Technology 01 · WBE
Written by the team
Could an emulated human brain improve alignment or preserve human agency?
Whole brain emulation would create digital minds derived from human brains. We assess the claim that those minds would improve alignment or preserve human agency. How exactly they would do that is still unresolved, and so are WBE's feasibility and its welfare implications.
Where we land
Full scorecardAll six ratings are Unclear for now. For four of them, the reason is that the proposal isn't specific enough to assess.
- Alignment case
- Intelligence CurseThe risk that increasingly capable AI reduces the economic and political value of human work and participation, and so weakens human agency. case
- Human welfare risk
- Artificial-mind welfare and rights risk
- Governance risk
- Safeguard adequacy
How to read the marks
- From a sourceQuoted word for word from a cited source.
- Our readThe team's own interpretation.
- AI-suggestedSuggested by an AI model, set in a dashed outline.
- UnclearWe don't know yet, and the tag gives the reason.
At a glance
The team's read unless marked
- What it claims to solve
Both alignment and the Intelligence Curse
- How it would help alignment
Potentially value specification, oversight or interpretability.
- Strongest case
The team hasn't written the strongest case yet.
- Strongest objection
Conventional alignment research may solve the relevant problems sooner.
From the team's scorecard counterarguments.
- Artificial minds
Unclear. We haven't found a source that assesses whether an emulation would be conscious or sentient.
- Main welfare risks
Consent violations, harms to identity and continuity, copying or exploitation, suffering of digital minds, concentration of power, value lock-in and unsafe deployment.
How it would work
The team's read unless marked
What it claims to solve
From the team’s scope
Both alignment and the Intelligence Curse
The steps it depends on, in order
Brain scanning, functional modeling and simulation become accurate enough.
A human brain is emulated, with legitimate consent and governance arrangements in place.
The emulation is deployed as an AI system, an overseer, a researcher or a contributor.
Its human-derived cognition improves value specification, oversight or institutional decision-making.
Safeguards prevent exploitation, abuses of copying, welfare harms and concentration of power.
Strongest case and objections
The team's read unless marked
For
The strongest case
The team hasn't written the strongest case yet.
Against
The strongest objections
From the team's scorecard: the most credible alternatives and counterarguments.
- 01
Conventional alignment research may solve the relevant problems sooner.
- 02
WBE may end up advancing AI capabilities more than safety.
- 03
Regulatory, moral and technical bottlenecks may keep it from being deployed in time. S02 (Thiergart 2023) points to the regulatory and moral bottlenecks explicitly.
- 04
S13 (Mineault et al. 2025) presents brain emulation as one of several underdeveloped NeuroAI pathways, and makes no case that it's an established safety solution.
Artificial minds
The team's read unless marked
- Consciousness
Unclear. We haven't found a source that assesses whether an emulation would be conscious or sentient.
- Moral status
Unclear. S02 (Thiergart 2023) identifies moral bottlenecks but doesn't set out a framework for moral status or a standard for protecting emulated minds.
- Identity
Unclear. How an emulated mind relates to the original person isn't addressed in the sources we have.
- Welfare
Unclear. The proposal could create digital minds whose welfare is at stake, but no source assesses suffering, consent, copying, confinement, acceleration, deletion or labor conditions.
Welfare and governance risks
The team's read unless marked
- Who could be harmed
Emulated minds, if any are created; people whose brains are scanned; users; workers; the wider public; and future generations.
- Possible harms
Consent violations, harms to identity and continuity, copying or exploitation, suffering of digital minds, concentration of power, value lock-in and unsafe deployment.
- Severity and reversibility
We haven't assessed it yet. It could be high, and hard to reverse if emulations can be copied or modified, or can persist indefinitely.
- Detectability and remedies
Unclear. Our sources don't specify any way to detect harms or remedy them.
- Control, oversight and exit
Unclear. No governance arrangements have been specified.
- Confidence
- Low
- Basis
S02 (Thiergart 2023) identifies moral and regulatory bottlenecks, and S15 (Drago 2025) describes the risks of concentrated power and of less investment in people. But we don't have enough WBE-specific welfare or governance evidence to tell likely harms from speculative ones.
See Thiergart 2023 (opens in a new tab); Drago 2025 (opens in a new tab)
Milestones
None of the 8 has a date yet, and nothing is mapped at the Actually helping stage. Each row opens the milestone on the timeline.
- WBE-01A human brain emulation reproduces relevant behavior in controlled testsStage: Possible
- WBE-02An emulation behaves stably and reproducibly across relevant tasksStage: Possible
- WBE-03A defensible welfare and consent framework exists before any emulation is createdStage: Possible
- WBE-04Cost and infrastructure allow controlled research or deploymentStage: Widely available
- WBE-05A cost-benefit case holds up without exploitative labor conditionsStage: Widely available
- WBE-06Clear legal authorization and safeguards existStage: Widely available
- WBE-07Independent oversight and legitimate authorization processes existStage: Widely available
- WBE-08Access and returns reach beyond a narrow groupStage: Widely available
Full assessment
The full scorecard with the basis for each rating, the scope, the evidence register, the alignment questions and the Intelligence Curse questions.
