In Robots We Trust - or do we?

On a recent edition of the BBC’s The Life Scientific, the physicist Jim Al-Khalili recalled being told off by a chatbot. Interviewing a Heriot-Watt system called Alana some years ago, he mentioned the science-fiction author Philip K. Dick — and the bot flagged him for being “inappropriate.” His guest, Professor Helen Hastie of the University of Edinburgh, has spent a career making sure the machines of the future earn a warmer response than that. Hastie studies human–robot interaction — specifically, the slippery, subjective, deeply human business of trust. Her work is a useful mirror to hold up to our own industry because hospitality is now buying robots by the container load. So it is worth asking the question that her research keeps circling back to: in robots we trust — or do we?

The scientist who teaches robots and humans to trust each other

Helen Hastie is Professor of Human-Robot Interaction and Head of the School of Informatics at Edinburgh, a Fellow of the Royal Society of Edinburgh, and co-founder of the National Robotarium — the UK’s largest applied research facility for robotics and AI. Her career runs from the early dialogue systems that were the ancestors of today’s generative AI, through to robots that serve coffee with small-talk, tutor struggling children with something like empathy, and act as calm, confident triage nurses.

Two of her lab’s creations should make every hotelier sit up. The first is a robot receptionist — a Furhat-based agent with computer vision installed at the National Robotarium that greets visitors, gives directions, registers attendees, and answers questions. The second is the Robo-Barista — an expressive robot head (named Alex in trials) wired to a good Italian coffee machine in the university common room, deliberately placed so people had a real reason to use it. It takes your order, reads your social cues, and makes small talk, and it ran for six weeks to study how guest attitudes shift over time. A robot concierge and a chatty barista, studied in the wild: this is not a thought experiment for hospitality. It is a preview.

Hastie’s Central Insight — In Her Own Words

She defines trust plainly: “the willingness to be vulnerable to another entity’s actions.” And the goal, she stresses, is not maximum trust but the appropriate amount: “it’s important not to over-trust robots, and it’s also important not to under-trust them.”

Her two examples say it all. A surgical robot that is more accurate than a human should be trusted and adopted. But a ship’s autopilot should never be over-trusted — you always keep a human eye on it.

Five things the science actually knows about trust

Strip away the marketing, and Hastie’s own research — much of it run live in the wild at the National Robotarium — converges on five findings. Each one has a direct, uncomfortable read-across to a hotel lobby.

1. Calibrate trust — don’t maximize it

The danger is not only that guests under-trust a capable robot. It is that they over-trust an incapable one, leaning on it for things it cannot do, only to feel betrayed when it fails. Hastie is alert to the sharper edge of this, too: as people grow more reliant on conversational AI, over-trusting becomes a risk of manipulation — particularly for vulnerable or isolated guests. The job is to make the machine’s confidence track its real capability, so trust lands where it belongs and nowhere else.

2. Failures are inevitable — so apologize, explain, and fail gracefully

Robots break. What matters is the recovery. Work by Hastie’s student Birthe Nesset found a clear hierarchy: an apology beats silence, but an apology with an explanation beats a bare one — “sorry, I got that wrong; I thought I saw a QR code, but I was mistaken” rebuilds more trust than “sorry” alone. The wider literature adds that a genuine compensation (“let me make that right”) outperforms words, that repeated failures are rarely fully repaired, and that denial is the worst move of all. The modern advantage, 

Hastie notes, is that today’s models are good at failing gracefully — responding sensibly rather than with a blunt “computer says no.”

3. Transparency is trust — the robot must explain its reasoning

Hastie’s clearest evidence comes from a robot triage nurse, trialed during COVID, that sat at a hospital reception, asked about symptoms, and advised people whether to go home or sit. The finding was unambiguous: when the robot explained the reasoning behind its decision, participants were markedly more accepting and trusting. A reception robot that explains itself — read that twice, because it is your front desk. A black box that says yes or no leaves the guest with nothing to calibrate against, and trust swings between blind faith and flat rejection.

4. Form follows trust — appearance must fit the task

How a robot looks is not decoration; it sets expectations. Hastie warns of the uncanny valley — the more closely a machine imitates a human, the more unsettling it becomes — and insists the right look is entirely dependent on the job. A machine doing serious work should look like it knows what it’s doing: in a warehouse, wheels beat legs. The signals must also be congruent — a warm line in a dead monotone reads as creepy rather than caring. As Hastie puts it, you would not want a cute, hapless film-robot handing out drugs in a hospital. Charm and authority are different design briefs; a lobby needs to choose deliberately.

5. Novelty is not destiny — a good robot can beat the curve

This is where the primary source corrects the conventional wisdom. The novelty effect is real — people get excited by a new robot, then drift away. But Hastie’s six-week barista trial, measured using a recognized negative-attitudes-to-robots scale, did not show it. Guest attitudes held steady. Why? Genuine utility (very good coffee), plus her telling observation that trustworthiness, likability, and usability are closely linked. The lesson for hoteliers is the optimistic one: novelty fades by default, but a robot that is genuinely useful and likable can hold its welcome well past the opening-week selfie.

The hospitality mirror: a cautionary tale at full scale

We have already run this experiment. In 2015, Japan’s Henn-na Hotel — aptly “Strange Hotel” — opened with a Guinness World Records title as the first robot-staffed hotel: dinosaur receptionists, robotic porters, and an in-room assistant called Churi in every room. Guests loved the novelty and fed their social feeds. Then reality arrived.

