How to Design AI for Social Good: Seven Essential Factors
Open accessThe DOI matches its Crossref record; 73 of the 75 references it cites were found in a registry or an index; 9 of the 80 sources its claims cite were checked for support.
Abstract
The idea of artificial intelligence for social good (henceforth AI4SG) is gaining traction within information societies in general and the AI community in particular. It has the potential to tackle social problems through the development of AI-based solutions. Yet, to date, there is only limited understanding of what makes AI socially good in theory, what counts as AI4SG in practice, and how to reproduce its initial successes in terms of policies. This article addresses this gap by identifying seven ethical factors that are essential for future AI4SG initiatives. The analysis is supported by 27 case examples of AI4SG projects. Some of these factors are almost entirely novel to AI, while the significance of other factors is heightened by the use of AI. From each of these factors, corresponding best practices are formulated which, subject to context and balance, may serve as preliminary guidelines to ensure that well-designed AI is more likely to serve the social good.
Overview
CoChat Summary
TL;DR
This article identifies seven ethical factors essential for designing AI for social good (AI4SG), supported by 27 case examples and seven representative projects. It derives corresponding best practices, while stressing that the factors are necessary but not sufficient and must be balanced contextually.
Key findings
- The paper proposes seven essential AI4SG factors: (1) falsifiability and incremental deployment; (2) safeguards against manipulation of predictors; (3) receiver-contextualised intervention; (4) receiver-contextualised explanation and transparent purposes; (5) privacy protection and data subject consent; (6) situational fairness; and (7) human-friendly semanticisation.
- The analysis is supported by 27 AI4SG case examples, from which 7 projects were selected as most representative in terms of scope, variety, impact, and ability to corroborate the factors.
- Each factor is linked to at least one of five ethical principles—beneficence, nonmaleficence, justice, autonomy, and explicability—with beneficence treated as a necessary precondition for AI4SG.
- The authors argue that the seven factors are not a checklist guaranteeing socially good outcomes; they require proactive consideration and contextual balancing, both within and between factors.
Methods
- The authors conducted a systematic literature review across five databases—Google Scholar, PhilPapers, Scopus, SSRN, and Web of Science—between October 2018 and May 2019.
- They analysed 27 projects identified as successful or unsuccessful AI4SG examples, then selected 7 representative cases for detailed discussion.
- The analysis is conceptual and normative: the authors derive seven factors and corresponding best practices, mapping each factor to broader AI ethics principles.
Limitations
- Stated in the paper: the seven factors are not sufficient for socially good AI, should not be treated as a checklist, and require context-dependent balancing; the article does not address how factors should be evaluated, by whom, or with what supporting mechanisms such as regulation or codes of conduct.
- Stated in the paper: the analysis is not meant to cover all potential contexts and does not answer in the abstract whether one is morally obliged to design, develop, or deploy a specific AI4SG project.
- Evident from the text: the proposed best practices are not empirically tested or validated as producing socially good outcomes; the detailed discussion relies on 7 representative cases from a wider sample of 27, so the findings are conceptual guidance rather than causal or quantitative evidence.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
9 Valid · 0 Not valid · 63 Not checked · 90 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Furthermore, many projects that generate socially good outcomes using AI are not (self-)described as such (Moore 2019).”
§ How to Design AI for Social Good: Seven Essential Factors > Introduction
What the sources say
- Partly supportedcites Moore 2019 sourceExperimental — accuracy not yet measured
In 'The good and bad of "AI for good"' Moore states the point in the claim's direction — the section's argument list includes 'Following 1, there are projects that are good, but that are not labeled as such,' and the body says 'there are projects which do not use the label "AI for the good" that might be classified as such,' illustrated by federated learning, which 'does not carry the moniker social good.' However, the source asserts existence, not quantity: nothing in our copy supports 'many' projects. It also frames these as conditional/hypothetical ('might be classified as such', 'one can imagine a fully-specified federated learning project that meets the criteria of 1', a imagined smartphone birth-defect screening project) rather than as documented cases that 'generate socially good outcomes', and 'good' there is relative to Moore's own capability-approach definition (criteria 1), not a neutral description of outcomes. (LLM, conf=0.72)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Tensions may arise between different thresholds of consent (Price and Cohen 2019).”
