Rohit Prasad is the vice president and head scientist of Amazon Alexa and one of its original creators. This conversation is part of the Artificial Intelligence podcast. If you would like to get more information about this podcast go to https://lexfridman.com/ai or connect with @lexfridman on Twitter, LinkedIn, Facebook, Medium, or YouTube where you can watch the video versions of these conversations. If you enjoy the podcast, please rate it 5 stars on Apple Podcasts or support it on Patreon. This episode is presented by Cash App. Download it (App Store, Google Play), use code "LexPodcast". The episode is also supported by ZipRecruiter. Try it: http://ziprecruiter.com/lexpod Here's the outline of the episode. On some podcast players you should be able to click the timestamp to jump to that time. 00:00 - Introduction 04:34 - Her 06:31 - Human-like aspects of smart assistants 08:39 - Test of intelligence 13:04 - Alexa prize 21:35 - What does it take to win the Alexa prize? 27:24 - Embodiment and the essence of Alexa 34:35 - Personality 36:23 - Personalization 38:49 - Alexa's backstory from her perspective 40:35 - Trust in Human-AI relations 44:00 - Privacy 47:45 - Is Alexa listening? 53:51 - How Alexa started 54:51 - Solving far-field speech recognition and intent understanding 1:11:51 - Alexa main categories of skills 1:13:19 - Conversation intent modeling 1:17:47 - Alexa memory and long-term learning 1:22:50 - Making Alexa sound more natural 1:27:16 - Open problems for Alexa and conversational AI 1:29:26 - Emotion recognition from audio and video 1:30:53 - Deep learning and reasoning 1:36:26 - Future of Alexa 1:41:47 - The big picture of conversational AI
The host encourages listeners to leave reviews and mentions the impact that these reviews can have. Additionally, he discusses the role of voice assistants in introducing people to artificial intelligence and the challenges faced by the teams behind them.
The speaker discusses the importance of smart assistants having a persona of their own to make interactions feel more natural and evaluates what humans look for in their interactions with these assistants to determine their intelligence.
The Alexa Prize challenge was created to encourage university research teams to build a social bot that can engage in coherent and engaging conversation for 20 minutes. This challenge not only advances AI research but also provides a unique opportunity for academia to have access to resources like data and computing power that is typically available only in industry.
The speaker discusses how feedback mechanisms are crucial to improving the accuracy of AI systems and emphasizes the importance of incorporating different modalities, such as visual input, for optimal learning. They also note the need for a shift in mindset when conducting evaluations beyond standard research environments.
The focus of AI design should be on its capabilities and potential benefits for customers, rather than solely on making it human-like. AI can provide different ways to delight customers and enhance their experiences as a virtual assistant.
Alexis discusses the importance of trust in AI systems and autonomous vehicles and the need for them to act in a trustworthy manner, not just in terms of data privacy, but also in accurately carrying out commands and interactions.
Amazon Echo's far field speech recognition was made possible with a combination of large data sets, deep learning progress, and infinite GPUs on AWS, which solved the problem of detecting the right mentions of Alexa's address to the device versus other general statements.
Researchers express their excitement on far field speech recognition, despite initial skepticism, and its emergence in the home with improved interaction.
The co-founder of Alexa, Rohit Prasad, discusses how the AI assistant was developed using a statistical first approach, including entity recognition and intent classification, and the challenges of incorporating multiple domains and skills.
Alexa is becoming more proactive and expanding its capabilities in the smart home space with new features like hunches, and seamlessly integrating with services like buying movie tickets.
The usefulness of voice assistants like Alexa depends on their ability to remember information in the short and long term, such as previous commands and personal preferences, which is still a work in progress for these technologies.
The process of building skills through Alexa Skill Skip, which is a self-serve way of building experience on Alexa, requires tools that ensure skill quality. The challenges of making sure that the quality is high, especially when considering emotions, calls for nuggets of research like zero-shot learning or one-shot learning.
The space for modeling a person's goals in the context of interacting with AI is huge and includes attributes such as product reviews and consumer forums. In the next five years, AI will likely close the gap between goal-oriented dialogues and open domains.
The adoption of conversational agents is rapidly increasing, and there is hope to solve the complex issues associated with these agents within the next five years. The use of conversational agents in the home provides a unique experience.
The discussion included the challenges in research topic selection for budding researchers and the democratization of AI through the launch of Alexa and the Alexa skills kit.