Introduction to Modern AI Course Final Assessment Exam Answers

Introduction to Modern AI Course Final Assessment Exam Answers

The “Cisco Introduction to Modern AI Exam Answers” course is designed to provide learners with foundational knowledge and practical insights into the world of artificial intelligence. This final assessment evaluates students’ understanding of key AI concepts, including machine learning, neural networks, natural language processing, and ethical considerations in AI. In this collection, you will find verified answers to the final assessment exam, ensuring accuracy and alignment with the course objectives. These answers are intended to help students review, validate their knowledge, and achieve a thorough understanding of the material.

1. Imagine that you have opened the Teachable Machine app https://teachablemachine.withgoogle.com/ and train it to classify two kinds of objects, a pencil and an eraser. You then test it with your face and notice that it classifies your face as an eraser (60% eraser, 40% pencil). What are possible reasons for this? Select all that apply? (Choose two.)

• Your shirt is a similar color as the eraser.
• There was some background noise (a dog barking) when you were taking pictures of the eraser, and also when you were testing the model with your face.
• When the model is not sure, or has a 50% probability assigned to both classes, it will default to the second of the two classes (in this case, the eraser).
• When taking pictures of the eraser, your face also appeared in some of those pictures, but not in the pictures of the pencil.

In this scenario, there are several possible reasons why the model might misclassify your face as an eraser after training on images of a pencil and an eraser. The model could be influenced by color similarities or unintended associations created during training. For instance, if your shirt color resembles the eraser, the model may be using color as a key characteristic to identify the “eraser” class. Additionally, if your face was accidentally included in some training images of the eraser but not the pencil, the model might have inadvertently associated your face with the eraser class.

However, some factors are unlikely to affect the classification. For example, the model won’t favor one class over another in a tie (e.g., assigning 50% probability to each class) as it simply outputs probabilities for each class without any default bias. Furthermore, background sounds, like a dog barking, have no impact since this model relies solely on visual data for classification, not audio.

---

2. You use a photo app to identify a rock that you are holding in your hand. You’re not getting good results, or no results at all. What are some things you can do? (Choose four.)

• Close and re-open the app.
• Try to install another photo app.
• Take another picture at a different angle.
• Set the rock on a table or the ground and take another picture.
• Adjust the bounding box.

If a photo app isn’t providing good results when identifying a rock, several adjustments might improve the outcome. For instance, taking a picture from a different angle can provide a more comprehensive view of the object, making it easier for the app to recognize distinctive features. Adjusting the bounding box so it more closely surrounds the rock can help the app focus on the relevant area, reducing distractions from the background. Additionally, placing the rock on a flat surface, like a table or the ground, rather than holding it in your hand can remove background elements, making the rock easier to identify.

Alternatively, trying a different photo app might yield better results, as different apps use various models with unique approaches to recognition. However, simply closing and reopening the app is less likely to enhance the model’s performance, though it could help resolve any temporary app glitches or freezes.

---

4. You are creating Spanish captions for a one-hour audio podcast recorded in English. This involves transcribing the audio to text, then translating the text to Spanish. What are some steps you can take to improve the accuracy of the final translation? (Choose two.)

• Take the direct output of the English transcription and pass it through to the translation model, to give the model the full and unfiltered context of the podcast discussion.
• Run the final translation through a chatbot and asking it to revise the text so that it will align better with the views of the intended audience.
• Check the transcription for possible words that are transcribed incorrectly (for instance, words or phrase that sound the same or similar to the intended words.
• Run the English transcript through a chatbot, in small batches of text at each time, to identify transcription issues.

To improve the accuracy of Spanish captions for an English audio podcast, it’s essential to follow certain steps to ensure both the transcription and translation are accurate. First, reviewing the English transcription for errors, especially words that sound similar but have different meanings, can help. Correcting any transcription errors before translation prevents misinterpretations from carrying over into the Spanish captions. Additionally, running the transcription through a chatbot in smaller text segments allows you to identify and correct any issues more thoroughly, as smaller batches enable the model to give more detailed feedback and help ensure accuracy.

