Historical case v0.2 · Prepared
Therapy: what Illinois chose to keep human in 2025
Human by choice · One human choice
Booking an appointment, preparing notes and responding to someone's distress are different tasks. A rule about AI in therapy should tell us which of them it protects—and what people seeking help might gain or lose.
Imagine a service considering three changes: software to organise appointments, a tool to help prepare a therapist's records, and a chatbot to conduct the therapeutic conversation. Calling all three “AI therapy” conceals the choice. This is an invented situation, not a report about a particular service.
Illinois offers a concrete comparison. Its 2025 Wellness and Oversight for Psychological Resources Act took effect on 1 August 2025. This piece examines the rules enacted in 2025. Later amendments and the current consolidated law have not been verified. Human by choice has not adopted an equivalent UK policy.
1. Which task stays with a person?
The enacted Act required covered therapy services to be conducted by the licensed professionals it defines. It prohibited AI making independent therapeutic decisions, communicating therapeutically with clients or detecting emotions or mental states. Recommendations and treatment plans required professional review and approval; other restrictions still applied.
Permitted AI assistance included scheduling, billing, records and organising referrals. The professional remained responsible. Supplementary use involving session recording or transcription required written information about the AI and its purpose, and consent. Exceptions included religious counselling, peer support and non-therapy self-help resources. Its catch-all professional category expressly excepted physicians. IDFPR could investigate and impose civil penalties up to $10,000 per violation after a hearing. These details matter when describing the law's boundaries. Read sections 10–35.
The distinction combines two approaches: reserving some work for people while permitting assistance around it. Professional review is a condition on particular assistance, rather than a substitute for every retained human role.
2. What would keeping a person contribute?
There are at least two arguments to separate. One values the relationship: another person listening, responding and remaining involved. The other values answerability: someone with professional responsibilities who can explain and change what happens. Human by choice calls these human presence and answerability. They can support different rules.
A person checking a draft record might supply meaningful control without having typed every word. A person approving a chatbot conversation afterwards has not personally provided the interaction. The relevant question is what the person contributes, not whether their name appears somewhere in the process.
Illinois's regulator presented the law as protecting patients, including children, and qualified providers' jobs. That stated intention is not evidence that the law has improved outcomes. Nor does preserving a professional role establish that every human practitioner performs well, or that software could never help.
3. Could assistance make more room for people?
Consider a different invented service design. A therapist uses software to organise material, checks it and retains time with the client. Compare that with a service that uses the same software to increase appointments while leaving less time to check each record. Both have a human professional; their working conditions could be very different.
Assess the whole arrangement: preparation, checking, conversation, follow-up and handling errors. Time saved in drafting may be consumed in verification. A promise of professional responsibility needs enough time, relevant information and authority to reject an output. It should not become an excuse to transfer every technology risk to an individual worker.
4. Who gains, who pays and who waits?
The strongest access objection deserves a serious answer. If a reliable digital option could offer useful support sooner or at lower cost, excluding it could deny something valuable to people unable to obtain human care. Some people might prefer text, flexibility or less interpersonal exposure. Those are possibilities to investigate, not findings about everyone's preferences or claims that today's chatbots provide effective therapy.
A human-role rule therefore needs a resource plan. In a hypothetical NHS proposal, would the commissioner fund additional staff and checking time, reduce another service, or accept a longer queue? In private provision, would clients pay higher fees? Who would bear the software, training, privacy and monitoring costs? A protected service that only affluent people can reach would fall short of an ambition to share AI's gains.
Workers' interests also differ. Preserving a therapist's role does not guarantee pay, manageable caseloads or secure employment. Administrative assistance could change other staff members' jobs. Any productivity gains could become more time with clients, shorter waits, better conditions or higher provider margins. The rule alone would not decide their distribution.
5. Who could make a comparable choice here?
Start with a setting, rather than announce a UK-wide ban. For an NHS-funded service in England, one route to examine is an agreed service specification. NHS England's 2026/27 contract guidance, section 36, assigns commissioners responsibility for developing specifications with clinical engagement and encourages service-user involvement. Local requirements cannot override national contract terms.
A proposal could ask an integrated care board and provider to specify the therapeutic interaction retained with people, permitted assistance, checking time and evidence for monitoring compliance. That is a proposed contractual route to investigate, not an existing entitlement or a rule covering every private app. It would need procurement and legal assessment, funding and participation by people using and delivering the service.
Professional standards are another route within the relevant body's remit. But the HCPC distinguishes protected titles from terms such as counsellor and psychotherapist, which do not require its registration. One regulator cannot be assumed to cover the whole market. A general prohibition would require government to establish the appropriate legal powers and territorial scope, with legislation where necessary. Any UK-wide proposal would also need to establish its route in each territory, rather than extend this English contract model by assumption.
6. What would change the answer?
Compare specific arrangements for a defined group and purpose: human delivery, human delivery with assistance, and any lawful alternative being considered. Examine outcomes and harms, who drops out, waiting time, full costs, worker workload and people's own accounts of the relationship. Establish privacy-respecting, independent error reporting and a way to stop an unsafe arrangement. This is a proposed evaluation agenda, not a validated clinical protocol.
Evidence that assistance releases time for valued interaction would favour that design. Evidence that checking is ceremonial, or that access deteriorates for particular groups, would count against it. Reliable evidence for a different arrangement could justify revisiting a restriction; an average outcome alone would not settle how much people value human presence.
Before adopting a rule, state its beneficiary and reason, how it works and evasion is detected, who pays, and when it will be reviewed. Is it a lasting commitment to a human relationship, or a temporary measure to give workers time to adjust? Those need different justifications. Illinois's enacted choice makes the question concrete. It does not decide the answer for the UK.
Human by choice is Policy Window's advocacy programme on AI, work and who decides. Its analysis and proposals are distinct from the neutral Policy Window wiki. This case applies the programme's six questions; it does not originate the idea of reserving work for people.