The Age of Conversational Machine Learning, Digital Students and Natural Language Instruction

Professor Tom Mitchell’s parting message on conversational machine learning in the a16z podcast The History and Future of Machine Learning resonated with me. It’s a nice description: a piece of the puzzle that extends two ideas I’d been kicking around before:

• Digital Coaches: technology that holds conversations with us for self-improvement
• Knowledge 4.0: the idea that technology can take pre-recorded (written or spoken) resources like web pages or videos, and turn them into interactive educational material.

Conversational Learning, in the way that Prof Mitchell describes it, takes this further, by talking about how technology could also have conversations to learn for itself. The coach becomes the student: digital students.

Here’s the transcript of the relevant section below, where Mitchell talks about how on top of that, it empowers users to create little helpers just for their niche situation that otherwise wouldn’t be practical to invest engineering time in.

I think if you want to know the future of machine learning, all you need to do is look at how humans learn and computers don’t yet. So we learn for example. We do learn statistically like computers do. My phone watches me over time and statistically it eventually learns where it thinks my house is and where it thinks my work is it statistically learns what my preferences are. I also have a human assistant and if she tried to figure out what I wanted her to do, by statistically watching me do things that thousand times, I would have fired her. She doesn’t learn that way. She learns by having a conversation with me. I go into the office and I say

“Hey this semester, I’m teaching a course with Katarina on deep reinforcement learning. Here’s what I want you to do whenever this happens you do this whenever we’re preparing to hand out a homework assignment if it hasn’t been pre-tested by the teaching assistants two days before hand out you send a note saying get that thing pre-tested”.

So what I do is I teach her with and we have conversation. She clarifies. So one of the new paradigms for machine learning that I predict we will see in the coming decade is what I’ll call conversational learning. Use the kind of conversational interfaces that we have say with our phones. To allow people to literally teach their devices what they want them to do instead of have the have the device statistically learn it and if you go down that road. Here’s a really interesting angle on it: it becomes kind of like replacing computer programming with natural language instruction. So I give you an example of a prototype system that we’ve been working on together with Brad Myers one of our faculty in HCI, it allows you to say to your phone something like: “Whenever it snows at night, I want you to wake me up 30 minutes earlier.”

I live in Pittsburgh where it’s a useful app and none of the Californian engineers have created that app and today I could create that app. If I took the trouble of learning the computer language of the phone, I could program it. But only you know far less than 1% of phone users can actually have taken the time to learn the language of the computer right?

We’re giving the phone that chance to learn the language of the person. So with our phone prototype, if you say whenever it snows at night wake me up 30 minutes earlier. It says I don’t understand. Do you want to teach me and you can say yes, here’s how you find out if it’s snowing at night you.

Open up this weather app right here and where it says current conditions if that says “snow” it’s snowing. Here’s how you wake me up 30 minutes earlier. You open up that alarm app in this number you subtract 30 from it, so. With a combination of showing demonstrating and telling voice we’re trying to give users the opportunity to create their own amps their own programs with the same kind of instruction voice and demonstration that you would use if you’re trying to teach me.

This to me underscores why progress in general purpose conversation design (such as ConvAI.io or the dialog state technology challenge) is so important: for digital students to learn they first must be able to process (“parse”) everyday spoken and written English.

Related resources

Conversational Machine Learning talk

Prof. Mitchell’s talk on Conversational Machine Learning (April 2018)

David A Kolb

David A Kolb is a pioneer in “experiential learning”, having done the theoretical groundwork in the 70s and 80s. In essence (from Wikipedia):

• The learner must be willing to be actively involved in the experience;
• The learner must be able to reflect on the experience;
• The learner must possess and use analytical skills to conceptualize the experience; and
• The learner must possess decision making and problem solving skills in order to use the new ideas gained from the experience.