a service has
supplied it, or that a farmer can exercise a right. These distinctions matter because legal
empowerment involves more than access to a general statement: it may require understanding
a specific entitlement, identifying a responsible actor, obtaining advice and using a practical
remedy.
An evaluation could present a defined scenario, identify which information the system gives,
and test whether a participant understands the next available step. It could separately assess
whether a referral works. The current instrument contains no such test, baseline or follow-up.
The title and interpretation therefore refer to rights information and safeguards rather than a
demonstrated increase in legal awareness.
2.3 Trust, responsibility and human assistance
The questionnaire asks whether farmers should know who is responsible for important AI
advice, whether uncertainty and limitations should be communicated, and whether human help
should be available when advice is inadequate. These are related aspects of accountability but
are not interchangeable. A named provider may not make a response accurate; disclosure of
uncertainty may not give the user access to a qualified person; and a human-help link is not
evidence that help can actually be obtained.
Trust Theory describes willingness to rely on another actor under uncertainty.5 Trust should be
calibrated to evidence and service conditions, not inferred solely from a user’s support for
3 Amar Singh, Vinod Kumar Shukla and Chatter Singh, ‘From Pilots to Plots: A Critical Review of Artificial
Intelligence for Smallholder Agriculture in South Asia’ (2026) DOI: 10.1177/00307270261489284.
4 Laurens Klerkx, Emma Jakku and Pierre Labarthe, ‘A Review of Social Science on Digital Agriculture, Smart
Farming and Agriculture 4.0: New Contributions and a Future Research Agenda’ (2019) 90–91 NJAS:
Wageningen Journal of Life Sciences 100315; Carbonell (n 2).
5 Roger C Mayer, James H Davis and F David Schoorman, ‘An Integrative Model of Organizational Trust’
(1995) 20 Academy of Management Review 709.
4
transparency. A recent review of agricultural AI highlights data governance, skills,
explainability, reliability and cost as connected influences on trust and adoption. 6 Its cross-
context synthesis provides a research agenda, not a direct finding about the participants in this
workbook.
2.4 Theoretical framework
UTAUT2 provides concepts such as expected usefulness, effort and facilitating conditions. 7
Trust Theory addresses reliance under uncertainty, while a digital-governance perspective
examines responsibility, data practices and routes for correction or redress.8 These approaches
are complementary: perceived usefulness does not itself show that legal information is correct,
and support for governance does not establish whether a service is accessible.
The present questions do not operationalise the full UTAUT2 model or measure a validated
legal-empowerment construct. The study does not estimate structural paths or moderation and
does not claim to validate the theories. It uses them as interpretive perspectives for a
descriptive, item-level analysis.
2.5 Bangladesh legal context
Bangladesh’s Personal Data Protection Act 2026 supplies part of the legal context when a
service processes identifiable personal information. The Act defines personal data and includes
provisions concerning correction and security, subject to its scope, commencement and
applicable legal conditions. 9 Not all farm information is automatically identifiable personal
data, and the questionnaire does not identify a data controller, establish compliance or
demonstrate a breach. Whether particular rights or duties apply requires the facts of the
processing activity and the relevant provisions.
Farmers’ legal rights may involve matters beyond data protection. The questionnaire does not
identify a particular land, agricultural-input or contractual issue, so this article does not assert
a specific additional entitlement from a general survey answer. Future research should identify
the legal topic, applicable rules and a route for obtaining help before assessing rights
awareness.
2.6 Analytical expectations and measurement
The AFR items cover usefulness of advisory information, responsibility, uncertainty
disclosure, human assistance and inclusion of rights information. The AAR and ADG items
cover related but distinct questions about reporting and data-use transparency. Their breadth
6 Bhattarai and others (n 1).
7 Viswanath Venkatesh, James YL Thong and Xin Xu, ‘Consumer Acceptance and Use of Information
Technology: Extending the Unified Theory of Acceptance and Use of Technology’ (2012) 36 MIS Quarterly
157.
8 Mayer, Davis and Schoorman (n 5); Callum Eastwood and others, ‘Managing Socio-Ethical Challenges in the
Development of Smart Farming: From a Fragmented to a Comprehensive Approach for Responsible Research
and Innovation’ (2019) 32 Journal of Agricultural and Environmental Ethics 741.
