Subconscious
behind the conversation.
The next personal AI interface is not another participant in the thread. It is a private cognitive layer behind the user: remembering what matters, noticing what is relevant, and surfacing the next useful thought before the user has to ask.
This is a first-draft concept paper tied to Commons item TOS-396. Current reply-mode controls exist. The Subconscious framing, private ask surface, provenance copy, and driving-mode review affordances are product work to refine and implement.
A subconscious is not a chatbot, copilot, or executive controller. It sits behind the user. It holds context, patterns, memories, commitments, habits, latent intent, and standing preferences, then surfaces useful thoughts into the active moment.
| Role | Meaning | Status |
|---|---|---|
| Disabled | The subconscious is off for this context. | Planned · Phase 1 |
| Assisting | It can retrieve, suggest, draft, and answer private questions. | In progress · Phase 1 |
| Driving | It can act inside explicit delegated boundaries. | Planned · Phase 2 |
1. The interface behind the conversation
The question is not whether AI should join every conversation. Most of the time, that is the wrong shape. When a user is speaking to another person, they usually do not need a third voice in the room. They need access to the thinking behind their own reply.
That thinking may include memory, preference, judgment, relationship history, obligations, uncertainty, or the simple sense that something has happened before. In a normal interface, the user has to stop, search, ask a bot, or reconstruct the context. A subconscious should make that layer ambient.
Not a chatbot in front of you. Not a copilot beside you. A subconscious behind you.
The user remains the person in the interaction. The subconscious is the private cognitive system helping them remember, decide, respond, and act.
2. The problem with chat as the default interface
Most AI products treat interaction as a chat. The user opens a window, asks a question, and receives an answer.
That model works when the user is explicitly consulting a system. It breaks down when the user is already inside another interaction: replying to a person, negotiating a decision, remembering a commitment, or deciding what to say next.
In those moments, the user may need to ask:
- When did I promise this?
- What happened last time?
- What should I remember before answering?
- How would I normally respond?
- Is there a better way to say this?
- Should I let the system handle this one?
Those are private cognitive queries. They should not become messages in the conversation. They should happen behind the thread.
3. Subconscious, not assistant
An assistant is external. It waits for instructions.
A chatbot is conversational. It becomes another voice in the room.
A copilot sits beside the user and helps operate a visible task.
A subconscious is different. It is private by default. It does not need to announce itself as a speaker. It watches context, retrieves memory, detects relevance, and offers latent thoughts at the edge of action.
This distinction matters most in communication. When the user is talking to another person, the subconscious should not feel like a third participant. It should feel like the user has access to a deeper private layer of themselves: memory, pattern, preference, intention, and judgment.
4. The three roles
The product should be organized around a simple question:
What role is your subconscious playing?
This replaces older control language like “augmented mode” and “autopilot” with a model that maps to the user’s cognitive experience.
Disabled
The subconscious is off for this context. It does not observe, suggest, draft, retrieve, or act. The conversation proceeds without cognitive assistance.
This is important for privacy, emotional clarity, and user control. Sometimes the right mode is no subconscious.
Assisting
The subconscious is active but not in control. It can suggest replies, surface memories, identify relevant context, remind the user of commitments, and answer private questions. The user remains the visible actor and decides what to send.
This is the successor to augmented mode.
Driving
The subconscious is authorized to act. It can answer routine messages, handle low-risk interactions, follow known preferences, or execute defined communication patterns.
This is the successor to autopilot. The important framing is not that an AI is taking over. It is that the user has allowed their subconscious to drive in a particular context.
5. Asking your subconscious
A core interaction is the ability to ask the subconscious a question while remaining inside another context. This should not feel like sending a message into the main chat.
If the user is talking to someone else, asking the subconscious should feel private, lateral, and behind the scenes. The surface might be a side panel, an inline drawer, or a composer-level affordance. The principle matters more than the first shape: the answer is private unless the user turns it into an external action.
Examples:
- “When did I last do this?”
- “What did I tell them before?”
- “Do I owe them anything?”
- “What was the decision we made?”
- “How should I answer without overcommitting?”
- “Draft the answer, but do not send it.”
The answer should be visibly separate from the conversation itself. It is closer to asking yourself a question and having the relevant memory surface.
6. Suggestions from the subconscious
When the subconscious supplies an answer, the UI should avoid making it feel like “the AI said this.” The better framing is a surfaced thought.
Possible labels:
- Your subconscious suggests:
- Suggested from your subconscious:
- From your context:
The label should be light enough not to compete with the message, but clear enough that the user understands where the suggestion came from.
Expected actions include:
- Insert suggestion
- Edit suggestion
- Ask why
- Ask what memory it used
- Send
- Let subconscious handle similar replies in the future
The most important affordance is provenance. The user should know whether a suggestion came from recent context, long-term memory, inferred preference, or a standing instruction. A subconscious that cannot explain why it surfaced something will eventually feel intrusive.
7. Driving mode
Driving mode should be treated as delegated subconscious action, not generic automation.
The user is deciding where their subconscious is allowed to act without interruption. That might include routine scheduling replies, simple acknowledgements, known follow-ups, low-priority declines, repetitive operational communication, or continuity while the user is unavailable.
Driving mode requires explicit boundaries:
- Where it is enabled
- Who it can respond to
- What kinds of messages it can handle
- Whether it can send immediately
- When it must ask for confirmation
- How it reports what it did
The framing should make delegated agency feel controlled, reversible, and inspectable.
8. Privacy and trust
A subconscious is powerful because it is intimate.
Privacy cannot be treated as a secondary feature. It is central to the metaphor. Users need to understand when the subconscious is disabled, observing, assisting, retrieving memory, drafting, acting, or driving without immediate review.
A subconscious that is always silently active without clear controls will feel invasive. A subconscious that is clear about its role can feel empowering.
The user should always know whether their subconscious is off, assisting, or driving.
9. Product implications
This framing implies five product changes.
- Rename and reframe conversation-level modes around the role of the subconscious.
- Expose a private “ask my subconscious” interaction inside a conversation.
- Present reply suggestions as surfaced thoughts from the user’s own context.
- Make provenance visible: memory, context, preference, or standing instruction.
- Make driving mode inspectable: what it handled, why it acted, and where it stopped.
The old product question was: should this conversation use augmented mode or autopilot? The new product question is: what role should my subconscious play while I am in this conversation?
10. Read together
This paper sits after the first four. The sequence matters because Subconscious is not a standalone assistant concept. It depends on the architecture beneath it.
- Paper 1, From memory to cognition. The maturity ladder: forgetfulness, memory, cognition.
- Paper 2, Your AI has amnesia. The architecture of the Mind and durable cognitive state.
- Paper 3, Two minds, one room. The user-side runtime and private deliberation boundary.
- Paper 4, How many minds do you need? Minds as coherence boundaries and federated focus.