British consumers were already drawing boundaries around commercial artificial intelligence in 2017. A survey commissioned by digital agency SYZYGY found that people could see value in personalisation while still wanting consent, disclosure and enforceable standards.
The study covered 2,000 adults in the United Kingdom. Ninety-two percent supported regulation or a code of conduct for marketing AI, and 75 percent wanted companies to obtain explicit consent before using it. Those results challenged the idea that convenient services automatically created permission for unrestricted profiling.
Personalisation Was Not Rejected Outright
The findings were more nuanced than a simple backlash against automation. Seventy-nine percent of respondents were comfortable with some form of profiling to receive more relevant advertising.
That suggested people distinguished between a useful recommendation and invisible surveillance. Relevance could create value when the exchange was understood, but the same process could feel intrusive when data collection was hidden or unexpectedly sensitive.
The difference depended on context, control and proportionality. A retailer remembering a preferred size was not equivalent to inferring health, politics or financial vulnerability.
Consent Needed to Be Meaningful
The demand for explicit permission reflected concern about how easily automated systems could combine browsing, purchase and location data. A buried clause in lengthy terms was unlikely to provide genuine understanding.
Meaningful consent required a clear description of the purpose, the data involved and the likely effect on the customer. It also required a realistic way to refuse without losing access to an unrelated service.
Companies needed to treat consent as an ongoing relationship. If a system began using information for a new purpose, customers should not be expected to discover the change by accident.
People Wanted to Know When AI Was Involved
One proposed principle was that companies should not conceal the use of AI. Disclosure mattered when a customer was chatting with a bot, receiving an automated decision or seeing content selected by a model.
Transparency did not mean explaining every mathematical detail. It meant giving people information they could act on: whether automation was involved, why an outcome occurred and how to reach a person when necessary.
This was particularly important for customer service. Twenty-eight percent expressed a negative view of AI in that setting, where an efficient answer could be welcome but a rigid bot could intensify frustration.
A Practical Code Could Set Behavioural Limits
The proposed guidelines were straightforward: do no harm, build trust, do not pretend to be human and provide genuine help. Their simplicity was useful because they focused on the customer experience rather than a particular technical method.
“Do no harm” extended beyond physical safety. Manipulative targeting, discriminatory offers and exploiting vulnerable customers could all produce damage.
“Be helpful” required measuring more than engagement. A system that increased clicks by creating confusion might meet a narrow metric while failing the person it was meant to serve.
Autonomous Systems Made Values Visible
The survey also asked about self-driving cars. Fifty-four percent said a vehicle should sacrifice its passengers if doing so minimised overall harm, yet 71 percent said they would not ride in such a car.
The contrast illustrated a recurring problem in AI governance. People might endorse an abstract social principle but make a different choice when they personally carried the risk.
These hypothetical dilemmas did not describe most real driving decisions, which involve uncertainty rather than neat choices. They nevertheless showed why companies could not determine acceptable behaviour through engineering alone.
Regulation and Trust Reinforced Each Other
Rules were sometimes presented as a brake on innovation. The survey suggested the opposite possibility: clear expectations could make people more willing to use AI-enabled services.
A code of conduct could establish minimum requirements for consent, disclosure, security, testing and human review. Formal law could provide remedies where voluntary commitments failed.
Companies that met those standards could compete on service instead of asking customers to trust vague assurances.
The 2017 Message Remained Relevant
Consumers did not have to understand model architecture to recognise when an automated experience felt unfair. They wanted useful personalisation, but they also wanted agency over their information and honesty about how decisions were made.
That combination offered a workable design brief. Collect only what is needed, explain the benefit, provide control and make escalation possible.
The central lesson was that acceptance of AI could not be inferred from usage alone. People often used convenient digital services while remaining uneasy about their operation. Sustainable adoption required companies to earn permission through transparent, bounded and genuinely helpful behaviour.
