On AI Voices and Learning
TLDR:
Immersive language practice through daily speaking and listening significantly improved my Italian language skills more than an app like Duolingo. That said, the act of speaking to learn a new language without immersion and receiving personalized feedback is now possible with AI.
AI tutors can revolutionize education by providing personalized, one-on-one tutoring that can achieve significantly better outcomes compared to traditional classroom settings.
AI tutors are a net positive to society, and ultimately provide a future where learning is no longer constrained by geography, resources, or traditional classroom constraints.
I’ve been in Italy for almost a month since I graduated from Berkeley, and my Italian language skills have dramatically improved over time. Leading up to my trip, I diligently practiced on Duolingo almost every day for three weeks. I’m Italian American and studied the language in undergrad, so the practice was primarily to re-acquaint myself with conjugations, nouns, and sentence structures. While Duolingo was useful, I deleted the app about a week into my trip in favor of another mechanism: speaking.
Every day I’ve spoken Italian to order meals, ask for directions, or find out from locals the best things to do in my area. Speaking (and listening) have by far paid the biggest dividends on my ability to learn Italian. I’m not alone in this realization. One of the most well known polyglots who’s learned over 50 languages, Arieh Smith, recently posted a YouTube video saying that speaking 10-15 hours over a 3 week period is crucial to his learning. While I’ve been immersed in this new language, I’ve also felt that with AI this speaking mechanism could be abstracted away from traveling and solely focused on as a way to learn a new language. To my surprise, it has.
I think Speak is a brilliant idea, and I want to generally talk about the power of AI tutors, voice, and language. AI tutors will be a huge net positive to society. I recently read Ethan Mollick’s “Co-Intelligence: Living and Working with AI”. Mollick devotes a chapter of his book to AI as a tutor, and opens the chapter by highlighting Benjamin Bloom’s 1984 paper “The 2 Sigma Problem”. This paper outlines that the average student, when tutored one-to-one, performed two standard deviations (yes, 98% better) than students educated in a typical classroom environment. AI tutors possess the opportunity to solve this 2 Sigma Problem. They can provide individual, personalized attention to tutoring needs. What’s more, they can democratize access to previously cost prohibitive, individual tutoring lessons and promote a more equitable learning environment. Immersion in order to learn a new language can be quite expensive, but the act of learning a language doesn’t have to be. Tutoring, and in general how we learn things, will look very different in the coming years as individuals turn more to AIs than humans for knowledge and personalized help.
Last month, a16z posted their thesis on AI voice agents and sees this opportunity across both B2B and B2C companies. For B2C specifically, they highlight the ability for AI to provide human-like services without the need to pay an actual human. They’re interested in AI companies that have compelling business cases for why voice is a necessity, and have listed Speak on their EdTech market map.
I’m really excited by the opportunity presented by AI tutors (and more broadly AI voice agents) as a way to access “human-quality” services more cheaply and effectively. I’m also keeping my fingers crossed on Speak releasing Italian on their platform so I can get back to having edifying conversations. Ultimately, whether or not AI tutors attract large user bases comes down to adoption, but I’m looking forward to an AI future where learning is no longer limited by geography, resources, or traditional classroom constraints.
