Block, the company behind Square and Cash App, has unveiled Dessa — a new artificial intelligence model that simulates human speech using only audio. Unlike most voice AI tools, Dessa doesn’t convert speech to text first. It processes audio directly, making interactions feel more natural and emotionally responsive. Source: coindesk.com.
What Happened
Block introduced Dessa during a technical preview. Live demonstrations showed the AI listening and responding in real time. It handled interruptions, laughter, and tone shifts — behaviors that mirror human conversation.
The model was trained from scratch using a single Nvidia GB200 NVL72 system. This suggests a focused, high-performance setup, optimized for efficiency in early development stages.
How Dessa Works
Dessa processes audio directly, not through text. This lets it capture subtle emotional cues like hesitation, surprise, or joy — features often missed in text-based AI.
In one demo, the AI laughed mid-sentence. This natural reaction mimics how humans respond to humor or surprise. The interactions were recorded live, with no human prompting or direction.
The model runs locally on a MacBook Pro with an M4 Max chip or better. This means users don’t need internet access or cloud servers, improving privacy and reducing latency.
Public Reactions
Reactions have split sharply. Some users were amazed by how lifelike the voice sounded.
“I’ve never heard anything artificial that sounded so human,” said X user Alexander Hoff.
“Straight up sci-fi stuff,” wrote user Hesamation.
Jack Dorsey, CEO of Block, called it “talking with a person…full of emotion, nuance, pauses, and a natural cadence.”
Changpeng “CZ” Zhao, Binance founder, sees potential in education. He referenced his project GiggleAcademy. “Teaching kids in current ‘AI voice’ was a challenge,” he wrote. “Dessa’s natural tone could make learning more engaging for children.”
Others raised alarms. Filmmaker Justine Bateman warned that human-like AI could be used to deceive people. “So you can trick people into lowering their guard,” she wrote.
Journalist Shiv Aroor predicted scam calls could reach pandemic levels. “Scam calls about to attain pandemic level,” she wrote. This reflects growing fears about fraud as AI voices become harder to detect.
Block’s AI Background
Block’s AI research lab reports directly to Jack Dorsey. It includes former staff from Toronto-based startup Dessa, acquired by Square in 2020. This gives the team deep expertise in voice AI development.
Block has not shared Dessa’s model size, performance benchmarks, or a public release date. The current version is a research preview, available only to a small group of AI safety researchers.
Why It Matters
Dessa represents a shift in AI design. It prioritizes emotional authenticity over linguistic accuracy. By processing audio directly, it may better detect emotions like distress or pain — cues often missed in text-based systems.
This could improve mental health support, customer service, and education. For example, an AI that senses anxiety could respond with empathy, improving user experience.
Yet, the technology raises serious ethical questions. If AI voices sound human, how do we prevent misuse? What rules exist to stop impersonation, fraud, or manipulation?
Limitations and Unknowns
There is no public data on Dessa’s accuracy, reliability, or long-term performance. It hasn’t been tested in real-world settings. Behavior under stress or edge cases is unknown.
Block has not confirmed whether Dessa can impersonate individuals or generate content that mimics real people. These gaps leave misuse concerns unaddressed.
No formal oversight or ethical guidelines exist for how Dessa might be used in deceptive or harmful ways.
What to Watch For
Block will offer early access to a small group of AI safety researchers. These experts will test the model, evaluate emotional detection, and identify risks.
Public release is not expected soon. No date has been set. As Dessa evolves, questions about its use in education, customer service, and personal interaction will grow. So will fears about fraud and deception.
The technology is still experimental. Its real-world impact depends on governance, testing, and ethical standards — and on whether the industry agrees on how to use human-like AI responsibly.
Sources & further reading
Featured image: original publisher image. Via original publisher. Image source