Show the full assessmentHide the full assessment
The team's read unless marked
Full scorecard
The ratings use the scales in the team’s draft scorecard. If an Unclear risk could be severe, its row also says how severe. “Not claimed” means the proposal doesn’t claim that outcome, so there is no case to grade. We never rate a risk Low just because nobody has studied it, and the cutoffs for the risk ratings aren’t set yet. How we rate
| Dimension | Rating | Confidence | Basis |
|---|---|---|---|
| Alignment caseCase · Alignment | UnclearUnderspecified proposal | Low | S02 (Thiergart 2023) and S13 (Mineault et al. 2025) both name WBE as a possible pathway for alignment, but neither shows how an emulation would solve a specific alignment problem. The evidence we have isn't enough to tell whether WBE is a promising research direction or an intervention that matters for safety. See Thiergart 2023 (opens in a new tab); Mineault et al. 2025 (opens in a new tab) |
| Intelligence Curse caseCase · Intelligence Curse | UnclearUnderspecified proposal | Low | S15 (Drago 2025) describes the risk that powerful actors lose their incentive to invest in people. None of our sources says how WBE would help, or shows that it would counteract that risk. Who would own the scans, the emulations, the compute and the labor they produce is unresolved. |
| Human welfare riskRisk | Unclear, Potentially high severityNo relevant evidence found | Low | Scanned people, users, workers, the wider public and future generations could all be exposed to harm. |
| Artificial-mind welfare and rights riskRisk | Unclear, Potentially high severityNo relevant evidence found | Low | The proposal could create digital minds whose welfare is at stake. |
| Governance riskRisk | Unclear, Potentially high severityUnderspecified proposal | Low | No governance arrangements have been specified, and the sources we have describe no way to detect or remedy harms. See Thiergart 2023 (opens in a new tab); Drago 2025 (opens in a new tab) |
| Safeguard adequacyCase | UnclearUnderspecified proposal | Low | The theory of change depends on safeguards against exploitation, abuses of copying, welfare harms and concentration of power. |
Grading key
Main gap in the evidence
Our current sources don't include enough technical, empirical, welfare or governance literature to assess WBE as an intervention for alignment or the Intelligence Curse.
Scope
- Intervention
A high-fidelity digital emulation of a human brain, which might be used as a human-derived AI system or as an overseer.
- Deployment scenario
Research, then deployment if it turns out to be feasible: emulated minds working as trusted AI systems, as overseers or as contributors to alignment research.
- Who it affects
Emulated individuals, if any are created; the people whose brains may be scanned; developers and operators; users of AI systems; and society more widely.
- Time horizon
Long term. The current evidence doesn't support any defensible near-term date for deployment.
- Compared with
Conventional AI alignment methods, including interpretability, evaluations, oversight, control and value learning.
- Claims to address
Both alignment and the Intelligence Curse
Evidence register
Each row is one claim, with the sources for and against it, the kind of evidence, its directnessHow close a piece of evidence comes to the scenario being assessed: direct, adjacent or remote., its main limitation and the sources’ independenceWhether the author or funder has a stake in the result..
| Claim | Supporting | Contrary | Evidence type | Directness | Main limitation | Independence |
|---|---|---|---|---|---|---|
WBE could contribute to alignment. | S02S13 | None in our current sources | Argument / review | Remote | No one has shown that an emulation would be safer, aligned or deployable. | Mixed |
Regulatory and moral bottlenecks could delay WBE. | S02 | None in our current sources | Argument | Adjacent | A workshop-level claim, with no policy analysis or empirical test behind it. | Unclear |
Neuroscience-inspired methods may support safety. | S13 | None in our current sources | Research roadmap | Remote | It doesn't show that WBE in particular improves alignment. | Independent |
WBE could preserve or distribute human agency. | None in our current sources | S15 | Argument | Remote | No evidence links WBE to broad ownership, bargaining power or institutional change. | Not applicable |
The Supporting and Contrary columns list sources by the team’s register IDs. Each ID links to the source’s card in the source database.
The alignment case
The alignment problem it targets
Potentially value specification, oversight or interpretability.
Does the mechanism address that problem, or move it somewhere else?
Unclear. An emulation could inherit human values, but it could also inherit human biases, conflicts, limitations and governance failures.
Could a more capable AI undermine or get around the benefit?
Unclear. Nothing we've read tests whether an overseer based on an emulation would stay effective against a more capable AI.
Does it create another agent or decision-maker that would itself need to be aligned or governed?
Yes, potentially. An emulated mind would itself need moral treatment and governance, and possibly alignment safeguards too.
What's the benefit relative to the comparatorThe alternative a proposal is measured against. Here: conventional AI alignment methods, including interpretability, evaluations, oversight, control and value learning.?
Unclear. There's no direct comparison with conventional alignment methods to go on.
- Rating
- UnclearUnderspecified proposal
- Confidence
- Low
Decisive open question
Which specific alignment failure would a WBE-based system solve better than alignment methods that don't use WBE, and what evidence would show that?
The Intelligence Curse case
The mechanism it claims
Potentially wider ownership of productive AI capacity or of human-derived labor. Our current sources don't specify the mechanism.
Who gains lasting agency or bargaining power?
Unclear. It could be emulated individuals, biological people, developers or the owners of compute.
Who owns and controls the infrastructure, the data and the returns?
Unclear. Ownership of the scans, the emulations, the compute and the labor they produce is unresolved.
Does it complement human contributions, delay replacement or deepen dependence?
Unclear. WBE could expand human-derived capacity, or it could create a new class of digital labor that's open to exploitation.
What's the benefit relative to the comparator?
Unclear.
- Rating
- UnclearUnderspecified proposal
- Confidence
- Low
Decisive open question
What arrangements for ownership, consent, labor and access to compute would let WBE increase broad human agency without creating a new class of dependent digital workers?