Paolo Pirjanian is the CEO and cofounder of the social robotics company Embodied and a general partner at Calibrate Ventures, an early stage deep tech investment firm. The following interview has been lightly edited for clarity and length.
HAUS: What led you to build a robotics startup?
Paolo: After I finished my PhD, I decided to go into academia and do research. So I moved to the US, I was at USC and working at NASA JPL working on Mars rovers and stuff like that. But I always had a fascination with entrepreneurship. I’d read about all these startups, but I had no clue how to start a business. It just looked like a puzzle to me and unfortunately I get attracted to things that I don’t know about. So when I was at JPL in 2001 I was approached by recruiters working for Bill Gross because Bill wanted to start a robotics company. I thought that was interesting so i went and talked to him about it. All of the people I had talked about prior to him hadn’t really captured my imagination—none of it was exciting enough. But within five minutes of meeting Bill he painted this picture of a company that was going to become the Microsoft of robotics. I could see he had the vision, he had the experience. So I thought to myself that this whole vision probably wouldn’t come to fruition, but I’m going to learn about entrepreneurship. So I jumped ship from my dream job working at NASA and joined that company—Evolution Robotics—as CTO. We started working on some amazing technologies—really deep tech—and that’s where I learned the ropes by watching how these people were conducting business. Although I was CTO, I was inserting myself more and more in the business side of things.
The short version of the story is that the company was not successful after a lot of investment from Idealab so they decided to shut it down. I negotiated with Idealab for the core tech that my team and I had developed and spun it out into a NewCo. I gave Idealab 40% ownership in exchange for the technology and started the company in 2008. We went after a vertical that was successful and sold that company to iRobot. After the acquisition I was CTO there for three years helping them integrate our technologies into their portfolio of products. iRobot was a pretty established company and they were very generous with me and gave me a lot of room to navigate. But I realized my passion was startups and felt like it was time to start a company again. Social robots were something that were fascinating to me because so many companies had tried and failed. I believed that I had the insights to understand why they had failed and what they had done wrong. I wanted to take a stab at it myself. And that resulted in Embodied, where we created a robot for children called Moxie as our first product.
What is Embodied?
So, the overarching vision for the company was to help to build companion AI robots that enhance our life experiences. We all have to go through difficulties across various stages in our life. One of the most vulnerable stages in life is early childhood because we’re just innocent and can’t protect ourselves. Our brains are developing very rapidly and there are a lot of important essential developmental milestones we need to meet. Another vulnerable stage is old age. We start losing some of our abilities due to illness and aging, there’s loneliness, Alzheimer’s dementia, and all these things that make us vulnerable. So the idea was to create a companion robot that’s going to be there as the emotional and mental support for people. We first chose to go after supporting children with Moxie, an AI companion robot that helps with development. The original focus was on social and emotional development So teaching children all the skills that we would categorize as emotional intelligence or EQ: emotion regulation, positive thinking, coping with anxiety or fear, all that. Recently we started expanding into also incorporating educational content. That’s possible today because of the generative AI wave that has happened. We have amazing technologies that were not possible before, which opens up a very powerful opportunity to have an AI tutor to help your child grow academically. So Moxie can provide holistically for the child by helping them develop both EQ and IQ.
What was the core innovation that allowed Moxie to succeed where other social robots have failed?
I think it’s a combination of a lot of things. Primarily it’s really about the software and content. But hardware plays a role as well because the design of the hardware needs to make a connection with the user. So this is one of the insights I had when I started Embodied and why I thought I knew all these other social robotics companies were failing. A lot of it was hardware design. They created interesting gadgets, but if you want to provide social and emotional connection, well there was a problem: none of these robots had eyes on them. They had plastic parts with LED rings behind them for eyes. But they didn’t have the ability to make eye contact, blink, or look confused when they’re thinking for example. There’s a lot that happens in the eye and it’s super important for social interaction. It’s very hard to do that, but we decided that we were going to have these social traits that are very important in communication. As a matter of fact, there are studies that show that only 7% of our communication is verbal—93% is nonverbal. It’s in the intonation of your voice and facial expression. It’s micro-body language like nodding to indicate you’re listening to me or seeing if you look confused, you might have a question. We wanted to capture that nonverbal communication in the design of Moxie’s hardware. But software had to support that. Because if you have eyes that can make eye contact, but they’re off by even a few degrees, it’s going to look like the robot is looking through you or ignoring you. So there’s a level of accuracy that’s required for this system to work well. We spent a lot of time developing the technology to make that possible.
