Cloud-based deployment of AI often has which of the following advantages?
Answer(s): B
B is correct as cloud-based deployment facilitates dynamic resource management, allowing for efficient scaling up and down based on demand.In cloud environments, resources can be allocated in real-time to accommodate varying workloads, significantly enhancing operational efficiency. This flexibility allows organizations to manage costs effectively, as they can provision only the resources they need during peak load times and retract them during periods of low demand. Moreover, cloud infrastructure supports various scaling models, such as vertical and horizontal scaling, which are essential for AI models requiring substantial computational power during inferencing.Evaluation of Other Options:A: Reducing latency at model inferencing : While cloud solutions can improve latency through optimized routing and proximity to users, they often introduce additional network latency due to data transmission over the internet. Local deployments may offer better inferencing response times for time-sensitive applications, negating this advantage in specific scenarios.C: Enhancing privacy : Cloud platforms may expose data to external threats and compliance challenges, potentially compromising privacy. Although cloud providers implement advanced security measures, the inherent risk of data breaches and the complexities of complying with regulatory frameworks make this option less accurate for AI deployment.D: Increasing transparency : While cloud services can provide transparency through access to performance metrics and reporting, the complex nature of AI models often obscures transparency in decision-making processes. Opacity in model workings is a significant concern in AI governance, which is not inherently resolved by cloud deployment.References:https://www.forbes.com/sites/bernardmarr/2020/06/08/the-pros-and-cons-of-cloud-computing/ https://www.ibm.com/cloud/learn/cloud-computing https://www.gartner.com/en/information-technology/glossary/cloud-computing
References:https://www.forbes.com/sites/bernardmarr/2020/06/08/the-pros-and-cons-of-cloud-computing/ https://www.ibm.com/cloud/learn/cloud-computing https://www.gartner.com/en/information-technology/glossary/cloud-computing
After deploying an AI model, what must be documented?
Answer(s): D
D is correct because it encompasses the essential technical and performance-related aspects that ensure the operational legitimacy and integrity of the AI model post-deployment.Firstly, documenting the model’s architecture provides critical insight into its design and operational framework, enabling stakeholders to understand how inputs are processed and outputs generated. Performance metrics are crucial for gauging the model's effectiveness in real-world applications and identifying areas for improvement. Updates are important to track changes over time, ensuring continuous alignment with evolving governance standards and ethical considerations.Evaluating other options:A: While training data, hyperparameters, and stakeholder feedback are significant, they are more relevant during the development phase rather than post-deployment, thereby lacking the comprehensive operational focus required after deployment.B: Encountered issues, the monitoring process, and risks are essential for ongoing oversight but do not constitute a complete representation of the model's operational framework — they focus more on challenges rather than the architecture and performance.C: Licensing information and business goals are critical for compliance and alignment but do not detail how the model functions or performs, thus lacking technical depth which is paramount in governance contexts.In summary, D provides the necessary foundation for understanding, auditing, and ensuring the responsible use of AI systems, addressing both transparency and accountability.References:https://www.nist.gov/publications/guide-creating-ethical-ai https://www.oecd.org/going-digital/ai/principles/ https://www.weforum.org/reports/ethics-guide-ai
References:https://www.nist.gov/publications/guide-creating-ethical-ai https://www.oecd.org/going-digital/ai/principles/ https://www.weforum.org/reports/ethics-guide-ai
A company deploys an AI model for fraud detection in online transactions. During its operation, the model begins to exhibit high rates of false positives, flagging legitimate transactions as fraudulent. Which is the best step the company should take to address this development?
Answer(s): C
C is correct because deactivating the model ensures immediate mitigation of user dissatisfaction and potential financial loss while allowing for a thorough assessment of the model's performance.Reasoning: The spike in false positives undermines the user experience, trust, and operational efficiency,necessitating immediate action. Deactivating the model halts erroneous transactions and protects customers from undue inconvenience while providing an opportunity to conduct a detailed evaluation of the model's algorithms, data inputs, and thresholds. A swift assessment can identify whether the issue is due to model drift, inadequate training data, or improper parameter tuning, facilitating a more targeted re-deployment strategy.Evaluation of Other Options:A: Dedicate more resources to monitor the model: While ongoing monitoring is essential for AI model governance, merely increasing resources without first addressing the false positives does not resolve the underlying problem and can lead to customer frustration. B. Maintain records of all false positives: Documenting false positives is a good practice for auditing and understanding model performance over time; however, this is a reactive approach that does not provide an immediate solution to the operational issues currently impacting legitimate transactions. D. Conduct training for customer service teams to handle flagged transactions: Training staff to manage the fallout from the model’s errors does not rectify the root cause of the high false positive rate, thus allowing the flawed model to continue affecting user satisfaction and financial outcomes.References:1. https://www.jmlr.org/papers/volume18/16-642/16-642.pdf 2. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7306467/ 3. https://www.ai.gov/learn/2020/ai-governance-guidelines.html
References:1. https://www.jmlr.org/papers/volume18/16-642/16-642.pdf 2. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7306467/ 3. https://www.ai.gov/learn/2020/ai-governance-guidelines.html
CASE STUDYPlease use the following to answer the next question: A local police department in the United States procured an AI system to monitor and analyze social media feeds, online marketplaces and other sources of public information to detect evidence of illegal activities (e.g., sale of drugs or stolen goods). The AI system works by surveying the public sites in order to identify individuals that are likely to have committed a crime. It cross-references the individuals against data maintained by law enforcement and then assigns a percentage score of the likelihood of criminal activity based on certain factors like previous criminal history, location, time, race and gender. The police department retained a third-party consultant to assist in the procurement process, specifically to evaluate two finalists. Each of the vendors provided information about their system’s accuracy rates, the diversity of their training data and how their system works. The consultant determined that the first vendor’s system has a higher accuracy rate and based on this information, recommended this vendor to the police department. The police department chose the first vendor and implemented its AI system. As part of the implementation, the department and consultant created a usage policy for the system, which includes training police officers on how the system works and how to incorporate it into their investigation process. The police department has now been using the AI system for a year. An internal review has found that every time the system scored a likelihood of criminal activity at or above 90%, the police investigation subsequently confirmed that the individual had, in fact, committed a crime. Based on these results, the police department wants to forego investigations for cases where the AI system gives a score of at least 90% and proceed directly with an arrest. What is the best reason the police department should continue to perform investigations even if the AI system scores an individual’s likelihood of criminal activity at or above 90%?