9 Personal Data Protection Act 2026 (Bangladesh Act No 63 of 2026), s 1; Personal Data Protection Act 2026 (n
2); Personal Data Protection Act 2026 (Bangladesh Act No 63 of 2026), s 12; Personal Data Protection Act 2026
(Bangladesh Act No 63 of 2026), s 17.
5
means that averaging them into a single scale would require a defensible measurement model.
Internal consistency alone would not establish construct validity.10
Three exploratory paired comparisons examine the rights-information item against usefulness,
provider responsibility against human assistance, and rights information against human
assistance. Four exploratory rank correlations examine rights-information responses in relation
to usefulness, reporting, transparent data-use practices and awareness of AI inaccuracies. The
supplied materials do not establish a preregistered test family. The analyses do not measure an
intervention or a change in knowledge.
3. Methodology
This section explains the recorded sample, the meanings and limits of the questionnaire items,
ethics safeguards and statistical analysis. It distinguishes a stated service preference from
observed agricultural use and legal knowledge.
3.1 Design and study population
The study analyses a cross-sectional questionnaire. The unit is an individual record; no farm,
advice system, legal case, referral or legal outcome is observed. Respondent-status categories
comprise law students, law graduates, advocates or legal professionals and other relevant
respondents. The workbook does not identify farmers or verify whether respondents used
agricultural AI. Accordingly, the sample is not described as representative of rural farmers or
all users of agricultural services.11
3.2 Sample, eligibility and provenance
The workbook contains 385 records, each coded as consenting, adult eligible and AI-aware;
respondent IDs are unique. The supplied files do not document the recruitment dates, sampling
frame, invitations, response rate or independent duplicate-screening procedures. A target of
385 can arise from a conventional sample-size calculation under specified assumptions, but it
does not establish random selection or a national margin of error.12
The researcher identified the numerical answers as real survey responses. The Codebook labels
the respondent identifier sequence and optional contextual comments as synthetic. Their
preparation and relationship to source records should be documented; contextual text was
excluded from analysis. These labels do not independently establish the origin of numerical
data, so fieldwork and consent documentation remain important.
3.3 Questionnaire and analytical items
All focal items use five response categories from strongly disagree to strongly agree. AFR1
asks whether AI agricultural advisory services can help farmers access useful information.
10 Godfred O Boateng and others, ‘Best Practices for Developing and Validating Scales for Health, Social, and
Behavioral Research: A Primer’ (2018) 6 Frontiers in Public Health 149; Klaas Sijtsma, ‘On the Use, the
Misuse, and the Very Limited Usefulness of Cronbach’s Alpha’ (2009) 74 Psychometrika 107.
11 Erik von Elm and others, ‘The Strengthening the Reporting of Observational Studies in Epidemiology
(STROBE) Statement: Guidelines for Reporting Observational Studies’ (2007) 4 PLOS Medicine e296.
12 William G Cochran, Sampling Techniques (3rd edn, John Wiley & Sons 1977).
6
AFR2 concerns knowing who is responsible for important advice; AFR3 concerns
communication of uncertainty and limitations; AFR4 asks for human assistance when advice
is inadequate; and AFR5 concerns incorporating legal-rights information in responsible
services.
The exploratory associations use AAR4, which concerns ways to report harmful or inaccurate
recommendations; ADG2, which describes confidence under transparent data-use practices;
and AIL1, which concerns awareness of plausible but inaccurate AI information. Exact
wording appears in Appendix A. The items express attitudes or stated awareness, not objective
legal knowledge, observed service use or actual data practices. No pilot, adapted scale or
validation procedure is asserted without documentation.13
3.4 Data collection and screening
The focal numerical items contain valid answers for the 385 supplied records. The workbook
codes eligibility and consent but does not establish administration mode, fieldwork dates,
invitations, exclusions or response rate. Status and location are reported as recorded without
inferring agricultural occupation or experience with an AI advisory service. Optional
contextual text is not analysed.
3.5 Statistical analysis
For each item, the distribution and number endorsing agree or strongly agree are reported.
Neutral responses are included in non-endorsement and are not treated as opposition. Pointwise
Wilson intervals describe uncertainty under conventional assumptions that are not secured by
an undocumented sampling design; they should not be interpreted as national estimates.14
Three within-person contrasts compare AFR5 with AFR1, AFR2 with AFR4, and AFR5 with
AFR4 using exact McNemar tests and Holm adjustment across the three tests. Whole-record
bootstrap percentile intervals preserve pairing and are pointwise. Different wording means a
paired difference is not a direct ranking of service features or their causal effectiveness.15
Four exploratory Spearman correlations compare AFR5 with AFR1, AAR4, ADG2 and AIL1.
