ARR: Lesson - How AI Learns Natural Interaction

A robot chatbot listens and responds to several humans inputting prompts.How AI Learns Natural Interaction

Think About It...

As we begin this lesson, ask yourself a few questions:

  • What "inputs" do you use to receive information? In other words: how do you hear things, smell things, taste things, see things, etc.?
  • Will those same answers work for AI?

Humans Communicating with Technology

Having reviewed the different types of decisions machines can make and how decision trees work, we will now look at how we naturally talk to AI using voice commands and text. This move from ideas to real-life uses will help us see how AI is part of our everyday lives. By studying how AI represents the relationship between words and meaning, we can understand how easy and helpful these technologies are and how they can improve.

It is quite common to interact with AI via verbal commands. Big Idea 4: Natural Interaction examines how intelligent agents require many kinds of knowledge to interact naturally with humans. For example, depending on which AI assistant we employ (i.e. Alexa, Siri, etc.) we can simply say our AI assistant’s name and state “ directions to (your chosen destination)” and AI will utilize the route finding technique we learned about in our last activity about search trees. We can also request information, set reminders/alarms, play music, and even control smart home devices all through verbal commands.

Learning Natural Interaction

These natural interactions allow us to access information, perform tasks, and receive assistance more conveniently and efficiently in our daily lives. In this lesson we will explore how AI represents the relationship between words and meanings. It has been noted that the English language is one of the most challenging languages to learn. Some of the challenges are due to silent letters, homophones, and syntax complexities just to name a few. In order for AI to interact naturally with humans, it employs various types of knowledge.

Some of the ways in which AI gains the necessary knowledge are by utilizing the following:

  • Word Embeddings: a way to represent words as points (vectors) in a multi-dimensional space. The distance and direction between these points (vectors) show how similar and related the words are to each other. For example, the words “pizza” and “burger” would be close together because they are both types of food, while “pizza” and “computer” would be far apart because they are not related.
  • Natural Language Processing (NLP): translates unstructured data into structured data by understanding the context is which words are used. It also uses sentiment analysis to help determine people’s feelings based on the words they use.
  • Large Language Models (LLM): trained on the largest amount of information possible from the internet. It uses its knowledge of semantics, neural networks, and advanced technology to predict what might come next in a sentence and to generate new information based on what it has learned.

Word Embeddings Example

Take a look at the following image. This is an example of how an AI can learn about the relationships between words.

Vector embedding example. Full explanation provided in page text.
Image description: Title - Vector Embedding
Y-axis: Age
X-axis: Gender
All words indicated on the vector with a blue dot:
Top line: mother (above girl), father (above boy)
Bottom line: girl, child, boy

Video Lesson

For more on how Large Language Models work, and chatbots in general, please watch the following video from Code.org.

Conclusion

In conclusion, our exploration into how we interact with AI through voice commands and text has shed light on its consistent presence in our daily lives. Understanding how AI represents the relationship between words and meanings is imperative, especially given the complexities of the English language. Despite its challenges, AI leverages various tools such as word embeddings, Natural Language Processing, and Large Language Models to facilitate natural interactions with humans. As technology continues to advance, our comprehension of AI’s capabilities and limitations will deepen, thereby paving the way for more seamless and intuitive human-AI interactions in the future.

[CC BY-NC-SA 4.0 Links to an external site.] UNLESS OTHERWISE NOTED | IMAGES: LICENSED AND USED ACCORDING TO TERMS OF SUBSCRIPTION - INTENDED ONLY FOR USE WITHIN LESSON.
Chatbot: The img/Shutterstock.com. Image used under license from Shutterstock.com and may not be repurposed.
Vector Embedding: lovelyduck/Shutterstock.com. Image used under license from Shutterstock.com and may not be repurposed.