§ How to Design AI for Social Good: Seven Essential Factors > Seven Essential Factors for Successful AI4SG > Privacy Protection and Data Subject Consent
What the sources say
- Supportedcites Price and Cohen 2019 source
In “The role of the patient in data collection and access” (page unavailable), the paper contrasts approval for every data access with broad consent and intermediate governance models. It explains that requiring individual approval would maximize autonomy but could eliminate much big-health-data work, supporting the claim that different consent thresholds may create tensions. (LLM, conf=0.87)
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Moreover, AI-driven decisions may be amalgamated from factors that are not of obvious ethical importance, and yet collectively constitute unfairly biased decision-making (Pedreshi et al. 2008; Floridi 2012).”
§ How to Design AI for Social Good: Seven Essential Factors > Seven Essential Factors for Successful AI4SG > Situational fairness
What the sources say
- Not checked yetcites Pedreshi et al. 2008
- Partly supportedcites Floridi 2012 sourceExperimental — accuracy not yet measured
Our copy of Floridi (2012/2013) supports the structural half of the claim: it argues that actions 'morally negligible in themselves may become morally significant, if properly aggregated' (The Big Challenge: Harnessing the Power of DM), that aggregation of neutral courses of action can produce moral evil as well as good for artificial/hybrid multiagent systems including webbots (The New Scenario with Distributed Morality), and that MAS' individually negligible actions can combine into morally loaded outcomes evaluated from a receiver's perspective. However, no passage in our copy concerns AI-driven or algorithmic decision-making being built from ethically irrelevant variables that collectively yield UNFAIRLY BIASED outcomes: searches for bias, fairness, discrimination, unfairness and data mining returned nothing on that point (the paper's negative examples are the tragedy of the commons and P2P/network-neutrality cases). That specific bias claim is not stated by this source and is presumably carried by the co-citation Pedreshi et al. 2008. (LLM, conf=0.72)
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Every source the paper cites for this claim was checked and supports it.
In the paper
“Consequently, researchers have developed AI software that balances reminding the patient against annoying the patient (Chu et al. 2012).”
§ How to Design AI for Social Good: Seven Essential Factors > Seven Essential Factors for Successful AI4SG > Human-Friendly Semanticisation
What the sources say
- Supportedcites Chu et al. 2012 source
Section 3.2.2 (Adaptive options) states the trade-off explicitly: an optimal strategy avoids prompting too soon and too late, and 'the model is trading off between two objectives: trying to avoid unnecessary prompts and ensuring the task occurs in time', with bad prompts noted as able to 'frustrate the user'. The Introduction (p. 1) frames the same balance — minimizing 'the cost of interruptions' and training the system 'to be less intrusive so that it delivers proper prompts only when needed' — and the prompt cost c enters the expected-reward equation (Eq. 1), with experiments penalizing each prompt (-2) against task failure (-10) and success (+10) in Section 4.1. The software is the paper's own hierarchical POMDP/MDP prompting system with Q-learning for people with cognitive disabilities, so the identification holds; note only that its human evaluation (Section 5) used volunteer student actors rather than patients. (LLM, conf=0.85)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
In the paper
“At the same time, the genuinely socially good outcomes of AI may arise merely by chance, for example through an accidental application of an AI solution in a different context.”
§ How to Design AI for Social Good: Seven Essential Factors > Introduction
What the sources say
- Not checked yetcites Goel et al. 2015
hedgedstates a possible causeSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
54 more claims
Create a free account to see all 162 claims and to verify the 71 references still open.
- It may state two separate claims as one. 51
- Probably not a substantial claim. 64
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Although AI can augment or surpass human abilities by identifying, interpreting, making inferences, and learning from data to achieve predetermined organizational and societal goals (Mikalef & Gupta, 2021), it provides suggestions that are by nature uncertain as it identifies correlation relationships (Floridi et al., 2020).”
From Responsible AI for Digital Health: a Synthesis and a Research Agenda (2023) · cites this paper as Floridi et al., 2020
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“In these cases, AI is subject to significant uncertainty, which called for precautionary and safety principles in the data collection and analysis for decision making (Floridi et al., 2020).”
From Responsible AI for Digital Health: a Synthesis and a Research Agenda (2023) · cites this paper as Floridi et al., 2020
The cited source says what the claim attributes to it.
In the citing paper
“For example, focusing on non-causal indicators may distract attention from the underlying causes of a given problem (Floridi et al. 2020).”