Holm correction covers the four p-values; percentile bootstrap intervals remain pointwise. The
items are ordinal responses measured at the same time and are not latent scores. A non-
significant association does not establish that a relationship is absent, and no estimate identifies
the effect of an advisory intervention on knowledge or behaviour.16
13 Boateng and others (n 10).
14 Cochran (n 12).
15 Quinn McNemar, ‘Note on the Sampling Error of the Difference between Correlated Proportions or
Percentages’ (1947) 12 Psychometrika 153; Sture Holm, ‘A Simple Sequentially Rejective Multiple Test
Procedure’ (1979) 6 Scandinavian Journal of Statistics 65; Bradley Efron and Robert J Tibshirani, An
Introduction to the Bootstrap (Chapman & Hall 1993).
16 Holm (n 15); Douglas G Altman and J Martin Bland, ‘Absence of Evidence Is Not Evidence of Absence’
(1995) 311 BMJ 485; Ronald L Wasserstein and Nicole A Lazar, ‘The ASA Statement on p -Values: Context,
Process, and Purpose’ (2016) 70 American Statistician 129.
7
3.6 Measurement limitations
The AFR5 item does not ask participants to name a specific right, apply a law or report
receiving rights information. AFR2 and AFR4 ask for different safeguards; support for one
does not show that another is unnecessary. Opportunities to obtain human help, report errors
or access a data notice may vary by service and setting, none of which is specified in the dataset.
Self-report, common item context and unknown selection into the sample may affect the
observed associations.17
3.7 Ethical considerations
Ethical approval for this study was obtained from the Center for Academic & Professional
Career Development and Research (CAPCDR), an external institutional body, under approval
reference CAPCDR: SMUCT/LLB/SS-0020260908-20. External ethical review was sought
because Shanto-Mariam University of Creative Technology does not have an institutional
ethics committee responsible for reviewing this type of research.
4. Results
The results describe responses in the 385 supplied records. They concern stated preferences
and perceptions rather than measured awareness of legal rights among farmers or the effects of
a deployed AI service.
4.1 Respondent profile
Table 1 shows the recorded status and location categories. Neither variable identifies farmers
or agricultural AI users.
Table 1. Recorded respondent status and location
Measure Category Count (%)
Status Other relevant respondent 140 (36.4%)
Law student 136 (35.3%)
Law graduate 58 (15.1%)
Advocate or legal professional 51 (13.2%)
Location Dhaka 177 (46.0%)
Other city corporation 78 (20.3%)
District town 66 (17.1%)
Upazila-rural area 60 (15.6%)
Other 4 (1.0%)
17 Philip M Podsakoff and others, ‘Common Method Biases in Behavioral Research: A Critical Review of the
Literature and Recommended Remedies’ (2003) 88 Journal of Applied Psychology 879; Sijtsma (n 10).
8
N = 385. Status and location are reported as recorded; farming and agricultural AI use are not
identified.
The records include 140 other relevant respondents (36.4%), 136 law students (35.3%), 58 law
graduates (15.1%) and 51 advocates or legal professionals (13.2%). The upazila/rural category
contains 60 respondents (15.6%) and does not separately identify rural residence or agricultural
occupation.
4.2 Advisory information and safeguards
Table 2 reports the five AFR items separately. They measure views about useful information
and different service safeguards, not a validated legal-awareness scale.
Table 2. Agricultural advisory and rights-information items
Item Counts 1 / 2 / 3 / 4 / 5 Endorsed n (%) 95% CI (%)
AFR1 0 / 12 / 129 / 176 / 68 244 (63.4%) 58.5-68.0
AFR2 0 / 4 / 51 / 164 / 166 330 (85.7%) 81.9-88.9
AFR3 0 / 8 / 52 / 178 / 147 325 (84.4%) 80.5-87.7
AFR4 2 / 13 / 123 / 177 / 70 247 (64.2%) 59.2-68.8
AFR5 0 / 3 / 44 / 173 / 165 338 (87.8%) 84.1-90.7
N = 385. Endorsed means agree or strongly agree. Non-endorsement includes neutral
responses. Wilson intervals are pointwise; exact wording is in Appendix A.