B is correct because AI systems that affect fundamental civil rights should not be fully automated.The use of AI in policing, especially in making predictions about criminal activities, raises profound ethical and legal implications. Such systems potentially infringe on individuals' civil liberties, including due process rights and the presumption of innocence. Relying solely on algorithmic predictions could lead to wrongful arrests and perpetuate existing biases inherent in the data, disproportionately impacting marginalized communities. The principle of human oversight is vital in safeguarding against these risks and ensuring accountability within law enforcement practices. By conducting thorough investigations, the police can gather contextual evidence that AI systems cannot interpret or measure, thus supporting fair legal processes and fostering public trust in law enforcement.Critical Evaluation of Other Options:A: Impact Assessment : While it is important that an impact assessment is performed for transparency and accountability, the absence of such an assessment does not directly justify the need for continuing investigations. The essence of the issue lies more substantially in civil rights implications than in procedural compliance.C: Identifying Additional Individuals : While investigations can uncover additional criminal associates, the primary concern here revolves around the ethical implications of completely trusting AI predictions in serious matters like arrests. This point is relevant but secondary to the overarching issue of maintaining oversight to protect civil rights.D: Information Relevant to Sentencing : Investigations may certainly provide context for sentencing; however, the focus of the question is on ensuring the integrity of the arrest process itself. A flawed or biased AI prediction leading to an arrest is more pressing than post-arrest considerations.References:https://www.aclu.org/news/privacy-technology/its-time-to-regulate-police-use-of-ai https://www.brookings.edu/research/getting-algorithms-right-how-to-get-ahead-of-bias-in-ai https://www.ojp.gov/library/publications/impact-artificial-intelligence-policing-technology-and-impact-civil-liberties
References:https://www.aclu.org/news/privacy-technology/its-time-to-regulate-police-use-of-ai https://www.brookings.edu/research/getting-algorithms-right-how-to-get-ahead-of-bias-in-ai https://www.ojp.gov/library/publications/impact-artificial-intelligence-policing-technology-and-impact-civil-liberties
You asked a generative AI tool to recommend new restaurants to explore in Boston, Massachusetts that have a specialty Italian dish made in a traditional fashion without spinach and wine. The generative AI tool recommended five restaurants for you to visit. After looking up the restaurants, you discovered one restaurant did not exist and two others did not have the dish. This information provided by the generative AI tool is an example of what commonly called:
C is correct because the AI tool produced fabricated information that misrepresented reality, indicating a phenomenon known as hallucination.In the realm of generative AI, hallucination occurs when models generate outputs that are plausible but factually incorrect or entirely made up. This is particularly relevant to your scenario, where the AI-laden recommendation included non-existent restaurants and inaccurate dish information, highlighting a gap in its reliability and factual grounding. The presence of erroneous data suggests an overreliance on patterns learned during training without appropriate validation against real-world entities.Evaluation of Other Options:A: Prompt Injection : This refers to a manipulation technique where an adversary alters the input to influence outputs in unintended ways. In this case, the output was flawed due to inherent model limitations, not external tampering with prompts.B: Model Collapse : This is a concept related to the instability of generative processes where models stop producing diverse outputs over time. The current situation does not describe a lack of diversity but rather specific inaccuracies in recommendations.D: Overfitting : This occurs when a model learns to perform well on training data at the expense of generalizability on unseen data. Here, the problem relates to inaccuracies in outputs rather than an inability to generalize across data sets.The occurrence of hallucinations in AI models necessitates ongoing improvements in architecture and training methodologies to enhance their reliability in real-world applications, as even high-performing models may produce significant inaccuracies without proper context or verifying against factual databases.References:https://www.microsoft.com/en-us/research/publication/ai-hallucinations/ https://www.ibm.com/cloud/learn/hallucination-in-ai https://towardsdatascience.com/the-challenge-of-hallucination-in-generative-ai-6b5621781c16
References:https://www.microsoft.com/en-us/research/publication/ai-hallucinations/ https://www.ibm.com/cloud/learn/hallucination-in-ai https://towardsdatascience.com/the-challenge-of-hallucination-in-generative-ai-6b5621781c16
Post your Comments and Discuss IAPP AIGP exam prep with other Community members:
Our website is free, but we have to fight against AI bots and content theft. We're sorry for the inconvenience caused by these security measures. You can access the rest of the AIGP content, but please register or login to continue.