Artificial Intelligence (AI): Multidisciplinary perspectives on emerging challenges, opportunities, and agenda for research, practice and policy

This is not a controlled experiment. The paper is a **commentary and agenda-setting piece** that collected expert perspectives on AI's emerging impact. There is no intervention, no comparator group, and no outcome measure in the traditional sense. Instead, the "test" was a structured expert elicitation process:

**Intervention:** None. The authors convened a panel of 22 experts and asked them to contribute short position pieces on AI opportunities and challenges within their domains.

**Comparators:** None. This is a single-group, qualitative synthesis of expert opinion.

**Outcome measures:** Thematic categories of opportunities, challenges, and research/policy priorities, as identified by the experts and synthesised by the lead authors.

**Sample size:** 22 expert contributors.

**Population:** Academics (from business, information systems, computer science, public policy), industry practitioners (from technology, consulting, finance, healthcare), and government/policy advisors.

**Setting:** The contributors were invited by the lead authors to submit short essays or position statements. The paper does not specify a formal recruitment process, inclusion/exclusion criteria, or demographic details (age, gender, years of experience, geographic distribution). The contributors are listed in the acknowledgements but not systematically described.

No instruments or scales were used. The "measurement" was a qualitative thematic analysis:

Each expert submitted a written contribution (length not specified) on AI opportunities and challenges within their domain.

The lead authors (Dwivedi, Hughes, Ismagilova, Aarts, Coombs, Crick) read all contributions and identified recurring themes.

Themes were grouped into four broad domains: business and management, government and public sector, science and technology, and cross-cutting societal issues.

No quantitative coding, inter-rater reliability, or statistical analysis was performed. The synthesis is narrative and interpretive.

**Study design:** This is a **qualitative expert commentary**—sometimes called a "perspectives piece" or "agenda-setting article." It is not a systematic review, meta-analysis, randomised trial, or observational study. The authors explicitly state they "bring together the collective insight from a number of leading expert contributors."

**How it was conducted:**

1. The lead authors identified and invited 22 experts from their professional networks.

2. Each expert submitted a written contribution (likely 1–3 pages) on AI opportunities and challenges.

3. The lead authors read all contributions and synthesised them into a narrative organised by domain.

4. The final paper includes direct quotes from some contributors but does not report how many quotes were used, how disagreements were resolved, or whether any themes were excluded.

**What this design can and cannot prove:**

**Can prove:** That a group of selected experts holds certain views about AI. It can identify a range of opinions and highlight areas of consensus or disagreement among this specific panel.

**Cannot prove:** That these views are representative of the broader AI research community, industry, or public. It cannot quantify the prevalence of any opinion, test a hypothesis, or establish causal relationships. It cannot provide effect sizes, confidence intervals, or statistical significance. It is not a systematic review, so it does not claim to have searched all relevant literature or applied quality filters.

**Major methodological weaknesses:**

**Selection bias:** Experts were chosen by the lead authors from their own networks. This likely over-represents certain institutions, countries (likely UK-centric given author affiliations), and perspectives.

**No systematic search:** Unlike a systematic review, there was no pre-registered protocol, no search strategy, no duplicate screening, and no quality assessment of the contributions.

**No quantitative analysis:** The synthesis is entirely narrative. There is no way to know which themes were mentioned by 1 expert versus 20 experts.

**No conflict of interest disclosure for contributors:** The paper does not report whether any contributors had financial ties to AI companies, which could influence their views.

**No replication:** The synthesis is a single, unrepeatable process. Another team with different experts would likely produce different themes.

The paper organises findings into four domains. Below are the main themes reported, with the caveat that no frequencies, effect sizes, or statistical tests are provided.

**Business and management:**

AI will automate routine cognitive tasks (e.g., data entry, customer service) and augment complex decision-making (e.g., credit scoring, fraud detection).

Key opportunities: cost reduction, personalisation, predictive analytics.

Key challenges: data quality and availability, lack of AI talent, integration with legacy systems, ethical concerns around bias and transparency.

The authors state that "the impact of AI could be significant" across finance, healthcare, manufacturing, retail, supply chain, logistics, and utilities, but provide no quantitative estimates of impact magnitude.

**Government and public sector:**

AI can improve public service delivery (e.g., chatbots for citizen queries, predictive policing, traffic management).

Key challenges: regulatory frameworks lag behind technological capability, privacy concerns, risk of algorithmic bias against marginalised groups, lack of digital infrastructure in some regions.

The authors note that "trust is a critical barrier" to AI adoption in government, but do not report any survey data on trust levels.

**Science and technology:**

AI accelerates scientific discovery (e.g., drug design, materials science, climate modelling).

Key challenges: reproducibility of AI-driven results, need for explainable AI, energy consumption of large models, potential for dual-use (e.g., autonomous weapons).

The authors highlight that "the pace of change is staggering" but do not provide any bibliometric or patent data to support this claim.

**Cross-cutting societal issues:**

Workforce displacement: AI will eliminate some jobs and create others, but the net effect is uncertain. The authors mention that "reskilling and upskilling" are essential but offer no estimates of job displacement numbers.

Inequality: AI could widen the gap between countries and individuals with access to AI and those without.