Rights-information inclusion was endorsed by 338 respondents (87.8%), knowing the
responsible provider by 330 (85.7%) and communication of uncertainty by 325 (84.4%).
Human help when AI advice is inadequate was endorsed by 247 (64.2%), while 244 (63.4%)
endorsed the usefulness statement. Different statements and response contexts prevent treating
these percentages as a direct ranking of respondents’ needs.
4.3 Reporting and data-use context
Table 3 presents the additional AAR4, ADG2 and AIL1 items used in the exploratory analysis.
These items describe an anticipated reporting route, conditional confidence and self-reported
awareness respectively.
Table 3. Reporting, data-use transparency and AI awareness
Item Counts 1 / 2 / 3 / 4 / 5 Endorsed n (%) 95% CI (%)
AAR4 0 / 9 / 113 / 179 / 84 263 (68.3%) 63.5-72.8
ADG2 0 / 5 / 59 / 167 / 154 321 (83.4%) 79.3-86.8
AIL1 3 / 34 / 131 / 157 / 60 217 (56.4%) 51.4-61.2
9
N = 385. Endorsed means agree or strongly agree. Non-endorsement includes neutral
responses. Wilson intervals are pointwise; exact wording is in Appendix A.
Ways to report harmful or inaccurate recommendations were endorsed by 263 respondents
(68.3%), confidence under transparent data-use practices by 321 (83.4%), and awareness that
AI can produce plausible inaccuracies by 217 (56.4%). Endorsement does not establish that a
reporting channel or transparent data policy exists for an actual service.
4.4 Paired endorsement and joint responses
Table 4 compares three selected item pairs within the same respondents. The differences should
be read with the item wording and the pointwise intervals.
Table 4. Paired endorsement comparisons
Contrast First only / second only Difference pp 95% CI pp Holm p
AFR5 - AFR1 123 / 29 24.4 18.7-30.4 <.001
AFR2 - AFR4 118 / 35 21.6 15.6-27.5 <.001
AFR5 - AFR4 115 / 24 23.6 18.2-29.1 <.001
N = 385; pp = percentage points. Exact McNemar tests; Holm adjustment across three tests.
Intervals are pointwise percentiles from 10,000 respondent-level bootstrap resamples. Items
differ in wording.
Rights-information inclusion exceeded the usefulness statement by 24.4 percentage points;
provider responsibility exceeded human assistance by 21.6 points; and rights information
exceeded human assistance by 23.6 points. The adjusted tests describe different response
distributions. They do not establish the relative effectiveness of any service feature, and neutral
answers are included in non-endorsement.
Table 5 shows the joint distribution of rights-information and usefulness endorsements. The
categories represent answer combinations rather than fixed types of respondents.
Table 5. Joint endorsement of rights information and usefulness
Response configuration Count (%)
Both endorsed 215 (55.8%)
Rights information only 123 (31.9%)
Usefulness only 29 (7.5%)
Neither endorsed 18 (4.7%)
N = 385. Neither includes neutral responses; categories describe answer combinations, not
fixed respondent types.
10
A total of 215 respondents endorsed both statements, while 123 endorsed the rights-information
statement without endorsing the usefulness statement. This pattern does not show whether
rights information changes perceptions of usefulness or whether the respondent has used an
advisory service.
4.5 Exploratory associations
Table 6 presents four Spearman rank correlations with AFR5. The four p-values are Holm
adjusted; confidence intervals are pointwise and are not adjusted for multiplicity.
Table 6. Exploratory associations with rights-information endorsement
Items Spearman ρ 95% CI p Holm p
AFR5 / AFR1 0.047 -0.053-0.149 0.356 0.923
AFR5 / AAR4 0.052 -0.049-0.154 0.308 0.923
AFR5 / ADG2 0.019 -0.086-0.117 0.703 0.923
AFR5 / AIL1 0.095 -0.003-0.192 0.064 0.254
N = 385. Four correlation p-values Holm adjusted. Pointwise percentile intervals use 2,000
respondent-level resamples and are not adjusted for multiplicity.
The estimated correlations are small and none provides clear evidence against a null association
after adjustment. The results do not prove that service usefulness, reporting, transparency or
awareness are unrelated to rights-information preferences. The questionnaire cannot establish
whether a change in any of these factors would affect legal knowledge or a person’s ability to
use a remedy.
