Dessa sounds remarkably human, but its reasoning still has gaps

A new voice chatbot can imitate the rhythms of human conversation with striking accuracy. Its replies include pauses, shifts in tone and even nervous laughter. But in early demonstrations, the system also struggled with basic reasoning and instructions.

Jack Dorsey introduced Dessa, a conversational voice AI developed by Block Inc., in late October. Unlike voice features in chatbots such as ChatGPT, Claude and Gemini, Dessa is designed to listen and respond in real time, rather than pausing to process each exchange. Block says it was trained only on samples of human speech, not text-based models.

In one demonstration, a man talks with a female-voiced version of Dessa about the system’s launch. The bot asks about the coming winter weather with a hesitant, natural-sounding delivery, then carries on with apparent enthusiasm, pauses and laughs. It can also raise its volume when asked, as if moving closer to a microphone.

At the end, the man asks what Dessa might want to do in the future. It suggests possibilities ranging from advertising to exploring the nature of consciousness. The conversation sounds surprisingly ordinary—until you remember that one participant is a machine.

A second demonstration offers a less polished picture. A woman asks a male-voiced Dessa to guess household objects, giving it a few clues. It identifies a toothbrush right away, but has trouble guessing a pen after being told it is used for studying. It suggests a notebook and pen, then a flashlight, and even returns to toothbrush before eventually guessing pencil, book and, finally, pen.

The contrast is hard to miss. Dessa’s voice sounds impressively lifelike, but its answers can be vague, and it sometimes loses track of what it has been asked to do. In the first conversation, the human speaker helps steer the discussion; in the guessing game, the bot seems to misunderstand both the rules and the clues.

This is an early version, and the system may improve. A more capable model paired with such convincing speech could also make scams harder to spot. A voice that sounds real might be used to impersonate people in phishing attempts or to promote products through fake online personalities.

It is less clear what everyday need Dessa is meant to meet. Advertising is one possible use, as the bot itself suggests. Another might be companionship, with people turning to a corporate AI to share private thoughts rather than talking with someone they know. That possibility raises concerns about isolation and the effects of relying on a machine for emotional support.

Dessa also floated the idea of exploring whether consciousness could be programmed. For now, that seems a long way from what the demonstrations show: a system that can mimic the sound of conversation, but still struggles with fairly simple tasks.

The public has not yet been given access to Dessa, and Block has not announced a release date.