However, directly passing the unedited English transcription to the translation model without corrections is not advisable, as any transcription errors may distort the intended meaning in the Spanish translation. Likewise, using a chatbot to revise the final Spanish translation to align with audience views risks altering the podcast’s original message. For accurate captions, it’s crucial to retain the podcast’s original tone and content without significant changes.

---

5. You are at a restaurant in a country where you do not speak the language. You are trying to figure out which menu items are recommended. What can you do? (Choose three.)

• Take a picture of the menu and use a photo app to extract the text from the image. Copy and paste the text into a translation app to read more about the menu items.
• Take a picture and upload it to a multimodal chatbot. Ask it to translate the menu items and to indicate which ones are recommended.
• Use speech to text translate what nearby customers are ordering, and use it to pick your items.
• Translate your question to the country’s local language and use text to speech to ask the waiter at the restaurant.

If you’re at a restaurant in a foreign country and want to understand which menu items are recommended, there are several effective approaches. Asking the waiter directly is a straightforward way to get recommendations, and using a translation app with text-to-speech can facilitate communication if you don’t speak the language. Additionally, taking a photo of the menu and using a photo app to extract the text for translation allows you to read through the menu and learn more about each item. Uploading the menu photo to a multimodal chatbot is another option, as it can both translate the text and provide insights about popular items.

However, using speech-to-text to capture what nearby customers are ordering is not recommended. This approach may invade others’ privacy, and even if you do identify what they order, it may not reliably reflect what most people would recommend.

---

6. You are at a restaurant in a country where you do not speak the language. You are trying to figure out which menu items have peanuts because someone from your group has a peanut allergy. What can you do? (Choose two.)

• Translate your question to the country’s local language and use text to speech to ask the waiter at the restaurant.
• Ask three chatbots which menu items are free of peanuts. If 2 out of 3 chatbots say it’s free of peanuts, then it’s probably safe to order it.
• Ask a chatbot which menu items are considered free of peanuts, and then use a translation app to ask the waiter if those dishes are indeed free of peanuts.
• Take a picture of the photos on the menu, and ask a chatbot to scan the images for signs of peanuts. If the chatbot does not detect peanuts, then the dish probably does not contain peanuts.

When trying to identify peanut-free menu items at a restaurant in a country where you don’t speak the language, certain steps can improve your confidence in finding safe options. Asking the waiter directly in the local language is likely the most reliable approach, as they will have specific knowledge about how each dish is prepared. Using a translation app with text-to-speech can help you communicate this question clearly. Alternatively, you could use a chatbot to help identify dishes that are generally peanut-free and then confirm these with the waiter using a translation app. This approach allows you to ask the right questions while relying on the waiter’s expertise for accuracy.

However, relying solely on chatbots to determine peanut-free dishes can be risky. For example, consulting multiple chatbots and assuming a majority answer is correct does not guarantee safety, as chatbots may lack specific information about how dishes are prepared at this restaurant. Similarly, asking a chatbot to scan menu photos for signs of peanuts is unreliable; even if peanuts are not visible in an image, they could be present in a form that isn’t detectable by simply analyzing a photo. Direct confirmation from the restaurant staff is essential for accurate and safe information regarding allergens.

---

7. You ask a chatbot to perform a task. For example, you ask it to split a restaurant bill among you and several friends. Which of the following can you add to the end of your instruction to improve the chances that the chatbot does the task correctly?

• Work through the task first, and show your work, then give your answer.
• Give your answer, then detail how you arrived at the answer.
• Think about the problem silently, and do not output anything except your answer.

To improve the chances that a chatbot performs a task correctly, it’s helpful to instruct it to **”work through the task first, and show your work, then give your answer.”** By doing this, the chatbot can use its intermediate steps and reasoning as context to help arrive at a more accurate final answer. Showing its work allows the chatbot to build on each step, creating a logical path to the solution.

On the other hand, asking the chatbot to **”give your answer, then detail how you arrived at the answer”** is less effective. If it outputs the answer first, it can’t use the later explanations as part of its reasoning process, which may reduce accuracy. Similarly, instructing the chatbot to **”think about the problem silently and do not output anything except your answer”** is also not ideal. While it encourages the chatbot to think through the problem, withholding intermediate steps means the chatbot can’t use previous output as context, which can negatively impact the quality of its final answer.

