Who do you trust most: a robot, an alarm clock or your partner?

When Mark Zuckerberg first addressed the Cambridge Analytica situation, he described what had happened as “a breach of trust”. In early April, a documentary featuring Elon Musk came out called “Do You Trust This Computer?” And as experts on the ethical use of algorithms like Cathy O’Neil and Virginia Eubanks have discussed, people often wrongly assume they can trust computers and algorithms to be unbiased and fair. In fact, we are now entering an era where we cannot necessarily trust what we ourselves see and hear due to recent advances in audio/video forgery.

We’ve spoken with a number of consumer enterprise organizations that worry about a conflict between innovating in AI and maintaining trust with their customers. It’s a false opposition. Trust is what makes relationships possible, whether they are personal, commercial, or international.

We’ll look at three different datasets, a total of 16,152 uses of trust and its related forms

To build technologies that justify trust, it’d be good to know what trust is, so I’m going to look at how people actually use the word trust in conversation.

I’ll kick off this post with a definition of trust, but focus on an analysis of trust in everyday and not-so-everyday situations: about 12,000 conversations among friends, family members and strangers. About 10% of all these conversations make some mention of trust. Then I’ll turn to more extreme situations exemplified by characters in 135 different TV shows, episodes are longer than conversations but on average 53% of TV episodes make at least one mention of trust.

Trust: an initial definition

When it comes to defining trust, my favorite resource has been Carolyn McLeod’s overview. What I like most about her approach is that it focuses on the inherent riskiness of trusting, which leads philosophers like her to think through the conditions when trust is warranted. This seems crucial for people building AI and other technologies to grapple with: what is the nature of the relationships you’re building with people affected by your systems?

For McLeod, the archetype of trust is interpersonal and that feels right to me, too. It’s also what we’ll see in the data below. By considering interpersonal, one-on-one trust, we can see how “trust” requires the possibility of betrayal. Once we are alive to the vulnerability inherent in trusting, we can see how different it is from reliance. If your alarm clock breaks overnight, you can be disappointed and frustrated, but you cannot be betrayed.

You can rely on an alarm clock but can you trust it?

Contrast an alarm clock with a friend who you’ve asked to do something important. In both cases, you implicitly assess competence before you either rely OR trust them. Trust involves something more. For example, if you are trusting your friend, you have also sensed that they are committed. There’s a bit more controversy on a third condition: you are optimistic about your friend’s goodwill. That is, if you thought your friend would help you because they have some selfish ulterior motive, you may correctly predict that they will do the thing you’re asking, but that’s not really trust.

Trustworthiness doesn’t come just because you can be counted on to do (or not do) something. An example from Nancy Nyquist Potter makes this obvious: imagine you’re a woman with an awful sexist boss who understands legal/HR sanctions he’d face if he acted out. So he never acts out but that doesn’t make him trustworthy. The parallel in corporations would be asking why you’re not polluting a lake with mercury. If the motivation for not polluting is because you don’t want to face steep fines or public relations disasters, then you’re not a trustworthy organization. In the data realm, do you protect your customers because of regulations and fears of fine or because you actually care about the privacy and well-being of your customers?

How friends and families talk about trust

First things first: unlike in the next two sections, the data I’m reporting in this section is fairly small. Across 221 conversations between people who know each other well, about 10.2% of them make mention of some form of trust. Here I’m including all parts of speech and words based on trust like entrust and distrusted, but I’m ditching word forms that have to do with financial arrangements (trustee, trustafarians, antitrust). All of these friends and family members know they are being recorded, so how much people talk about trust in private may be higher.

With so few occurrences, I could read all the transcripts. The overwhelming conclusion is that people bring up trust when there’s an issue with it. That’s different than other words. For example, you can certainly problematize love but few instances of I love you occur because someone is doubting love and it’s uncommon to negate love (“I don’t love him”) than it is to negate trust. My own favorite negative from these conversations: “I would never trust myself to shoe a horse.”

The archetypical — or at least the most common — use of trust is interpersonal. Of the 46 occurrences, 18 of them are about people that the speaker knows (39%, these are commonly about fidelity) and only 7 are about things or systems (15%, e.g., sonograms, recording equipment, the Dutch version of Quicken). The rest of the tokens are about people that the speaker doesn’t have a direct relationship with — like presidential candidates or unspecified somebody’s.

It’s probably worth pausing on how most of the impersonal uses like trust somebody/anybody/anyone get used. In this small data set, most of them come from a speaker who has taken one betrayal of trust and, sadly, extrapolated from one person to everyone:

A: And it was just like I couldn’t trust anybody because you know if you decide to trust somebody so much {lipsmack} you just comple- completely never think that they could do that to you
B: Right
A: Right. And then they do.

