Mira Murati still he wants to build AI intelligence. But the former OpenAI CTO sees human intelligence as an important part of the equation.
A time of growing concern about AI eliminate jobs and increasing the power of a few large companies, the beginning of Murati, Laboratory Thinking Machinesoffers a completely different vision of technology.
“At some point we’ll have super-intelligent machines,” Murati tells WIRED. “But we think that the best way to have the most possible future – the best future – is to keep humans in the loop.”
Murati says AI doesn’t need to automatically put humans out of the equation. The most promising approach, he suggests, is to let people build and customize their own models of AI boundaries, then work with those models to achieve their goals.
This week, Thinking Machines previewed a new type of AI model that it says is pointing toward a more human-inclusive reality. The company’s “interaction models” are trained to communicate with a person through a camera and microphone. Unlike many existing voice-mode interfaces, the new models don’t just capture and transcribe speech, then feed it into a language model that processes it in the same way as a chatbot. Interaction models naturally understand continuous, messy, human communication—meaning they can better understand the meaning of pauses, interruptions, and changes in tone. This allows them to get used to skipping when someone explains something or changes the subject. The company showed several videos demonstrating this capability, although the examples have not been made public.
Murati’s approach stands in contrast to how many large AI companies seem to be pursuing intelligence today. OpenAI, Anthropic, Google they are developing large models that perform increasingly complex tasks, including writing entire applications from scratch, via text notification. This requires little help from a human.
Thinking Machines isn’t the only startup envisioning a more inclusive future. Other labs, including Human&, also aim to develop AI systems that prioritize human interaction. Some prominent economists have called for AI researchers and companies building systems in this way, focus on human empowerment rather than replacement.
Murati left his role as chief technology officer of OpenAI in 2024, co-founding Thinking Machines with several prominent engineers. The Thinking Machine has raised billions of dollars building the AI frontier.
So far, however, the company has only released one product. Tinkerlaunched in October 2025, it makes it possible to improve the AI design of the frontier using special data. Today, it is available as an API that researchers and engineers can use to standardize open source models.
Alexander Kirillov, a founding team member of Thinking Machines and an expert in multimodal AI, meaning models that handle audio and video and text, says the lab’s new interactive models also have the potential to enable more personalized and personalized AI.
“The model always recognizes what you’re doing and is always there to respond and give you information or search for information or use other tools,” Kirillov says. “This is something that none of the models (today) can do. The turn (in conversation) is determined by a system that is not very intelligent.”
Mira says it’s all part of a bigger AI vision.
“This represents the first bet on human cooperation,” he explains. “Where this is going is developing people’s own preferences and values, and AI understanding intent and predictive intent.”