---

8. You want to check if a chatbot is making up things that are not true (hallucinating). What are some things that can help you to detect or reduce the chance that the chatbot makes up things that are not true? (Choose four.)

• Ask another chatbot to perform the same task and check if the two chatbots give significantly different answers.
• Give the chatbot a role that includes the following: “You do not hallucinate.”
• Ask the chatbot a follow up question: “are you sure?”
• Ask the chatbot to say “I do not know” when it’s not sure.
• Ask the same chatbot the same question three times. If it answers significantly differently each time, it may be hallucinating.

To help detect or reduce the chances that a chatbot makes up information, there are several effective strategies. First, asking the chatbot to say “I do not know” when it’s unsure can prevent it from making up answers just to satisfy your request. This option encourages honesty when the chatbot lacks the required knowledge. Additionally, following up with a question like, “Are you sure?” gives the chatbot a chance to review its previous response and correct any potential errors.

Other strategies include asking the chatbot the same question multiple times and comparing the answers. If the responses vary significantly, it might indicate that the chatbot is generating information inconsistently, suggesting possible hallucination. Likewise, asking a different chatbot to perform the same task and comparing responses can be helpful. If the two chatbots give contradictory answers, it’s a signal that one or both may be inaccurate, prompting further verification.

Simply giving the chatbot a role that includes “You do not hallucinate” is ineffective, as chatbots may not inherently recognize when they are generating inaccurate information. Instead, a role description encouraging specific actions like “check your work” or “admit when you do not know” is more likely to reduce hallucination.

---

9. When are some instances when you may want to use a compressed model that is downloaded onto your computer? (Choose three.)

• When you do not have an internet connection.
• When you want to keep your chats private to you.
• When you are using the chatbot to perform a relatively easy task, such as reviewing spelling and grammar.
• When you want to ask the chatbot to help you solve a challenging math problem.

Using a compressed model that’s downloaded onto your computer can be beneficial in several situations. If privacy is a concern, running the model locally means your chats remain on your computer, providing an additional layer of security. For simpler tasks, such as checking spelling and grammar, a smaller, compressed model may still perform well, making it a good choice for such cases. Additionally, having a model downloaded allows you to use it even without an internet connection, making it convenient in offline settings.

However, for complex tasks, such as solving challenging math problems, a larger, uncompressed model may perform better due to its enhanced capacity and accuracy. A compressed, local model might struggle with the complexity of more advanced tasks, so an internet-connected, more robust model would be preferable in those cases.

---

10. You are working in the marketing team of a company, and wish to get a chatbot to write some marketing material that is in a similar style as similar marketing emails produced by your team. What are some things that you can do to get the chatbot to write in a particular style? (Choose three.)

• Give the chatbot the names of some of your teammates who are writing marketing copy, and ask it to write like them.
• Review and describe how the company’s existing marketing emails are written. Then give the chatbot a list of instructions about how you want it to write; for example, which kinds of words to use over others, or whether to be concise or use more expressive language.
• Give the chatbot an existing draft of the marketing email you are writing, as well as other writing samples that are representative of the desired writing style. Ask the chatbot to revise your draft to use the writing style of the other samples.
• Give it examples of marketing emails written by your team, and ask the chatbot to follow a similar style.

To guide a chatbot in writing marketing material in a specific style, you can take several steps. Providing examples of marketing emails written by your team helps the chatbot understand and emulate the style directly. Additionally, sharing an existing draft alongside representative writing samples and asking the chatbot to revise your draft in line with the samples can help it align with the desired tone and approach.

Another effective method is to review and describe your company’s existing marketing style, giving the chatbot detailed instructions on tone, language preferences, and level of expressiveness. You can even ask the chatbot to analyze and describe the style from sample emails to better understand and apply it.

However, simply providing the names of your teammates who write marketing copy is unlikely to help, as the chatbot won’t have specific knowledge of their writing styles. Concrete examples of the style you want are much more effective for achieving consistent results.