Facing Off Against the CEO of the AI Firm That Mimicked My Identity

Today, I’m diving into a conversation with Shishir Mehrotra, the CEO of Superhuman, the company that has evolved from its roots as Grammarly, which remains its flagship product. Shishir previously served as the chief product officer at YouTube and is currently on the board of directors at Spotify. We initially set up this interview to explore the impact of AI on software, platforms, and creativity, but things took an unexpected turn.

Last August, Grammarly introduced a feature called Expert Review, which provided writing suggestions from AI-generated “experts,” including names like mine. This sparked outrage among many, including investigative journalist Julia Angwin, who filed a class-action lawsuit after discovering our names were used without permission. In response, Superhuman quickly offered an opt-out option and ultimately decided to discontinue the feature entirely. Shishir has expressed regret over the situation, and you’ll hear him address it during our chat.

As we prepared for this discussion, I wondered if Shishir would still show up, given the complexity of the questions surrounding decision-making, AI, and platform ethics. To his credit, he did, and we had a candid conversation that sometimes got tense, revealing our differing views on the implications of AI.

So, without further ado, here’s Shishir Mehrotra, CEO of Superhuman.

Shishir, welcome to the show.

Thanks for having me.

I’m glad you’re here, though I admit I’m a bit surprised. I have a lot of questions about AI, public sentiment towards it, and the controversial feature you launched in Grammarly that stirred up quite a bit of emotion. Let’s start with the basics. Superhuman encompasses Grammarly and Coda, among other products. Can you give us a quick overview of Superhuman’s structure and offerings?

Superhuman is an AI-native productivity suite designed to integrate AI into the tools people already use. We rebranded from Grammarly to Superhuman last year to reflect our expanded scope. Alongside Grammarly, our suite includes Coda, a document space, and Mail, a popular email client.

We recently launched Superhuman Go, a platform that connects users with proactive AI assistance tailored to their work environments. Think of Go as an extension of Grammarly, allowing users to create agents that function similarly, whether for sales, support, or other tasks. Our goal is to embed AI seamlessly into your workflow, so you don’t have to think about it.

You mentioned that Grammarly is often experienced as a keyboard tool, integrating with various applications like Google Docs. How do you differentiate your approach from others that also incorporate AI?

Many tools attempt to integrate AI, but few do it effectively. Grammarly acts as your assistant across a multitude of platforms, whether web-based or desktop applications. We’ve developed a way to observe user behavior and provide unobtrusive suggestions, making changes on your behalf. This consistent experience across different applications is what sets us apart, contributing to our 40 million daily active users.

It seems like the promise of your tool is to create a more intelligent experience by understanding the various environments in which users work.

Absolutely. For many users, it’s about having a familiar experience that feels like a virtual assistant by their side, enhancing their productivity.

Is the focus on consistency of experience, or is it about delivering better results?

It’s both. Grammarly’s omnipresence is crucial, and we pride ourselves on delivering high-quality suggestions. As we evolve, we’re transforming Grammarly into a platform that allows others to create agents and experiences that maintain that high standard.

Let’s talk about decision-making. How do you approach it at Superhuman?

We have various frameworks for decision-making. One key concept I developed is called Eigenquestions, which emphasizes framing the right questions. We also use a process called Dory and Pulse to gather feedback and avoid groupthink.

Now, let’s address the Expert Review feature. It generated suggestions using names without permission, including mine. Can you explain the decision-making process behind launching this feature and the subsequent choice to remove it?

I anticipated this question. I understand the challenges experts face today, and it pains me that we fell short in this instance. I want to apologize for that. The feature itself was poorly executed; it had low usage and was misaligned with our strategy. We decided to discontinue it quickly after receiving feedback, well before any legal action was initiated.

How many people were involved in the decision to launch this feature?

It was a small team, likely just a product manager and a couple of engineers.

Given your decision-making process, how did it not come up that using people’s names without permission could lead to backlash?

The team was motivated by user feedback and the desire to enhance the Grammarly experience. They believed users wanted insights from experts, but ultimately, the execution didn’t meet expectations.

How do you view the issue of compensation for using someone’s name or likeness?

Attribution is crucial, and using someone’s work without permission is a different matter. We believe in giving credit where it’s due, but we also maintain that our feature did not constitute impersonation. We clearly indicated that the suggestions were inspired by specific individuals and their works.

If you were to use my likeness, how much should you pay me?

We should not impersonate anyone. If we use someone’s work, we should attribute it properly. The idea of impersonation is a separate issue, and we believe our feature did not cross that line.

Let’s pivot to public sentiment about AI. Recent polls show that people have a negative perception of AI. What do you think is driving this?

People are understandably concerned about job security. Many fear that AI will replace their roles, which contributes to the negative perception. However, I believe AI can create new opportunities and enhance productivity rather than eliminate jobs.

You’re in the midst of a controversy regarding the appropriation of work. How do you reconcile that with the broader public sentiment?

Creators are facing significant challenges, and I empathize with them. However, the general public’s concerns about AI are more about job displacement than about creators’ rights. We need to help people understand how AI can be a tool for empowerment rather than a threat.

As we wrap up, what’s next for Superhuman?

We’re focused on expanding Superhuman Go and have several exciting launches planned in the coming months.

Thank you for joining me today, Shishir. I appreciate your insights and willingness to engage in this important conversation.