By January 2019, the hotel had cut more than half of its 243 robots. Churi mistook a guest’s snoring for a command and woke him repeatedly through the night. The front-desk robot could not answer basic questions. The dinosaur could not photocopy a passport — a legal check-in requirement — so humans stepped in every time. The porters could not handle slopes. Staff worked overtime repairing the machines that were meant to replace them. The robots, as one analysis put it, were better at creating work than reducing it.

Henn-na is every one of Hastie’s five findings failing at once: over-trust in robot capability; repeated failures with no credible repair; no transparency when things went wrong; novelty that curdled into irritation; and a snore read as a command — the textbook failure to understand human context.

Set that against Hastie’s barista. Both were robots placed in a real-world service setting; one was abandoned, the other was embraced. The difference was not the hardware — it was whether the machine was genuinely useful, transparent, and likable. Henn-na’s novelty curdled because the robots could not deliver; the barista’s welcome held because it could. Novelty is not destiny. Competence is.

And yet the machines are coming anyway, for hard commercial reasons. The American Hotel & Lodging Association reported that 87% of hotels are facing staffing shortages in 2025, with labor accounting for nearly a third of revenue. The hospitality-robotics market is forecast to climb from roughly USD 0.6 billion in 2025 toward USD 1.8–3 billion within five years. Closer to home, service robots are projected to grow from USD 80.6 million to around USD 208 million in the Philippines alone by 2029. Hilton has trialed “Connie”; Aloft has “Botlr”; delivery robots from Relay, Bear, and Pudu now roam corridors worldwide. The question is no longer whether but how well.

Trust by Design: a framework for hoteliers

Trust is not a feature you bolt on at go-live. It is designed in, or it is absent. Here is how Hastie’s science translates into decisions a General Manager or owner can actually make.

The economics of trust: where TCPG meets trust-per-guest

Every robot and every AI agent carries a running cost, which is exactly what Token Cost Per Guest (TCPG) was built to surface: total AI spend divided by guests served, a unit metric a controller can defend. But TCPG only tells you the price. It says nothing about the return, and the return on a guest-facing machine is denominated in trust.

A robot that erodes trust is expensive at any token price, because trust is the asset hospitality actually sells. A robot that earns calibrated trust — reliable, transparent, gracious in failure — converts spenders into loyalty, reviews, and repeat stays. So the metric to pair with cost-per-guest is trust-per-guest: are your machines adding to the relationship, or quietly eroding it? The first is an investment. The second is the Churi problem, with a monthly invoice attached.

So — in robots we trust?

Yes — conditionally. Hastie’s hope is for robots that are not merely in our world but a welcome part of it. That word, “welcome,” is the whole game in hospitality. A robot earns its welcome the way a good employee does: by being reliable, by being honest about what it can’t do, by making it right when it falls short, and by reading the room. None of that is automatic. All of it is designed.

Hastie is equally clear about the guardrails. Build robots so they can explain themselves and leave an audit trail, embed responsible design from inception rather than bolting it on, and take the darker risks seriously — robots that can be hacked, or turned to surveillance. For a hotel, that is not abstract: a guest-facing robot sees rooms, faces, and routines. Trust is also a security posture.

The hotels that win the next decade will not be the ones with the most robots. They will be the ones whose robots are trusted — because someone took the science of trust as seriously as the spec sheet. Hastie’s own dream is modest and human: she wants tech her ninety-something mother will actually accept so that she can live well and independently — rather than leave it in a drawer. That is the real test. Buy the capability if you must. But earn the trust. The guests can tell the difference, even when they can’t name it.

The intelligence may be artificial. But the experience is human.

Sources & further reading - ctto

Primary source: Helen Hastie, interviewed by Jim Al-Khalili, The Life Scientific, BBC, 8 June 2026 — full episode transcript. All quotations and the surgical-robot, ship-autopilot, triage-nurse, Robo-Barista (“Alex”), apology-plus-explanation, and novelty-effect findings are drawn directly from this episode.

• Robo-Barista (“Alex”) and the Furhat robot receptionist — six-week in-the-wild trust trial at the National Robotarium / Heriot-Watt University.

• Birthe Nesset, Marta Romeo, G. Rajendran & Helen Hastie — “Robot Broken Promise? Repair strategies for mitigating loss of trust for repeated failures,” IEEE RO-MAN 2023 (apology vs. apology-plus-explanation).

• Nesset, Robb, Lopes & Hastie — “Transparency in HRI: trust and decision making in the face of robot errors,” ACM/IEEE HRI 2021 (triage-nurse transparency).

• Hastie, H. — Trust in Autonomous Systems node; “Why teaching robots and humans to trust each other is essential,” The Scotsman / Royal Society of Edinburgh, 2022.

• SPRING (Socially Pertinent Robots in Gerontological Healthcare), Horizon 2020, National Robotarium; Alana conversational-AI system (Heriot-Watt, Amazon Alexa Prize).

• Wider trust-repair literature: ScienceDirect “Three strikes and you are out” (2023); compensation vs. apology studies (ACM THRI).

• Henn-na Hotel robot reduction (secondary): Hotel Management, The Wall Street Journal, Raconteur (2019–2022).

• Hospitality service-robot acceptance and market data (secondary): ScienceDirect meta-analyses (2024–2026); AHLA staffing survey (2025).

Note on sourcing: this version is built on the verbatim episode transcript; Hastie’s own words and her lab’s findings are the primary basis, with the named papers as their published underpinning and press items clearly marked as secondary commentary.

Made with the help of various AI tools, but with a HITL.