5. Discussion
The findings separate support for legal-rights information from perceived usefulness and other
safeguards. Their interpretation is limited to the questionnaire responses and the mixed sample,
not measured outcomes among farmers.
5.1 Rights information and useful advice
A substantial majority endorsed including legal-rights information in responsible services. The
usefulness item received a lower endorsement, and 215 respondents endorsed both. These
patterns are consistent with the possibility that respondents can view rights information and
useful advice as distinct features, but the items are not a forced-choice test. Differences in
wording, specificity or assumptions about agricultural AI could contribute to their
distributions.
The survey does not test whether information is legally accurate, understandable or actionable.
A system that displays a generic statement about rights may still leave a user unable to identify
an applicable rule or obtain help. Future evaluations should specify the rights issue and measure
comprehension, appropriate referral and the user’s ability to act. A broader digital -agriculture
11
literature emphasises the importance of accountability and power over data and service
design.18
5.2 Responsibility and human assistance
The provider-responsibility statement attracted more endorsement than the human-assistance
statement. This does not establish that respondents value a named provider more than access
to a person: one item asks who is responsible for advice and the other specifies help when
advice is inadequate. Both can be necessary, and the questionnaire does not explain how they
would be implemented.
A meaningful human-help pathway should identify who can respond, the type of question they
can address, any cost or delay, and what happens if the problem is outside their expertise. It
should not depend solely on a label that an AI tool is supervised. Responsible service evaluation
requires evidence about whether a user actually reaches an appropriate person.19
5.3 Data transparency, reporting and legal context
The exploratory association between rights-information endorsement and the transparent-data-
use statement was small and uncertain after adjustment. The same is true of the relationship
with the reporting item. The findings cannot be used to claim that transparency or reporting has
no relationship with legal-information preferences. They also cannot show that either safeguard
causes awareness or that a remedy works.20
The Personal Data Protection Act 2026 is relevant where information is identifiable personal
data and the relevant legal provisions apply. Its definitions and rules cannot be applied to an
unspecified service without knowing the data flows and actors. 21 A legal-information feature
should identify its legal scope and date, explain limitations, and offer a route to qualified
assistance rather than treating every agricultural record as personal data.
5.4 Mixed sample and experience
The sample includes law students, graduates, legal professionals and other respondents. Only
60 records are coded in the combined upazila/rural category; that category does not
independently identify farmers or rural residence. Participants may have different knowledge
and experience of agriculture or digital services, and the file does not identify actual
agricultural AI use. Accordingly, the results should not be described as farmers’ legal
knowledge, a rural population estimate or an evaluation of a service.
The numerical records were identified by the researcher as real survey data, while the
Codebook labels identifiers and optional comments synthetic. Documentation should explain
the creation of those fields and the relationship between the workbook and the source
18 Klerkx, Jakku and Labarthe (n 4); Carbonell (n 2).
19 Eastwood and others (n 8); Bhattarai and others (n 1).
20 Altman and Bland (n 16); Wasserstein and Lazar (n 16).
21 Personal Data Protection Act 2026 (n 2); Personal Data Protection Act 2026 (n 9); Personal Data Protection
Act 2026 (n 9).
12
responses. Fieldwork dates and recruitment procedures are also needed for readers to assess
selection and reproducibility.
5.5 Interpreting small associations
The four rank correlations with rights-information endorsement were small and none remained
distinguishable from zero under the chosen Holm adjustment. The intervals allow a range of
possible relationships, and lack of statistical evidence is not evidence of no effect. The items
are individual, cross-sectional responses and do not provide a test of the proposed structural
model.22
The questionnaire does not directly measure the effect of a service on farmers’ rights
awareness. The item-level results should therefore be interpreted as stated perceptions, with
inferences limited to the measures actually collected.
5.6 Integrated interpretation
UTAUT2 highlights anticipated usefulness and support, Trust Theory concerns reliance under
uncertainty, and digital governance directs attention to accountability and rights. 23 The study
does not validate a model connecting them. It offers a focused distinction between including
information about rights, knowing or exercising a right, and receiving useful agricultural
advice. A future study can measure these elements independently in a defined user population.
