ISTQB - AI Testing(CT-AI) - 1. Introduction to AI -Test 02

What was the original objective of "Artificial Intelligence (AI)" when the term was coined in the 1950s?. To create systems capable of processing big data. Building and programming "intelligent" machines capable of imitating human beings. To develop advanced statistical models. To automate industrial manufacturing processes.

How does the "AI effect" relate to systems like Deep Blue, the chess-playing computer?. Deep Blue is still considered the pinnacle of AI due to the AI effect. The "brute force" approach of Deep Blue is now not considered true AI by many, illustrating the AI effect where perceptions change. The AI effect ensures that definitions of AI remain constant over decades. The AI effect primarily concerns the hardware development for AI.

Which category of AI is characterized by systems that can reason and understand their environment as humans do?. Narrow AI. General AI. Specific AI. Applied AI.

As of 2021, what is the status of General AI systems?. They are widely deployed in consumer electronics. No general AI systems have been realized. They are commonly used for natural language processing tasks. They form the basis of most AIaaS offerings.

What distinguishes Super AI from General AI?. Super AI is limited to specific tasks, while General AI is broad. Super AI replicates human cognition and uses massive processing power and knowledge, expected to surpass human wisdom. General AI is theoretical, while Super AI is currently available. Super AI relies on conventional programming, unlike General AI.

What is a key difference in how AI-based systems using ML and conventional systems are developed?. Conventional systems use patterns in data to determine future reactions. AI-based systems using ML are programmed by humans using imperative languages like if-then-else for their core logic. In AI-based systems using ML, the system determines from patterns in data how to react to new data. Conventional systems are typically harder for humans to understand.

Why might the prediction-making procedure of an AI-based system be less easy for humans to understand?. Because they use outdated programming languages. Because the system determines on its own what patterns or features in data to use, which may not be intuitive to humans. Because they are always implemented with very few lines of code. Because they rely on simple rule engines exclusively.

Which of the following is recognized as a technology to implement AI?. HTML rendering engines. Operating system kernels. Neural networks. Firewall configurations.

Which of the following Al development frameworks is described as a deep learning open-source framework used by Amazon for AWS?. TensorFlow. PyTorch. Apache MxNet. CNTK.

What is Scikit-learn?. A high-level API capable of running on top of TensorFlow. An open-source machine ML library for the Python programming language. A suite of tools by IBM for AI solution development. A type of purpose-built ASIC for AI.

What is a common reason for choosing a particular AI development framework?. Its inability to run on GPUs. The programming language used for implementation and its ease of use. Its focus on non-scalable solutions. Its lack of support for data preparation.

For ML, why is "low-precision arithmetic" often beneficial?. It requires more complex CPU operations. It uses fewer bits for computation (e.g., 8 instead of 32), which is usually sufficient for ML and can be faster/more efficient. It improves the interpretability of the model. It is only supported by CPUs, not GPUs.

What type of hardware is an Edge TPU?. A general-purpose CPU with many cores. A purpose-built ASIC designed to run AI on individual devices. A high-performance GPU for cloud-based training. A neuromorphic processor mimicking brain neurons.

Which company produces the EyeQ family of SoC devices for vision processing in vehicles?. NVIDIA. Google. Intel. Mobileye.

What is a hybrid approach in the context of using AI components?. Using only open-source AI models. Some AI functionality is provided from within the system, and some is provided as a service (AIaaS). Training models on both CPUs and GPUs simultaneously. Combining multiple narrow AI systems to create general AI.

What type of support can be provided when ML is used as a service (AIaaS)?. Only access to a pre-trained model via an API. Support for data preparation, model training, evaluation, tuning, testing, and deployment. Primarily hardware rental for running AI models. Legal advice on AI ethics and regulations.

What is a typical payment model for AIaaS?. A one-time perpetual license fee. Often paid on a subscription basis. Based purely on the accuracy achieved by the model. Free for all non-commercial uses.

During a free trial period for AIaaS, what is the consumer generally expected to do?. Develop new features for the AI service. Market the AI service to other potential customers. Test whether the provided service meets their needs in terms of functionality and performance (e.g., accuracy). Provide their own hardware for the service to run on.

What is an advantage of using a pre-trained model like one trained on the ImageNet dataset?. It guarantees the model will be free of any documented defects. It reduces the risk of consuming significant resources with no guarantee of success and saves training costs. It ensures the model is perfectly tailored to any specific, niche task. It eliminates the need for any further testing.

How can a pre-trained model be used if no modifications are needed?. It must always be retrained with new data. It can simply be embedded in the AI-based system or used as a service. It requires conversion to a different programming language. It is only suitable for academic research purposes.

In transfer learning for deep neural networks, which layers are typically reused?. Only the final, specialized task layers. The early layers that perform basic tasks (e.g., identifying lines in an image classifier). All layers must be retrained equally. The layers related to data input and output.

What significantly impacts the effectiveness of transfer learning?. The programming language of the pre-trained model. The similarity between the function performed by the original model and the function required by the new model. The size of the pre-trained model in megabytes. The number of times the pre-trained model has been downloaded.

What is a risk if a pre-trained model lacks transparency?. It becomes easier to integrate into new systems. It may be harder to understand its behavior, limitations, or inherited biases. It generally performs better than transparent models. It is less likely to have undocumented defects.

If an AI-based system is known to contain a specific pre-trained model, what security risk might arise?. The model might become too popular and slow down. Vulnerabilities associated with that pre-trained model may already be known by potential attackers. The model provider might revoke access. It becomes more difficult to apply transfer learning.

What is the role of ISO/IEC JTC1/SC7?. It is a subcommittee specifically for AI standards development. It covers software and system engineering and has published a technical report on "Testing of Al-based systems". It focuses on ethical considerations in AI, similar to IEEE. It develops the AI Quality Metamodel with DIN.

What does the GDPR require regarding individuals and automated decision-making?. Individuals must always be subjected to automated decision-making for efficiency. It includes requirements for ensuring individuals' rights not to be subjected to automated decision-making under certain conditions. It mandates that all automated decisions be made by narrow AI. It requires AI systems to explain their automated decisions in source code.

The German national standards body (DIN) has developed which of the following?. The GDPR. The IEEE Global Initiative for Ethical Considerations in AI. The AI Quality Metamodel. ISO 26262 for automotive systems.

When are standards like ISO 26262 typically made mandatory?. They are always voluntary and never mandatory. Their use is normally only made mandatory by legislation or contract. Only when an AI system achieves Super AI status. When they are published by industry bodies like IEEE.

What is an example of Narrow AI mentioned in the syllabus?. A system with human-like cognitive abilities across all domains. Spam filters. A hypothetical superintelligent machine. An AI that has achieved technological singularity.

Which AI development framework is an open-source ML library operated by Facebook and supports Python and C++?. Keras. Apache MxNet. TensorFlow. PyTorch.