Following our previous discussion on the fundamentals of AI, it is important to consider how we can harness the full potential of AI responsibly. Responsible AI enables the design, development, and deployment of ethical AI systems in our society.1
As AI continues to evolve, it raises several ethical concerns. Short-term issues include fairness, transparency, accountability, privacy and data protection. Looking ahead, there are also worries about AI-induced unemployment and the equitable distribution of AI-generated economic benefits.2
Despite these valid concerns, many people are not well-informed about the current regulations and evaluations governing AI, leading to myths such as the fear of AI robots rebelling against humans.3
In this blog, we will first explore the purpose and potential of AI, including its content generation capabilities. Following that, we will discuss how AI is regulated and how its performance is rigorously evaluated and monitored throughout its development process.
The purpose and potential of AI
AI is designed to enhance human efficiency by automating complex tasks, analysing vast amounts of data, and making decisions faster than humans. It also expands and enriches our world with insights and innovations that were previously unimaginable, serving as a new avenue for self-expression. Despite concerns from science fiction and movies about AI dominating the world, this remains far-fetched, provided AI remains supervised and regulated. However, it is still crucial to maintain meticulous oversight.
Who checks on AI? Self monitoring and regulations
A common myth is that AI operates entirely autonomously without any form of self-monitoring mechanisms in place. But that is not true: it can detect when real-world changes diverge from its programmed model. For instance, just as how cars have evolved from their designs a century ago, AI algorithms can identify and adapt to such shifts. This self-regulation ensures AI remains relevant and accurate in our rapidly changing world.
However, the role of responsible and well-informed regulators remains crucial, and nations have been taking initiatives to address the technology. On December 9, 2023, the EU Parliament and Council provisionally agreed on the AI Act, pending formal adoption to become EU law. It ensures that fundamental rights, democracy, the rule of law and environmental sustainability are protected from high risk AI.4
Open communication is important. It is imperative that dialogue is maintained amongst all stakeholders to empower them with responsibility. This encompasses everyone from the recruiters who gather training data from participants to the software engineers who meticulously review lines of code.
Evaluating AI performance
Contrary to popular belief, AI does not operate on autopilot once it is programmed. It must pass through rigorous, multi-stage testing to ensure it functions as intended. This process starts with testing the software used for curating training data and developing models, often by writing the code in different ways to verify that it yields consistent results. Such practices ensure the foundational integrity of AI systems, preparing them for further examination.
In addition, the taxonomies of labels defining correct or incorrect behaviour, undergo thorough testing to confirm they accurately represent the data used to train models. This step is crucial for maintaining the quality and relevance of AI decisions. Furthermore, experts who assign these labels are regularly assessed and qualified, ensuring that their judgments are reliable and consistent.
Finally, the performance of AI models and algorithms is compared against human capabilities, focusing on their precision, consistency, and speed in making predictions. The effects of an AI model in the intended application will be assessed. In mental health support settings, for example, certain metrics exist for evaluating how someone is feeling before and after therapy. These same metrics will be applied to evaluate Cam AI’s services, guaranteeing accurate context interpretation and the generation of suitable responses.
Conversational AI
The primary service offered by Cam AI is conversational AI, which is a type of AI that can simulate human conversation.5 Its core function is to predict the next response. This involves integrating previous dialogue and contextual information about the users and the conversation’s purpose.
To prevent inappropriate responses, the uncertainty and risk tied to predictions are meticulously considered within mental health settings. Cam AI is committed to minimising inappropriate responses by closely controlling what the AI says. This is achieved through training with data from genuine therapy and psychological mentoring sessions and by offering transparent explanations for how each response is generated.
References
- Responsible AI | AI Ethics & Governance. (n.d.). Accenture. Retrieved March 1, 2024, from https://www.accenture.com/gb-en/services/applied-intelligence/ai-ethics-governance
- Stahl, B.C. Embedding responsibility in intelligent systems: from AI ethics to responsible AI ecosystems. Sci Rep 13, 7586 (2023). https://doi.org/10.1038/s41598-023-34622-w
- 10 most common myths about AI. (2022, February 10). Spiceworks. Retrieved March 1, 2024, from https://www.spiceworks.com/tech/artificial-intelligence/articles/common-myths-about-ai/
- Artificial Intelligence Act: deal on comprehensive rules for trustworthy AI | News | European Parliament. (2023, December 9). European Parliament. Retrieved March 1, 2024, from https://www.europarl.europa.eu/news/en/press-room/20231206IPR15699/artificial-intelligence-act-deal-on-comprehensive-rules-for-trustworthy-ai
- What is conversational AI: examples and benefits. (n.d.). Google Cloud. Retrieved March 8, 2024, from https://cloud.google.com/conversational-ai
Image Source: https://www.pexels.com/photo/a-woman-with-number-code-on-her-face-while-looking-afar-5473956/
Author: Julie Wang, Cam AI Volunteer
I am a first year undergraduate student from Cambridge University studying Psychological and Behavioural Sciences. I am currently volunteering as an assistant at Cam AI, reading papers about AI and mental health, engaging in outreach activities and writing blogs. I am curious about the ways in which AI can enhance the mental health services provided to humans, and am very excited to be part of this team.