AI Twin of the Deceased: Cloning Technologies

# AI Clone of the Deceased: How a Chinese Company Recreated a Digital Person for His Family

The family of a man killed in a car crash in Shandong Province created an AI clone of him to keep his death secret from his elderly mother, who has heart disease. Super Brain used photos, videos, and audio recordings to precisely recreate his appearance, voice, and mannerisms—including his habit of leaning forward during conversations. Daily video calls with the "son" let the mother remain blissfully unaware.

Technologies for Creating the Digital Avatar

Super Brain developers applied generative AI methods to synthesize a realistic likeness. Key components:

• Visual reconstruction: deepfake technologies based on neural networks processed photos and videos to create a face model with expressions and gestures.
• Voice synthesis: TTS models (text-to-speech) cloned his timbre and intonations from audio recordings.
• Behavioral AI: scripts and LLM (large language models) mimic his communication style, responding to the mother's typical questions about health, work, and plans.

The "clone" holds real-time conversations: promising to take care of himself, complaining about being busy, and dodging in-person meetings. Project lead Zhang Zewei emphasizes the ethical side—the technology offers temporary comfort but doesn't replace real relationships.

Ethical Dilemmas of AI Imitation

The case sparked debates on Chinese social media. Users are split:

• Some see value in delaying grief for vulnerable relatives.
• Others warn of psychological harm: a fake connection could intensify mourning when the truth comes out.

AI ethics experts highlight risks:

• Emotional manipulation without consent.
• Dependency on a digital ghost.
• Legal issues: rights to the deceased's digital copy.

In China, similar services are gaining traction. Super Brain positions the product as grief tech—a tool for bereavement support.

Development Prospects

The technology relies on advanced models like diffusion models for video and fine-tuned transformers for dialogue. For mid/senior-level developers, key challenges include:

• Handling edge cases in behavior (unpredictable questions).
• Reducing latency for seamless video calls.
• Integrating with AR/VR for greater immersion.

Example cloning workflow:

# Pseudocode: basic cloning pipeline

def clone_avatar(media_files):
    face_model = train_facegan(media_files['images'])
    voice_model = fine_tune_tts(media_files['audio'])
    behavior_llm = adapt_llm(media_files['transcripts'])
    return integrate_models(face_model, voice_model, behavior_llm)

Future versions could include emotion analysis to tailor responses to the interlocutor's mood.

Key Takeaways

• Super Brain created an AI clone based on photos, videos, and audio, mimicking the deceased's appearance, voice, and gestures.
• Goal: shield an 80-year-old mother with heart disease from shock; daily calls continue.
• Ethical debates: comfort vs. worsening grief and manipulation.
• Technologies: deepfake, TTS, LLM; grief tech popularity rising in China.

— Editorial Team