From The ethics of algorithms: key problems and solutions (2022) · cites this paper as Floridi et al. 2020
The cited source says what the claim attributes to it.
In the citing paper
“Transparency can enable individuals to game the system (Martin 2019;Magalhães 2018;Floridi et al. 2020).”
From The ethics of algorithms: key problems and solutions (2022) · cites this paper as Floridi et al. 2020
In the citing paper
“However, when there is a lack of understanding of the exact data used and a lack of transparency and interpretability of the processes followed to generate those results (Floridi et al., 2020), AI provides inscrutable evidence.”
From Responsible AI for Digital Health: a Synthesis and a Research Agenda (2023) · cites this paper as Floridi et al., 2020
2 more claims about this paper
Create a free account to see every claim other papers make about this one.
Citations
Cites
Floridi, L., Cowls, J., Beltrametti, M., Chatila, R., Chazerand, P., Dignum, V., Luetge, C., Madelin, R., Pagallo, U., Rossi, F., et al. (2018) · Minds and Machines · doi:10.1007/s11023-018-9482-5 · Cited by 4,774 · cited 5×
“Each factor relates to at least one of five ethical principles of AI— beneficence , nonmaleficence , justice , autonomy , and explica-bility —identified in the comparative analysis mentioned above (Floridi et al. 2018).”
BackgroundInfluentialSelf-citationSemantics derived automatically from language corpora contain human-like biases
Caliskan, A., Bryson, J. J., & Narayanan, A. (2017) · Science · doi:10.1126/science.aal4230 · Cited by 2,887 · cited 2×
“Training datasets may contain natural language that carries unfair associations between genders and words which, in turn, carry normative power (Caliskan et al. 2017).”
BackgroundPrivacy in Context: Technology, Policy, and the Integrity of Social Life
Nissenbaum, H. (2009) · doi:10.1515/9780804772891 · Cited by 2,129 · cited 1×
“…is considered to be an essential condition for safety, human dignity, and social cohesion, among other things (Solove 2008), and because earlier waves of digital technology have already had a major impact 1 3 How to Design AI for Social Good: Seven Essential Factors on privacy (Nissenbaum 2009).”
BackgroundExplaining collaborative filtering recommendations
Herlocker, J. L., Konstan, J. A., & Riedl, J. (2000) · Proceedings of the 2000 ACM conference on Computer supported cooperative work · doi:10.1145/358916.358995 · Cited by 1,731 · cited 1×
“Explainability breeds trust in, and fosters adoption of, AI4SG solutions (Herlocker et al. 2000; Swearingen and Sinha 2002; Bilgic and Mooney 2005).”
BackgroundA Unified Framework of Five Principles for AI in Society
Floridi, L., & Cowls, J. (2019) · Harvard Data Science Review · doi:10.1162/99608f92.8cd550d1 · Cited by 1,631 · cited 1×
“The seven factors analysed in the previous pages are summarised in Table 1, together with the corresponding best practices, and the five principle(s) of AI ethics identified in (Floridi and Cowls 2019) to which each factor is most closely identified.”
BackgroundSelf-citationPrivacy in the age of medical big data
Price, W. N., & Cohen, I. G. (2019) · Nature Medicine · doi:10.1038/s41591-018-0272-7 · Cited by 1,522 · cited 1×
“Tensions may arise between different thresholds of consent (Price and Cohen 2019).”
BackgroundA model of inexact reasoning in medicine
Shortliffe, E. H., & Buchanan, B. G. (1975) · Mathematical Biosciences · doi:10.1016/0025-5564(75)90047-4 · Cited by 1,447 · cited 1×
“It has been a focus of research since at least 1975 (Shortliffe and Buchanan 1975).”
BackgroundWachter, S., Mittelstadt, B., & Floridi, L. (2017) · International Data Privacy Law · doi:10.1093/idpl/ipx005 · Cited by 1,308 · cited 1×
“This is likely to include information about both the general functionality and logic employed by a system and the reasons for the specific decision being made (Wachter et al. 2017).”
BackgroundSelf-citationFriedman, B., & Nissenbaum, H. (1996) · ACM Transactions on Information Systems · doi:10.1145/230538.230561 · Cited by 1,203 · cited 1×
“These decisions may be based on factors of ethical importance (e.g., ethnic, gender, or religious grounds) and irrelevant to the decision-making at hand, or they may be relevant but legally protected as a nondiscriminatory characteristic (Friedman and Nissenbaum 1996).”
