Responsible AI: Difference between revisions

From Wiki for Sustainable IT
Newer edit →
VisualWikitext
Creation of the Responsible AI page
 
Expanded
Line 1: Line 1:
'''Responsible AI''' refers to the design, deployment and use of artificial intelligence systems that are human-centred and human-controlled, trustworthy, inclusive, non-discriminatory and environmentally responsible.
'''Responsible AI''' refers to the design, deployment and use of artificial intelligence systems that are human-centred and human-controlled, trustworthy, inclusive, non-discriminatory and environmentally responsible.


== Definition ==
== Why it became a subject ==


The rapid spread of generative AI has brought these systems into the everyday operation of organisations. Responsible AI addresses this by setting requirements across the whole lifecycle of a system: choice of use case, training data, transparency, human oversight and environmental footprint.
Generative AI changed how these systems enter organisations. Enterprise software went through procurement, therefore through a decision. An AI service is reachable from a browser, without installation or budget: adoption precedes the decision.


The Institut du Numérique Responsable has formalised this approach in its '''AI Charter''', which complements the Responsible Digital Charter and aligns with the European AI Act.
The question is no longer whether to adopt AI, but under what conditions, and who decides.


== The four principles of the AI Charter ==
== The four principles of the AI Charter ==


=== 1. AI at the service of humanity ===
The Institute for Sustainable IT has formalised the approach in an AI Charter, complementing the Responsible Digital Charter and aligned with the European [[AI Act]].


* Develop AI that strengthens the central role of employees while extending their capabilities
=== AI at the service of humanity ===
* Build uses together with staff, through social dialogue and feedback
* Train personnel in the proper use of AI and its impacts
* Respect autonomy at work, the quality of social interaction and the meaning of work
* Include respect for international treaties protecting human rights, democracy and the rule of law in procurement specifications


=== 2. Inclusive and ethical AI ===
* develop AI that strengthens the central role of employees while extending their capabilities ;
* build uses with the teams, through dialogue and feedback ;
* train staff in appropriate use and its consequences ;
* respect autonomy at work, the quality of social interaction and the meaning of work ;
* include respect for international treaties on human rights and the rule of law in procurement specifications.


* Respect equity, diversity and non-discrimination
=== Inclusive and ethical AI ===
* Put in place mechanisms ensuring data quality and preventing gender, ethnic or religious bias
* Organise responses to discrimination arising from biased data
* Make AI-based digital services accessible to people with disabilities


=== 3. Trustworthy AI ===
* respect equity, diversity and non-discrimination ;
* put in place mechanisms ensuring data quality and preventing gender, ethnic or religious bias ;
* organise the response to discrimination arising from biased data ;
* make AI-based services accessible to people with disabilities.


* Inform users that they are interacting with an AI system, and disclose data sources
=== Trustworthy AI ===
* Explain how the algorithms work in a clear and accessible way
* Ensure human oversight of high-risk activities
* Strengthen system robustness against cyberattacks


=== 4. Eco-responsible AI ===
* inform users that they are interacting with an AI system, and disclose data sources ;
* explain how the algorithms work in clear and accessible terms ;
* ensure human oversight of high-risk activities ;
* strengthen robustness against cyberattack.


* Verify that the use case is genuinely useful before any deployment
=== Eco-responsible AI ===
* Favour '''frugal AI''' models with efficient resource usage
* Measure and minimise the environmental footprint across the system lifecycle
* Apply eco-responsible practices to both training and deployment


== Signing ==
* verify that the use case is genuinely useful before deployment ;
* favour '''frugal''' models with efficient resource use ;
* measure and minimise environmental footprint across the system life cycle ;
* apply eco-responsible practices to training and deployment alike.


The AI Charter may be signed by any public or private organisation, either on its own or together with the Responsible Digital Charter.
== Beyond compliance ==


== External references ==
The [[AI Act]] sets a floor. Two of the charter's commitments go beyond it: preserving autonomy at work, which no regulation requires, and the environmental footprint, which the AI Act addresses only marginally.


* [https://charter.isit-europe.org/charte-ia/?lang=en_GB AI Charter — Institut du Numérique Responsable]
== See also ==
 
* [[AI Act]] · [[AI governance]] · [[Shadow AI]] · [[Algorithmic bias]] · [[Generative AI]] · [[Sustainable IT]]
 
[[Category:Artificial intelligence]]

Revision as of 05:37, 6 August 2026

Responsible AI refers to the design, deployment and use of artificial intelligence systems that are human-centred and human-controlled, trustworthy, inclusive, non-discriminatory and environmentally responsible.

Why it became a subject

Generative AI changed how these systems enter organisations. Enterprise software went through procurement, therefore through a decision. An AI service is reachable from a browser, without installation or budget: adoption precedes the decision.

The question is no longer whether to adopt AI, but under what conditions, and who decides.

The four principles of the AI Charter

The Institute for Sustainable IT has formalised the approach in an AI Charter, complementing the Responsible Digital Charter and aligned with the European AI Act.

AI at the service of humanity

  • develop AI that strengthens the central role of employees while extending their capabilities ;
  • build uses with the teams, through dialogue and feedback ;
  • train staff in appropriate use and its consequences ;
  • respect autonomy at work, the quality of social interaction and the meaning of work ;
  • include respect for international treaties on human rights and the rule of law in procurement specifications.

Inclusive and ethical AI

  • respect equity, diversity and non-discrimination ;
  • put in place mechanisms ensuring data quality and preventing gender, ethnic or religious bias ;
  • organise the response to discrimination arising from biased data ;
  • make AI-based services accessible to people with disabilities.

Trustworthy AI

  • inform users that they are interacting with an AI system, and disclose data sources ;
  • explain how the algorithms work in clear and accessible terms ;
  • ensure human oversight of high-risk activities ;
  • strengthen robustness against cyberattack.

Eco-responsible AI

  • verify that the use case is genuinely useful before deployment ;
  • favour frugal models with efficient resource use ;
  • measure and minimise environmental footprint across the system life cycle ;
  • apply eco-responsible practices to training and deployment alike.

Beyond compliance

The AI Act sets a floor. Two of the charter's commitments go beyond it: preserving autonomy at work, which no regulation requires, and the environmental footprint, which the AI Act addresses only marginally.

See also

Retrieved from "https://en.wiki.isit-europe.org/index.php?title=Responsible_AI&oldid=59"