Mira Murati still wants to build AI superintelligence. However, OpenAI’s former CTO sees human intelligence as a key part of the equation.
At a time of growing fears that artificial intelligence will eliminate jobs and escalate the power of a few huge companies, Murati’s startup, Thinking Machines Lab, offers a radically different vision of the technology.
“At some point we will have superintelligent machines,” Murati tells WIRED. “But we believe that the best way to ensure many possible futures – good futures – is to inform people about them.”
Murati says artificial intelligence doesn’t have to automate humans. He suggests that a more hopeful approach is to enable people to create and customize their own pioneering AI models, and then work with those models to achieve their goals.
This week, Thinking Machines unveiled a novel kind of artificial intelligence model that it says points to a more inclusive reality for humans. The company’s “interaction models” are trained to communicate with a person via camera and microphone. Unlike many existing voice interfaces, the novel models do not just capture and transcribe speech and then feed it into a language model that processes it in the same way as a chatbot. Interaction models natively understand continuous, messy human communication, which means they are better able to capture the meaning of pauses, pauses, and changes in tone. Thanks to this, they can constantly adapt to the situation when someone clarifies an issue or changes the topic. The company showed several videos demonstrating these capabilities, although the models have not been publicly revealed.
Murati’s approach contrasts with how most huge AI companies currently seem to be pursuing superintelligence. OpenAI, Anthropic and Google are developing huge models that do increasingly elaborate work, including writing entire applications from scratch using text prompts. This requires little human assistance.
Thinking Machines isn’t the only startup with a vision for a more inclusive future. Other labs, including Humans&, also aim to develop artificial intelligence systems that prioritize human collaboration. Some eminent economists they called AI researchers and companies will build systems this way, focusing on human empowerment rather than replacement.
Murati left her position as OpenAI’s chief technology officer in 2024 and co-founded Thinking Machines with several prominent engineers. Thinking Machines has raised billions of dollars to build pioneering artificial intelligence.
However, so far the company has only released one product. Launched in October 2025, Tinker enables you to refine your pioneering AI model using custom data. It is currently available as an API that researchers and engineers can exploit to tune open source models.
Alexander Kirillov, founder of the Thinking Machines team and an expert in multimodal artificial intelligence, i.e. models that support audio and video as well as text, claims that novel interaction models developed in the laboratory also have the potential to enable more individualized and personalized artificial intelligence.
“The model constantly sees what you are doing and is constantly present to be able to respond and provide information, search for information or use other tools,” says Kirillov. “This is something that doesn’t exist [today’s other] models can actually do this. Corners [in a conversation] are determined by a much less intelligent system.”
Mira says it’s all part of a larger vision for artificial intelligence.
“This shows the first step towards interpersonal cooperation,” he explains. “It actually goes towards amplifying people’s own preferences and values, with the AI actually understanding intentions and predicting them.”