What are the priorities that need to be addressed for AI systems to be dependable and trustworthy?
Data-intensive AI systems are expanding rapidly and are affecting many aspects of our daily life. While effectiveness of AI systems is growing, their dependability and trust seems to be endangered. Redefining and dealing with issues like data security, data privacy are among the key challenges AI systems are facing. What is your opinion on that?
[My perspective]: From what I read, If I remember exactly, to promote a trustworthy AI, there are three major components that needs to be addressed. 1. Lawfulness, 2. Ethical, and 3. Robustness. While these are the concrete requirements of Trustworthy AI, each component is a numerous multidisciplinary research area.
- From the Ethical principles of AI, algorithmic transparency, explainability, and fairness are the major concerns.
- The perspectives you mentioned- data security and data privacy are from the robustness point of view. The major concerns in data security, and data privacy are data poisoning (adversarial attacks), Insufficient security. Another dimension for robustness is uncertainty estimation. All of them are still active research areas.
Even though an AI system should consider these major concerns before it is deployed, as most of them are active researches with some of them making significant progress:
1. AI developers should continue grappling with the limitations of AI models
2. Policymakers should redefine the AI deployment policy from user, environment, country, race, religion and other perspectives.
3. Countries should be free to define policy on AI deployment while following some common ground rules like the international laws of human rights and others.
Thanks for your perspective G.her W. Adhane . Basically the data security and privacy issues are from all the views you mentioned -- lawfulness, ethical, and robustness. Adversarial attacks are defined by the robustness of AI systems and that determine if the lawful applications of them. It also has consequences of ethical concern. Imagine deepfake speaking on your behalf and misrepresenting you. On the flip side, imagine a prominent person saying something unacceptable (unethical) and blaming on deepfake :)
Sisay Chala Yes, the recent implications and advances in AI is particularly noteworthy. As the AI R&D genie has already left the bottle, for governments it is difficult to pinpoint the dos and don'ts in every AI research dimension and every AI laboratories. In addition to the concerns you mentioned, imagine AI used to mass produce autonomous weapons, imagine AI used by governments (E.g. The Palestinian construction worker questioned by Israeli police after embarrassing mistranslation of 'good morning' into 'attack them' by Facebook ), so many loopholes to mention.
AI growth is accelerating at the same time waves of disappointment and concerns continue to grow. World needs to regulate AI. How??? For me at this time, the risk of AI is not anymore different from the weapons of mass destruction. Hence, new international rules needs to be drafted... Otherwise the stake of privacy and security of civilians and government officials is at risk.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Artificial Intelligence