The Woman Behind the Vision
Mira Murati spent years as one of the most influential figures in artificial intelligence, serving as Chief Technology Officer at OpenAI before stepping down in September 2024. Now, her new venture — Thinking Machines — is beginning to show the world what she's been working on, and it's a fundamental rethink of how AI systems experience reality.
What Are Interaction Models?
On Monday, Thinking Machines announced it is developing what it calls "interaction models" — a new class of AI designed to collaborate with people the way humans naturally collaborate with each other.
The core idea is surprisingly simple but technically ambitious: current AI models experience the world in a single thread. They wait, passively, until you finish typing or speaking before doing anything. They have no awareness of what you're doing while you're doing it — no sense of hesitation, no ability to read the room in real time.
Interaction models, as described by Thinking Machines, would change that entirely. These systems would continuously take in audio, video, and text simultaneously, then think, respond, and act in real time — not after the fact.
Think of it less like texting someone and more like talking to a colleague across a desk who can see your screen, hear your tone of voice, and jump in naturally when the moment is right.
Why This Matters
The announcement is notable for a few reasons. First, it signals that Murati's team is focused not just on making AI smarter in the traditional benchmark sense, but on making it feel more like genuine collaboration.
Current large language models — even the most advanced ones — are fundamentally reactive. You prompt, they respond. The interaction model concept pushes toward something more proactive and perceptually rich: an AI that can process a video call, hear uncertainty in your voice, watch you struggle with a spreadsheet, and offer help without being explicitly asked.
Second, it shows Thinking Machines is willing to stake out territory that's distinctly different from what OpenAI, Google, and Anthropic are building. Rather than racing to score highest on benchmarks, the company seems to be betting on a new interaction paradigm altogether.
Still Early Days
Thinking Machines hasn't released a product yet, and Monday's announcement was light on technical specifics. The company shared its vision and framing but stopped short of demos, timelines, or detailed architecture breakdowns. That's not unusual for an early-stage AI lab — these companies typically build quietly before going public with working systems.
Still, the framing is striking. The phrase "collaborate with AI the way we naturally collaborate with each other" is a high bar. Human collaboration involves eye contact, tone, intuition, and improvisation. Building AI that can genuinely participate in that kind of exchange — not simulate it — is one of the hardest open problems in the field.
What's Next
The AI industry will be watching closely. Murati's credibility in the field is significant — she helped ship GPT-4, DALL-E, and Codex during her time at OpenAI. If Thinking Machines can back up this vision with real technology, interaction models could become one of the defining concepts of the next wave of AI development.
For now, the company has given us a direction. The details, presumably, are coming.
Source: The Verge