Ethics and governance: Need for "responsible AI" frameworks, but no specific proposals are evaluated.

This paper does not report any effect sizes. The "effect" is the set of expert opinions, and its magnitude cannot be quantified. The authors' claim that AI's impact "could be significant" is a qualitative judgement, not a measured outcome. For context, other sources (e.g., McKinsey Global Institute, 2017) estimate that AI could add $13 trillion to global GDP by 2030, but this paper does not cite or replicate such estimates.

**Acknowledged by authors:**

The paper is explicitly positioned as an "agenda for research, practice and policy," not as a definitive empirical study.

The authors note that "the pace of change is staggering" and that their insights may become outdated quickly.

They call for "further research" on most topics, implicitly acknowledging the preliminary nature of their synthesis.

**Critical reader observations:**

**No systematic methodology:** The paper does not meet the standards of a systematic review (PRISMA) or a Delphi study. It is essentially an edited collection of invited essays.

**Selection bias:** All 22 experts were chosen by the lead authors. The paper does not report how many were invited versus how many declined, nor does it describe the expertise distribution (e.g., how many were computer scientists versus business professors versus policymakers).

**No quantitative data:** The paper contains zero numbers—no sample sizes, no percentages, no p-values, no confidence intervals. This makes it impossible to assess the strength of any claim.

**No conflict of interest reporting:** The paper does not disclose whether any contributors had financial interests in AI companies (e.g., as consultants, board members, or shareholders).

**Geographic bias:** All lead authors are based in the UK (Swansea University, Loughborough University, University of Bradford, University of Leicester). The expert list appears UK-heavy, which may limit generalisability to other regions.

**No time horizon:** The paper discusses "emerging challenges" but does not specify whether these are expected in 1 year, 5 years, or 20 years.

**No comparison to alternative views:** The paper does not engage with critics who argue that AI's impact is overhyped (e.g., the "AI winter" perspective) or that current AI is narrow and brittle.

For someone running their own n=1 experiment, this paper offers no direct experimental findings to replicate. However, you can use its themes to design personal experiments around AI tools. Below are actionable suggestions based on the paper's identified opportunities and challenges.

**What to test:**

Test whether an AI-powered productivity tool (e.g., a writing assistant like Grammarly, a scheduling assistant like x.ai, or a code completion tool like GitHub Copilot) actually improves your output quality or speed compared to your usual workflow.

Test whether using an AI chatbot for customer service (e.g., in a small business) reduces response time or customer satisfaction compared to human-only service.

**Minimum meaningful duration:**

For productivity tools: 2–4 weeks of daily use, with a 1-week baseline period before the intervention.

For customer service: 4–8 weeks to account for seasonal variation in customer inquiries.

**What to measure (specific metrics):**

**For writing tools:** Words written per hour, error rate (typos/grammar mistakes), subjective writing ease (1–10 scale), final document quality score (e.g., from a blind reviewer).

**For scheduling tools:** Time spent scheduling per week (minutes), number of scheduling errors (double-bookings, missed meetings), user satisfaction (1–10 scale).

**For customer service:** Average response time (minutes), customer satisfaction score (1–5 scale), resolution rate (percentage of issues resolved on first contact).

**Key confounds to control for:**

**Learning curve:** AI tools often require a setup and learning period. Run the experiment long enough (at least 2 weeks) to move past the initial friction.

**Task difficulty:** If you test a writing tool, ensure the tasks are comparable across baseline and intervention periods (e.g., same type of document, same length).

**Time of day:** Productivity varies by time of day. If possible, use the AI tool at the same time each day, or randomise which tasks get AI assistance.

**Placebo effect:** Simply trying a new tool can make you feel more productive. Consider a blinded design where you don't know whether the AI is active (e.g., use a tool with an on/off toggle that you don't control).

**Tool quality:** Not all AI tools are equal. Choose a well-reviewed tool and use the same version throughout.

**What a positive result would look like:**

For writing tools: A 15–20% increase in words written per hour AND a 1–2 point improvement in subjective ease (on a 10-point scale), with no decrease in quality.

For scheduling tools: A 30–50% reduction in time spent scheduling per week, with fewer than 1 error per month.

For customer service: A 20–30% reduction in average response time, with no decrease in customer satisfaction (or a <0.5 point drop on a 5-point scale).

**Confounds specific to this paper's themes:**

**Trust in AI:** The paper highlights trust as a barrier. In your n=1 experiment, measure your own trust in the tool (e.g., "How confident are you that the AI made the correct decision?" on a 1–10 scale) and see if it correlates with performance.

**Bias awareness:** If you test an AI tool for hiring or screening (e.g., resume parser), check whether it systematically favours or penalises certain demographics. This is hard to test in an n=1, but you can compare the AI's output to your own judgement for a set of test cases.

**Over-reliance:** The paper warns about automation bias. Track whether you accept the AI's suggestions without checking them. A positive result would be that you use the AI as a tool, not a crutch—i.e., you still catch errors the AI misses.

**Bottom line:** This paper is not a source of experimental data. Use it as a brainstorming tool to identify which AI applications are worth testing in your own life, but do not treat its claims as evidence. For actual effect sizes, look to empirical studies (e.g., randomised trials of AI in radiology, customer service, or writing).