Research
Luciano Floridi & Josh Cowls
Harvard Data Science Review, 1(1), 2019
Paraphrase: surveys several prominent sets of AI ethics principles and finds substantial convergence around the four classic bioethics principles, adding explicability (intelligibility and accountability) as a fifth that enables the others.
SourceTom L. Beauchamp & James F. Childress
Oxford University Press, first ed. 1979; 8th ed. 2019
Paraphrase: the foundational account of respect for autonomy, non-maleficence, beneficence and justice as mid-level principles for clinical and research ethics.
SourceAnna Jobin, Marcello Ienca & Effy Vayena
Nature Machine Intelligence, 1, 389–399, 2019
Paraphrase: maps 84 guideline documents and reports convergence on themes such as transparency, justice and non-maleficence, alongside real divergence in how principles are interpreted and implemented.
SourceBrent Mittelstadt
Nature Machine Intelligence, 1, 501–507, 2019
Paraphrase: a critique arguing that the medical-ethics analogy is limited for AI, because AI development lacks medicine's shared goals, professional history and accountability mechanisms.
SourceConvergence and disagreement
Reviews of guideline documents find recurring themes. Floridi and Cowls group them as:
Agentic AI
Paraphrased from Prof. Stephanie Dick (Simon Fraser University; co-editor, Harvard Data Science Review), Harvard Data Science Initiative session on ethics and moral reasoning in agentic AI.
Factories automated physical work and computers automated mental work; agentic AI aims to automate the capacity to decide, initiate and act, which demands judgement and self-regulation.
Following Hume, moral reasoning always needs a step beyond the facts, taken by principles the data cannot supply. Consequentialism, deontology and virtue ethics give different answers.
Different ethical frameworks imply different liabilities and customer expectations, and different stakeholders judge an AI's choices differently.
Research by Atari, Xue, Park, Blasi and Henrich finds language-model responses resemble some populations' values far more than others, so a model's built-in ethics is not neutral.
Responsibility is enacted across workflows, human oversight, technical constraints, audits, refusal conditions and escalation paths.
Audits surface how responsibility, risk and harm are allocated by a system in operation.
Get involved
Most professional societies take open individual membership with yearly dues; research labs and journals are joined by contributing work. Terms change, so confirm on each site.
Bioethics
Open to anyone with an interest in healthcare ethics. Yearly dues are based on income, from $75 (under $35,000) to $475. Members can vote, hold office and attend the annual conference.
VisitBioethics
Open individual membership: $70 (high-income countries), $60 (middle-income), $25 (low-income countries and students), $35 (early-career researchers). Members can register for the World Congress of Bioethics and vote for the board.
VisitBioethics
Fellows are elected for distinguished work; anyone can subscribe to the Hastings Center Report, attend events, or apply for visiting scholar programs.
VisitAI & data science
Join through its executive education courses, public events and newsletter; affiliate status is generally for Harvard faculty and researchers.
VisitJournal
Open access to read; contribute by submitting articles or volunteering as a reviewer. No membership required.
VisitAI & normative theory
Research lab led by Seth Lazar, now a professor at Johns Hopkins University. Engage via its public seminars and publications, or apply for posted PhD and postdoc positions.
VisitAI fairness & accountability
No membership needed to attend or submit to the annual conference; ACM membership is optional.
VisitAI ethics
AAAI offers open membership for professionals and students, with discounts on conference registration; the AI, Ethics and Society conference accepts open submissions.
Visit