Vitalik reveals AI future blueprint: rejecting full automation, advocating for human real-time control and brain-computer interfaces

Vitalik Buterin, co-founder of Ethereum, recently posted on X (formerly Twitter), bluntly stating that the current development of artificial intelligence (AI) overly pursues "autonomy" while ignoring the value of human input in the decision-making and output processes. He called for the industry to increase human input pathways, promote open-weight models with editing capabilities, and proposed a concept of combining brain-computer interfaces (BCI) to achieve real-time feedback and adjustments, outlining a more human-centered blueprint for AI development.

Reject "fully automated agent-type" AI, emphasizing human involvement

Vitalik stated that the current trend of "agent-based" systems in the AI field, which attempt to complete tasks autonomously with minimal human supervision, frustrates him. He believes that human input not only enhances the quality of output but also significantly improves safety.

He particularly prefers open-weight models with "editing capabilities," allowing developers and users to intervene, correct, or guide the output in real-time during the generation process, rather than having to create results from scratch.

Echoing Karpathy's point of view: AI needs "controllability"

Vitalik's views align with those of former Tesla AI head Andrej Karpathy in a recent talk. Karpathy warned that large language models should not be seen as fully reliable autonomous agents, emphasizing the importance of human oversight, cautious prompting, and incremental development.

This idea is leading the industry towards a "human-machine collaboration" model:

1. Provide a more intuitive editing interface

2. Support real-time trimming and correction during the generation process.

3. Reduce illusions and biases to ensure that the output aligns with human intent and safety standards.

Brain-Computer Interface (BCI) + AI: The Future Vision of Instant Feedback

Vitalik further proposed that the medium-term vision is to integrate BCI technology, allowing AI to instantly perceive users' emotional responses while generating content, and automatically adjust the output based on preferences.

This concept is not mere talk:

1. Companies like Neuralink and Synchron have initiated human trials and minimally invasive implants.

2. Wearable devices based on electroencephalography (EEG) are attempting to decode attention and emotional states.

3. The academic community has made progress in "affective computing" and emotion tagging technologies.

4. Although non-invasive emotional decoding still poses challenges, technological advancements are gradually bringing this vision closer to reality.

Open Weights and Security: The Double-Edged Sword Debate

Recently, OpenAI unexpectedly released open-weight models such as gpt-oss-120b and gpt-oss-20b, sparking heated discussions. Supporters believe that this helps democratize AI development, making it easier for teams to establish human-centered workflows and assess security risks; however, critics worry that the widespread distribution of powerful models may lead to abuse risks.

Vitalik's position is clear: better outputs and safer systems come from placing human preferences and complex signals at the core of AI workflows.

Conclusion

Vitalik Buterin's voice reflects a core debate in the development of AI: should models be fully autonomous, or should humans maintain real-time control? With the advancement of open-weight models, brain-computer interfaces, and emotion detection technologies, the possibilities for human-machine collaboration are rapidly expanding. The future of AI may no longer be a black box that "makes decisions for you," but rather an intelligent partner that can create alongside you and interact in real-time.