So robots today, eight years after we started our company, they adopt what I call the Alexa interface, where they start with a command like ‘Hey Alexa, what’s the weather?’ That is a fantastic interface and it was a paradigm shift when it was introduced, but it’s not a great interface for social interaction. It’d be really awkward if I had to say ‘Hey Alexa’ every time I want to say a sentence. Instead, we should look at body language and all the other cues that let me know when a person is done speaking and now it’s your turn to speak and continue the conversation. That’s also something we implemented in Moxie, which created a lot of engineering headaches to make that happen. Because what happens is that in a real world environment, you may not be the only person speaking. So we cannot just look at the signal of speech and say, ‘okay, when, when the person is done speaking, then Moxie can take over’ because there could be someone else in the background talking to a third person, right, or the TV could be running in the background. So we had to look at our cues and involve visual cues, body language, and all these things so that Moxie knows when a user is done speaking. So all of these things are innovations that allow us to have a much more intuitive social interaction with a robot that were not possible before.
Since you’ve started Moxie, there’s been a lot of evidence that people develop parasocial relationships with AI tools like ChatGPT or the growing number of AI girlfriends. Some people are concerned that these relationships could supplant genuine human relationships. What is your take on this as someone building advanced social robots?
There is still a huge stigma against this. It’s like what is your relationship going to be with other humans if you’re going to be building a relationship with machines? I think that’s healthy to have that kind of attitude, but the kind of things we are trying to solve, we just don’t have enough human labor. For instance, we talked about elderly care earlier and we do not have enough human labor that can sit and talk for two hours about their memories or whatever. Even in nursing and medical care this is viewed as low value high touch interactions. We have a massive shortage of nurses, and they want them to only focus on low touch high value tasks like giving medication. The same thing applies for children. If we talk about mental health for example, there are a lot of children that suffer from various developmental challenges like autism, anxiety, and depression. But we have a massive shortage of therapits. For every 350 people that need therapy, there’s only one therapist. And when you’re in that much demand, the cost of therapy can go up significantly so now it’s not even affordable. There is a massive unmet need for human care. The way you can close that labor gap is with some forms of automation. Today, we have the technology that can make machines more human. They may be pretending to have empathy, but in my opinion that’s not much different from going to a therapist that meets with 20 clients in one day. They pretend they care about your particular struggles—they express sadness, empathy, and compassion. But I’m not sure if they would survive as human beings if they had to really feel all those emotions with 20 different clients every day. So Moxie is emulating that and has unlimited patience. Moxie is never going to have a bad day. A bad day for them is if Google Cloud crashes and causes an outage.
In addition to Embodied, you’re also a general partner at Calibrate Ventures. How has your experience building robotics startups affected your approach to investing in early stage deep tech startups?
Every deep tech startup has to invent a bunch of different things and then prove it in the marketplace. I think we underestimate what it takes to actually create a market, especially if it’s a new category of product that still requires a lot of education. People are skeptical, they may not believe in it. So it takes time to actually find support and build a business around it in my experience. So from my perspective, if someone is going into deep tech you want to partner with founders that are top of their field. But even that may not be enough.
Deep tech usually requires a lot of capital and you have to be able to sell it. So you have to be really good storytellers. One of the things we usually look for is, how good are these people at storytelling because the first round of financing is just to get them started, but they need to raise a lot more money before they are successful. And that takes really good storytelling to be convincing. So strong storytelling is an extremely important part of being a deep tech founder.
From my perspective, there are not enough founders who appreciate the power of storytelling. I may have been one of those founders earlier in my career, but I’ve learned the hard way that it is absolutely important. We have some amazing founders in the Calibrate portfolio and some of them do a lot better than the others—not necessarily because they’re smarter, just because they’re much better storytellers. It’s a skill that is completely underestimated because the people that need it the most are probably the ones that appreciate it the least.
Finally, deep tech founders also need perseverance because these things tend to take much longer than anyone could have imagined. And you want to make sure that you have a founder and a team that is dedicated and has the grit to deal with all the challenges. These are not easy things to do because it’s like building an industry almost all by yourself.