6. Theoretical and Practical Implications
The following implications are prospective recommendations grounded in the item-level
findings and relevant literature. The survey does not establish that an intervention improves
legal empowerment or agricultural outcomes.
6.1 Theoretical implications
First, rights-information inclusion should be distinguished from legal awareness. A respondent
can favour inclusion without being able to identify or apply a right. A future scale should define
the legal domain and assess comprehension rather than relying on agreement that information
is desirable. Second, usefulness, provider responsibility, data transparency and human help
should be measured as separate concepts before testing whether they form any common
construct.24
Third, the full UTAUT2 framework requires measures of its underlying constructs; the current
instrument does not justify claims that its paths are supported. Trust and governance can be
integrated into a later model if their content and measurement are defined and the study design
supports the intended inference.
22 Wasserstein and Lazar (n 16); Sijtsma (n 10).
23 Venkatesh, Thong and Xu (n 7); Mayer, Davis and Schoorman (n 5); Eastwood and others (n 8).
24 Boateng and others (n 10).
13
6.2 Practical and policy implications
Service providers could identify the organisation responsible for advice, present sources and
uncertainty, and explain relevant data practices in language users can understand. Rights-
related information should be linked to a defined issue and authoritative sources, with the
applicable legal context and date made clear. A functioning route to human assistance or
complaint should be tested with the intended users rather than assumed from an interface.
Agricultural extension bodies and technology providers can evaluate accuracy for local crops
and conditions, accessibility across devices and languages, and whether users reach an
appropriate person after a referral. Studies should include people with different connectivity,
literacy, disabilities and agricultural circumstances. A provider’s legal obligations must be
assessed against the actual service and applicable rules.25
7. Conclusion, Limitations and Future Research
Of 385 recorded respondents, 87.8% endorsed including legal-rights information in responsible
digital agricultural services, 85.7% endorsed knowing who is responsible for important AI
advice, 84.4% endorsed communication of uncertainty and 64.2% endorsed human assistance
when advice is inadequate. Rights-information endorsement exceeded endorsement of the
usefulness statement by 24.4 percentage points; 215 respondents endorsed both. These are
responses to different statements and do not establish the relative effectiveness of service
features or an effect on legal awareness.
The study is limited by cross-sectional self-report, an undocumented sampling and fieldwork
process, a mixed law-oriented sample, no identified farmer or agricultural-AI-user variable,
and no direct test of legal knowledge. The researcher identified the numerical records as real,
while the Codebook labels the identifier and contextual-comment fields synthetic; their
preparation should be documented. No causal or farmer-population claim follows from the
analysis.
Future research should recruit and identify agricultural users and document fieldwork dates,
eligibility and recruitment. It should define specific legal scenarios, measure baseline and
follow-up understanding, and evaluate whether a user can locate and use a relevant remedy. A
service study could vary provider identity, uncertainty disclosure, data information and human
referral, then measure comprehension, referral completion, accuracy and practical outcomes.
Interviews and accessible user testing would help reveal barriers that general agreement items
cannot capture.
Declarations
Ethical approval and oversight
Ethical approval for this study was obtained from the Center for Academic & Professional
Career Development and Research (CAPCDR), an external institutional body, under approval
reference CAPCDR: SMUCT/LLB/SS-0020260908-20. External ethical review was sought
25 Bhattarai and others (n 1); Eastwood and others (n 8).
14
because Shanto-Mariam University of Creative Technology does not have an institutional
ethics committee responsible for reviewing this type of research.
Informed consent
Informed consent was obtained from adult participants before questionnaire completion. The
questionnaire explained the study purpose, voluntary participation, anonymity and the option
to discontinue. All analysed records contain affirmative consent and adult-eligibility responses.
Data availability
The numerical data supporting the reported analyses are held by the research team. Requests
for de-identified data and analytical specifications may be directed to the corresponding author.
Any sharing is subject to participant confidentiality and applicable institutional requirements.
The contextual-text field was excluded from the analysis.
Conflict of interest
The authors declare that they have no conflicts of interest.
Author contributions
Sumtahin Pushpita: Lead student researcher; conceptualisation; preparation of study
materials; organisation and curation of survey data; manuscript development. The project was
conducted under the supervision of Prof. Dr Kazi Abdul Mannan.
Kazi Abdul Mannan: Conceptual and methodological guidance; supervision; interpretation
of statistical findings; critical review and editing of the manuscript.
