Professional Documents
Culture Documents
How AI can be a force for good to improve the systems to avoid mistakes and
mitigate the risks of misuse. To be success-
ful, similar projects must include an ethical
An ethical framework will help to harness the potential impact analysis from the beginning, to assess
AI’s benefits and risks and define guiding
of AI while keeping humans in control principles for an ethically sound design and
use of AI.
By Mariarosaria Taddeo1,2,3 to health care, understanding which tasks The effects of decisions or actions based
and Luciano Floridi1,2 and decisions to entrust (delegate) to AI and on AI are often the result of countless in-
how to ascribe responsibility for its perfor- teractions among many actors, including
A
rtificial intelligence (AI) is not just a mance are pressing ethical problems. At the designers, developers, users, software, and
new technology that requires regula- same time, as AI becomes invisibly ubiqui- hardware. This is known as distributed
tion. It is a powerful force that is re- tous, new ethical challenges emerge. The pro- agency (10). With distributed agency comes
shaping daily practices, personal and tection of human self-determination is one distributed responsibility. Existing ethi-
professional interactions, and environ- of the most relevant and must be addressed cal frameworks address individual, human
The ethical debate on AI as a new form of has become infamous. Robust procedures sis. In 2017, the International Federation of
agency dates to the 1960s (2, 4). Since then, for human oversight are needed to minimize Robotics suggested that by 2020, more than
many of the relevant problems have con- such unintended consequences and redress 1.7 million new AI-powered robots will be
cerned delegation and responsibility. As AI is any unfair impacts of AI. installed in factories worldwide. In the same
used in ever more contexts, from recruitment Still, human oversight is insufficient if it year, the company Juniper Networks issued
deals with problems only after they occur. a report estimating that, by 2022, 55% of
1
Oxford Internet Institute, University of Oxford, 1 St Giles, Techniques to explain AI and predict its households worldwide will have a voice as-
Oxford OX1 3JS, UK. 2The Alan Turing Institute, outcomes are also needed. The Explainable sistant, like Amazon Alexa.
96 Euston Road, London NW1 2DB, UK. 3Department of
Computer Science, University of Oxford, Oxford OX1 3QD, UK. Artificial Intelligence program of DARPA As it matures and disseminates, AI blends
Email: mariarosaria.taddeo@oii.ox.ac.uk (Defense Advanced Research Project Agency) into our lives, experiences, and environ-
ments and becomes an invisible facilitator regulations, and uses of this technology. seized for lack of clarity or fear of backlash.
that mediates our interactions in a conve- Once identified, ethical principles must Ethical regulation of the design and use of
nient, barely noticeable way. While creating be translated into viable guidelines to shape AI is a complex but necessary task. The alter-
new opportunities, this invisible integration AI-based innovation. Such translation has native may lead to devaluation of individual
of AI into our environments poses further precedents, especially in medicine, where rights and social values, rejection of AI-based
ethical issues. Some are domain-dependent. translational research goes “from bench to innovation, and ultimately a missed oppor-
For example, trust and transparency are cru- bedside,” building on research advances in tunity to use AI to improve individual well-
cial when embedding AI solutions in being and social welfare. Humanity
homes, schools, or hospitals, whereas learned this lesson the hard way when
equality, fairness, and the protection it did not regulate the impact of the
of creativity and rights of employees industrial revolution on labor forces,
are essential in the integration of AI and also when it recognized too late
in the workplace. But the integration the environmental impact of massive
of AI also poses another fundamental industrialization and global consum-
risk: the erosion of human self-deter- erism. It has taken a very long time,
mination due to the invisibility and social unrest, and even revolutions to
influencing power of AI. protect workers’ rights and establish
This invisibility enhances the in- sustainability frameworks.
fluencing power of AI. With their The AI revolution is equally signifi-
predictive capabilities and relentless cant, and humanity must not make
beneficence, nonmaleficence, autonomy, and gating strategies. This has a dual advantage.
justice. There is reason to be optimistic about As an opportunity strategy, foresight meth- ACKNOWLEDGME NTS
further convergence, as other principles may odologies can help leverage ethical solutions. M.T. and L.F. are members of the Partnership on Artificial
be extracted from the Universal Declaration As a form of risk management, they can help Intelligence to Benefit People and Society; L.F. is also chair of
the scientific committee of AI4People.
of Human Rights. This convergence will fos- prevent or mitigate costly mistakes, by avoid-
ter coherence, and hence compatibility, of dif- ing decisions or actions that are ethically SUPPLEMENTARY MATERIALS
ferent ethical frameworks for AI and provide unacceptable. This will lower the opportu- www.sciencemag.org/content/361/6404/751/suppl/DC1
overarching ethical guidance for the design, nity costs of choices not made or options not 10.1126/science.aat5991
Published by AAAS
How AI can be a force for good
Mariarosaria Taddeo and Luciano Floridi
SUPPLEMENTARY http://science.sciencemag.org/content/suppl/2018/08/22/361.6404.751.DC1
MATERIALS
RELATED file:/content
CONTENT
REFERENCES This article cites 10 articles, 2 of which you can access for free
http://science.sciencemag.org/content/361/6404/751#BIBL
PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions
Science (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement of
Science, 1200 New York Avenue NW, Washington, DC 20005. 2017 © The Authors, some rights reserved; exclusive
licensee American Association for the Advancement of Science. No claim to original U.S. Government Works. The title
Science is a registered trademark of AAAS.