Trustworthy AI

Trustworthy AI refers to artificial intelligence systems that are designed to have transparent reasoning, are explainable (XAI), accountable, robust, fair and honest, respectful of data privacy, and steerable or alignable with human goals. ^{[1] [2] [3] [4]}

Trustworthy AI creation is a goal of AI governance and policymaking. To achieve transparency and data privacy, several privacy-enhancing technologies (PETs) can be used. These include: ^{[5] [6]}

• Homomorphic encryption for computing with encrypted data without ever decrypting it.
• Federated learning and secure multi-party computation (MPC) for distributing the model training without sharing information between the learning centers and computing servers.
• Differential privacy for exposing statistical data while guaranteeing that no private information is exposed.
• Zero-knowledge proof - providing proven validity for statements without disclosing any extra information.

Transparency in AI involves making the processes and decisions of such systems understandable to users and stakeholders. Accountability ensures that there are protocols for addressing adverse outcomes or biases that may arise, with designated responsibilities for oversight and remediation. ^{[nb 1]} Robustness and security aim to ensure that AI systems perform reliably under various conditions and are safeguarded against malicious attacks. ^[7]

Fairness is achieved by refusal training: training the models to avoid problematic requests, and by adding "bolting filters" to detect and prevent discussion on biased, unethical, or dangerous topics. ^{[2] [8]} Making AI that aligns with human goals is a task still being debated and developed (as of September 2025). It may be achieved (with some rate of success) through automatic anomaly detection during training, or during the interactive discussion with the user, as well as using filtering techniques for recommender systems that attempt to align the results with the user's request and profile, including content-based filtering and the use of collaborative filtering for classifying information. ^[3]

ITU standardization

A work programme for achieving Trustworthy AI was set up by the International Telecommunication Union, an agency of the United Nations, initiated under its AI for Good programme. ^[6] Its origin lies with the ITU-WHO Focus Group on Artificial Intelligence for Health, where a strong need for both privacy and analytics created demand for a standard in these technologies.

In 2020, AI for Good moved online, and the TrustworthyAI seminar series was established to initiate discussions on these topics. This eventually led to standardization activities. ^[9]

Multi-party computation

Secure multi-party computation (MPC) is being standardized under "Question 5" (the incubator) of ITU-T Study Group 17. ^[10]

Homomorphic encryption

Homomorphic encryption allows for computing on encrypted data, where the outcomes or result is still encrypted and unknown to those performing the computation, but can be deciphered by the original encryptor. It is often developed with the goal of enabling use in jurisdictions different from the data creation (under, for instance, GDPR).^{[ citation needed ]}

ITU has been collaborating since the early stage of the HomomorphicEncryption.org standardization meetings, which has developed a standard on homomorphic encryption. The fifth homomorphic encryption meeting was hosted at ITU HQ in Geneva.^{[ citation needed ]}

Federated learning

Zero-sum masks as used by federated learning for privacy preservation are used extensively in the multimedia standards of ITU-T Study Group 16 (VCEG) such as JPEG, MP3, H.264, and H.265 (commonly known as MPEG).^{[ citation needed ]}

Zero-knowledge proof

Previous pre-standardization work on the topic of zero-knowledge proof has been conducted in the ITU-T Focus Group on Digital Ledger Technologies.^{[ citation needed ]}

Differential privacy

The application of differential privacy in the preservation of privacy was examined at several of the "Day 0" machine learning workshops at AI for Good Global Summits.^{[ citation needed ]}

See also

• Artificial intelligence
• Privacy-enhancing technologies
• Data science

Notes

1. For further reading about bias prevention, see the article about fairness in machine learning.

References

1. Ethics Guidelines for Trustworthy AI, The Independent High-Level Expert Group on Artificial Intelligence, set up by the European Union, April 8, 2019 (Spanish Government website)
2. Mehrabi, Ninareh; Morstatter, Fred; Saxena, Nripsuta; Lerman, Kristina; Galstyan, Aram (13 July 2021). "A Survey on Bias and Fairness in Machine Learning". ACM Computing Surveys. 54 (6): 115:1–115:35. arXiv: 1908.09635 . doi:10.1145/3457607. ISSN   0360-0300. S2CID   201666566.
3. Russell, Stuart J.; Norvig, Peter (2021). Artificial intelligence: A modern approach (4th ed.). Pearson. pp. 5, 1003. ISBN   9780134610993 . Retrieved September 12, 2022.
4. How to Govern AI — Even If It’s Hard to Predict, Helen Toner, TED
5. "Advancing Trustworthy AI - US Government". National Artificial Intelligence Initiative. Retrieved 2022-10-24.
6. "TrustworthyAI". ITU. Archived from the original on 2022-10-24. Retrieved 2022-10-24.
    This article incorporates text from this source, which isby the International Telecommunication Union available under the CC BY 4.0 license.
7. "'Trustworthy AI' is a framework to help manage unique risk". MIT Technology Review. Retrieved 2024-06-01.
8. Discrimination, Bias, Fairness, and Trustworthy AI, Daniel Varona and Juan Luis Suárez, CulturePlex Laboratory, London, April 2022, Applied Sciences Magazine (MDPI Open Access Publishing)
9. "TrustworthyAI Seminar Series". AI for Good. Retrieved 2022-10-24.
10. Shulman, R.; Greene, R.; Glynne, P. (2006-03-21). "Does implementation of a computerised, decision-supported intensive insulin protocol achieve tight glycaemic control? A prospective observational study". Critical Care. 10 (1): P256. doi: 10.1186/cc4603 . ISSN   1364-8535. PMC   4092631 .