Generative AI disclosure
Generative AI assisted manuscript drafting, language editing, organisation and the
development of analytical code. Statistical estimates were computed programmatically from
the numerical survey data. Responsibility for the study’s analysis, interpretation and reporting
rests with the authors.
Bibliography
Altman DG and Bland JM, ‘Absence of Evidence Is Not Evidence of Absence’ (1995) 311
BMJ 485
Bhattarai R and others, ‘Scale, Trust, and the Digital Divide: A Systematic Review of AI and
ML for Agricultural Applications’ (2026) 9 Frontiers in Artificial Intelligence 1798896
Boateng GO and others, ‘Best Practices for Developing and Validating Scales for Health,
Social, and Behavioral Research: A Primer’ (2018) 6 Frontiers in Public Health 149
Carbonell IM, ‘The Ethics of Big Data in Big Agriculture’ (2016) 5(1) Internet Policy
Review, DOI: 10.14763/2016.1.405
Cochran WG, Sampling Techniques (3rd edn, John Wiley & Sons 1977)
Eastwood C and others, ‘Managing Socio-Ethical Challenges in the Development of Smart
Farming: From a Fragmented to a Comprehensive Approach for Responsible Research and
Innovation’ (2019) 32 Journal of Agricultural and Environmental Ethics 741
15
Efron B and Tibshirani RJ, An Introduction to the Bootstrap (Chapman & Hall 1993)
Holm S, ‘A Simple Sequentially Rejective Multiple Test Procedure’ (1979) 6 Scandinavian
Journal of Statistics 65
Klerkx L, Jakku E and Labarthe P, ‘A Review of Social Science on Digital Agriculture,
Smart Farming and Agriculture 4.0: New Contributions and a Future Research Agenda’
(2019) 90-91 NJAS: Wageningen Journal of Life Sciences 100315
Mayer RC, Davis JH and Schoorman FD, ‘An Integrative Model of Organizational Trust’
(1995) 20 Academy of Management Review 709
McNemar Q, ‘Note on the Sampling Error of the Difference between Correlated Proportions
or Percentages’ (1947) 12 Psychometrika 153
Podsakoff PM and others, ‘Common Method Biases in Behavioral Research: A Critical
Review of the Literature and Recommended Remedies’ (2003) 88 Journal of Applied
Psychology 879
Sijtsma K, ‘On the Use, the Misuse, and the Very Limited Usefulness of Cronbach’s Alpha’
(2009) 74 Psychometrika 107
Singh A, Shukla VK and Singh C, ‘From Pilots to Plots: A Critical Review of Artificial
Intelligence for Smallholder Agriculture in South Asia’ (2026) DOI:
10.1177/00307270261489284
Venkatesh V, Thong JYL and Xu X, ‘Consumer Acceptance and Use of Information
Technology: Extending the Unified Theory of Acceptance and Use of Technology’ (2012) 36
MIS Quarterly 157
von Elm E and others, ‘The Strengthening the Reporting of Observational Studies in
Epidemiology (STROBE) Statement: Guidelines for Reporting Observational Studies’ (2007)
4 PLOS Medicine e296
Wasserstein RL and Lazar NA, ‘The ASA Statement on p-Values: Context, Process, and
Purpose’ (2016) 70 American Statistician 129
Appendix A. Analytical questionnaire items
Table A1 reproduces the survey statements so that each estimate can be interpreted against its
exact wording.
16
Table A1. Exact wording of analytical items
Code Questionnaire statement
AFR1 AI agricultural advisory services can help farmers access useful information.
AFR2 Farmers should know who is responsible for important advice delivered through AI systems.
AFR3 AI advisory services should clearly communicate uncertainty and limitations.
AFR4 Farmers should have access to human assistance when AI advice is inadequate.
AFR5 Legal-rights information should be incorporated into responsible digital agricultural services.
AAR4 Users should have ways to report harmful or inaccurate recommendations.
ADG2 I would use AI agricultural services more confidently if data-use practices were transparent.
AIL1 I understand that generative AI can produce plausible but inaccurate information.
Source: Group 16 Codebook. Answers range from strongly disagree to strongly agree.
Items are analysed separately and do not establish a single validated latent construct.
Table of legislation
Personal Data Protection Act 2026 (Bangladesh Act No 63 of 2026)
论文原文与出版信息
Research Square