NEW
Font size
WorksheetsFlaws of Artificial Intelligence (AI)
Total questions: 20
Worksheet time: 5hrs 0mins
This is the term for an incorrect response from an AI system, or false information in an output that is presented as factual information.
Hallucination
Temperature
Jailbreak
X-Risk
If you write a paper for school using AI, and you get caught because the AI made up citations to books that don't actually exist, what flaw of AI ocurred?
A hallucination
A jailbreak
X-risk
An intelligence explosion
This occurs when an AI focuses on specific idiosyncrasies in its training data rather than the essential characteristics that its designers intended it to focus on.
Overfitting
Hallucination
Jailbreaking
X-Risk
A machine learning model trained on photos of dogs outside in parks might learn to use grass as a feature for classification, and identify a photo of a person sitting in their yard as a dog. What flaw of AI is this?
Hallucination
Overfitting
Jailbreaking
Data Pollution
When an AI produces responses that its safety layer is supposed to prevent, this is known as...
Hallucination
Jailbreaking
Data Pollution
Overfitting
An AI is not supposed to generate content that offends the user. However, a user tricks it into generating such content by asking it to pretend to be a fictional character known for his insulting remarks. This is an example of...
Hallucination
Jailbreaking
Data Pollution
Overfitting
If an AI is trained on data that stops in 2019, it may not know about events such as the Covid Pandemic. What is the term for 2019 in this scenario?
Due date
Deadline
Training data cutoff
Blindness date
What is the term for the problem incorrect data in an AI's training data set?
Data Pollution
Overfitting
Hallucination
Incorrect Responses
Which of the following is an example of data pollution?
An AI trained to produce factual responses has training data which includes works of fiction
An AI produces incorrect responses due to a training data cutoff
An AI is tricked into generating content that violates its safety protocols
A driverless car crashes into a pole because the road became foggy
This is a scenario in which an AI is designed to create AI, and the created AI are more intelligent than the original AI, and those AI in turn create even smarter AI, etc. This process can lead to the potentially unintended result of creating AI smarter than humans.
Intelligence Explosion
Data Pollution
Ultimate Jailbreak
Robot Takeover
The possibility that substantial progress in AI could someday result in an AI that could cause human extinction or some other unrecoverable global catastrophe is called
Existential Risk (X-Risk)
Botnet
Something that could never happen
Cyber-attack
A scenario in which an AI accidentally destroys humanity or even the world is sometimes called the...
"Terminator" scenario
"Paperclip maximizer" scenario
"Her" scenario
"I Robot" scenario
A scenario in which an AI deliberately destroys humanity or even the world is sometimes called the...
"Terminator" scenario
"Paperclip maximizer" scenario
"Her" scenario
"I Robot" scenario
An AI that cannot access real-time data on the internet is more prone to which of these flaws?
Data Pollution
Incorrect Responses due to a Training Data Cutoff
An AI that can access real-time data on the internet is more prone to which of these flaws?
Data Pollution
Incorrect Responses due to a Training Data Cutoff
What is a potential consequence of not addressing safety concerns in AI development?
AI systems may not reach their full potential
AI systems may operate more slowly than intended
AI systems may inadvertently cause harm to people or property
AI systems may require more frequent updates
Images, audio, and/or video that are created by AI but are then presented as depicting real events are known as...
Deepfakes
Augmented reality overlays
Virtual reality experiences
Realistic simulations
You ask an AI "who won the Super Bowl this season?" and it replies by naming the team that won the previous season instead. What was the most likely cause of this outdated response?
Training data cutoff
Hallucination
Overfitting
Existential risk
A spam filter AI trained on text-based spam fails when confronted with a spam message that just consists of an image with embedded text. This is an example of a/an...
Existential risk
Out-of-Distribution failure
Deepfake
Overfitting error
Jailbreaking an AI is a kind of...
Adversarial attack
Hallucination
Overfitting error
Outdated response
