Thinking Machines and Mira Murati: Towards a More Human AI

Thinking Machines Announce a More Human-Like AI

What are "interaction models"?

Mira Murati, former CTO of OpenAI, is bringing a new dimension to the future of AI experiences. Innovation in AI is happening in many recent research labs, which has huge implications for the future of AI. What if AI models could provide more human-like experiences? As a frequent user of voice AI, I have often pondered this question. This week, we finally got to see what Thinky is preparing.

Thinking Machines Lab, the AI startup founded last year by Mira Murati, has managed to attract some of the best AI talent from Silicon Valley. Thinky aims to move beyond the era of "turn-based" AI interactions, and I believe they are on the right track. If they succeed, it will enable much more genuine and human-like AI interactions in real time. Think about it: traditional AI models operate sequentially—listening and then responding—while their new model aims for simultaneous processing of inputs and response generation, akin to a phone call.

Interaction Models

While Lilian tells a story, the interaction model can track when she thinks, hesitates, corrects herself, or invites a response; there is no specific dialogue management system integrated. Interaction models represent a new category of AI systems with a micro-turn architecture, a system that processes data in chunks of 200 ms. This could literally make AI interactions much more useful in the physical and social world we live in. It’s certainly more human-like, Thinky! Murati is no novice; she is simply Albanian, and the technical term for this is "full duplex." The company claims that its model, TML-Interaction-Small, responds in 0.40 seconds, which is roughly the speed of natural human conversation and significantly faster than comparable models from OpenAI and Google.

I found this story particularly intriguing, as it clearly points to many other interesting AI products to come. If you are a builder, developer, investor, or advanced AI user, this is incredibly exciting.

A More Natural AI Interface Emerges

Essentially, AI should serve us, but we have been forced into poor interactions. There is still much work to be done. Thinking Machines argues that back-and-forth interactions with current models force human users to "contort" themselves to fit the interface. If you are a regular user of voice AI, you know exactly what I’m talking about.

Could we have a benevolent AI that genuinely looks out for us, rather than trying to deceive us with flattering words? Tessa's quality of life has greatly improved with a few reminders. I found that quite amusing, and I would agree with an AI that reminds me of things in a positive way. Can we envision a time when AI would be a benevolent advisor and actor within human institutions, helping to guide humanity through civilization-scale challenges? Perhaps, like me, you are familiar with the concept of an "Oversoul" of artificial intelligence, explored by author Orson Scott Card in his five-volume science fiction series, the Homecoming Saga.

For me, interaction models point toward this world, perhaps decades away.

The Potential of Interaction Models

So, what would interaction with an AI that is not so false and artificial look like? Interaction models have real potential to humanize AI in a positive way. These models are essentially a new AI system designed to enable real-time, native collaboration across audio, video, and text, without relying on external structures. One can almost sense the feminine touch in Thinky’s product orientation. AI doesn’t have to be horrible, flawed, or flattering. Is there a better way to represent and build products with this technology?

Simultaneous "Full Duplex" Processing

Thinking Machines is essentially helping to solve the "collaboration bottleneck" in our relationship with AI in a more realistic manner.

“We believe that interactivity should evolve with intelligence; the way we work with AI should not be considered an afterthought. Interaction models allow people to collaborate with AI as we naturally collaborate with each other—they continuously take into account audio, video, and text, and think, respond, and act in real time.” - Thinking Machines Blog.