BackgroundThe grand challenges of Science Robotics
Yang, G.-Z., Bellingham, J., Dupont, P. E., Fischer, P., Floridi, L., Full, R., Jacobstein, N., Kumar, V., McNutt, M., Merrifield, R., et al. (2018) · Science Robotics · doi:10.1126/scirobotics.aar7650 · Cited by 1,177 · cited 2×
“…them is “more important” than any other, so we shall introduce them not in terms of priority, but somewhat historically, starting with factors that pre-date AI, and yet take on greater importance when AI technologies are used, owing to the particular capabilities and risks of AI (Yang et al. 2018).”
BackgroundSelf-citation
Research works printed but not linked: 2
- Thelisson, E., Padh, K., & Elisa Celis, L. (2017). Regulatory mechanisms and algorithms towards trust in AI/ML. In Proceedings of the IJCAI 2017 workshop on explainable artificial intelligence (XAI), Melbourne, Australia.
- Wachter, S., Mittelstadt, B., & Floridi, L. (2016). Why a right to explanation of automated decision-making does not exist in the general data protection regulation. SSRN Scholarly Paper ID 2903469. Rochester, NY: Social Science Research Network. https://papers.ssrn.com/abstract=2903469.
Cited by
Privacy in the age of medical big data
Price, W. N., & Cohen, I. G. (2019) · Nature Medicine · doi:10.1038/s41591-018-0272-7 · Cited by 1,522
Morley, J., Floridi, L., Kinsey, L., & Elhalal, A. (2020) · Science and Engineering Ethics · doi:10.1007/s11948-019-00165-5 · Cited by 882
Self-citationTrustworthy Artificial Intelligence: A Review
Kaur, D., Uslu, S., Rittichier, K. J., & Durresi, A. (2023) · ACM Computing Surveys · doi:10.1145/3491209 · Cited by 709
“Over the past few years, research institutes, private organizations, and government agencies have proposed various guidelines and frameworks to make AI trustworthy [60, 68, 69, 154, 198, 224].”
BackgroundBibri, S. E., Krogstie, J., Kaboli, A., & Alahi, A. (2024) · Environmental Science and Ecotechnology · doi:10.1016/j.ese.2023.100330 · Cited by 536
“[63,244,299,300]) in urban life domains in emerging smarter eco-cities.”
BackgroundThe Ethics of Artificial Intelligence: Principles, Challenges, and Opportunities
Floridi, L. (2023) · doi:10.1093/oso/9780198883098.001.0001 · Cited by 507
Self-citation
Metadata
- Authors
- Floridi, L., Cowls, J., King, T. C., & Taddeo, M.
- Affiliations
- University of Oxford (GB)
- Corresponding author
- Cowls, Josh
- Venue
- Science and Engineering Ethics
- Published
- 2020-04-03
- Volume
- 26
- Issue
- 3
- Pages
- 1771-1796
- ISSN
- 1353-3452 (ISSN-L)
- 1471-5546
- PMC
- PMC7286860
- OpenAlex
- W3014499801
- Cited by
- 430
- References
- 75
- Citations by year
- 2026: 82 · 2025: 80 · 2024: 79 · 2023: 70 · 2022: 50 · 2021: 54 · 2020: 13
- FWCI
- 63.63
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- Science and Engineering Ethics (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- PhilPapers (PhilPapers Foundation) (submittedVersion)
- Funders
- The Alan Turing Institute (Doctoral Studentship)
- Google UK Ltd
- Engineering and Physical Sciences Research Council (EP/N023013/1)
- Facebook Inc.
- Microsoft
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Social Sciences › Sociology and Political Science › Innovation, Technology, and Society
- Physical Sciences › Computer Science › Artificial Intelligence › Artificial Intelligence Applications
- Keywords
- AI ethics, design factors, best practices, social problems, AI policy, moral norms
- MeSH terms
- Artificial Intelligence, Humans, Morals
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (qwen3.8-omni-flash and gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: How to Design AI for Social Good: Seven Essential Factors. https://app.cochat.ai/papers/10.1007/s11948-020-00213-5
- The paper
- Floridi, L., Cowls, J., King, T. C., & Taddeo, M. (2020). How to Design AI for Social Good: Seven Essential Factors. Science and Engineering Ethics. https://doi.org/10.1007/s11948-020-00213-5
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.