The somebody here is not truly generic, it’s a generalization of a very specific and hurtful incident. It’s pretty obvious that if you (as an individual or a company) betray someone’s trust, they would be justified in not trusting you, at least until you rebuild the trust. But notice the possibility that someone else in the ecosystem can contaminate the trust that people place in others.

Later in the A&B conversation, Speaker A acknowledges it’s important to trust people. No potential lovers are going to convince her that they are trustworthy by browbeating her with logic or data. Partly, it’s going to take time. And partly, her future romantic partners need to take actions that are consistent with trustworthiness. As Nancy Nyquist Potter suggests, we must pair the risk-taking of the person who trusts (or would like to) with the responsibility-taking of the person or organization that would be trusted. And if there is care of any sort in your relationship with a friend, family member or customer, then you would care about the trustworthiness of the distant others that they interact with.

More data, this time strangers

Let’s jump over to a lot more data, strangers paired up over the phone to chat with each other. The Fisher Corpus has about 16,000 different conversations — pairs of people were randomly assigned one of ~40 topics. Overall, 9% of these conversations have some use of *trust*, but as you can imagine certain topics generate a lot more trust-talk than others. The data was collected circa 2003 — a few of the topics were about specific themes in the news, but most were broader.

Of the 40 topics talked about in the Fisher Corpus, here are the topics that mention *trust* most/least/typically

Among conversations about life partners, the tendency is for strangers to talk about trust by including it as core to what you want in a life partner. In utterances about trust, people also mention communication, integrity, friendship, compassion, sincerity, loyalty and hard work but the most commonly co-occurring word is honesty.

(1) that’s why i said a lot of trust that you gotta be honest in the relationship in order for it to last
(2) you can’t trust a dishonest person right

I believe that trust+honesty connects to motivation: someone who is honest tells you what’s going on, even if it’s difficult. You can’t trust someone who makes everything smooth and easy. If you are erecting the scaffolding of trust, you’re putting in effort and people know that sometimes hard hats are required. Luxurious fluffy, silky pillows feel wonderful but are terrible girders.

When we shift our attention to less interpersonal examples, there’s a lot more sighing involved. Some of the highest rates of sigh’s are when people talk about trust in conversations about arms inspections and there are even more sighs when people are talking about trust and corporate conduct. In these topics, there’s a kind of exhausting remove from the ability to know or affect what’s going on.

Another keyword in the corporate conduct topic is money. When you accept a dollar, you are trusting that it will be worth something. You’re vulnerable to economic forces like inflation that could render it practically valueless as folks in Zimbabwe and many other parts of the world know well. As we saw in the A&B conversation about fidelity, this topic shows a lot of extrapolation and contamination: people often say something along the lines of “you know you can’t you feel like you can’t trust anybody anymore” due to corporate scandals.

This also connects to motivations. When you believe that a corporation’s primary motivation is money, it becomes difficult to trust them. You may be able to rely on them to behave certain ways, but that doesn’t seem to be trust.

(3) we’re supposed to trust these people a- and you know you they don’t care whether you trust ’em or not they got their money
(4) to squeeze money out of customers and what happens is the customers very quickly lose trust in these companies and get irritated

You could go two directions with this analysis. The one I prefer is to try to alter motivations and increase responsibility-taking. Another approach would be to assert that trust is only relevant for one-on-one relationships and that everyone should be disabused of the notion that anyone is looking out for them. Part of me admires the realism of that, but it sounds a bit too much like an Ayn Randian horrorscape of unfettered self-interest.

Take me to TV

In our real-life conversation data, no one says I don’t trust you in exactly those words. But that’s one of the top five most frequent phrases in TV shows. And it may surprise you to learn which show uses it the most: Star Trek: The Next Generation.

It’s hard to trust The Other

If you’ve watched that show, you know that the crew gets along with each other very well. The characters that don’t trust each other are outsiders, in fact, it’s usually aliens not trusting the crew (or other aliens).

Sometimes this is justified mistrust, as when you’re dealing with an omnipotent mischief-maker. Other times it’s just lack of familiarity. Mostly, though, it has to do with what kinds of characters you’re surrounded by, as we’re about to see.

It probably won’t surprise you to know that Friends, I Love Lucy, South Park, The Simpsons, Big Bang Theory, Bob’s Burgers, M*A*S*H* and The Golden Girls rarely talk about trust — though it does still come up for them more than strangers on the phone. The show with the lowest frequency of trust in the 135 shows I looked at was for Sex and the City and some form of that word still occurs in 15% of the episodes.

Of the 135 shows, here are the ones that mention *trust* most/least/typically across episodes

But let’s jump to the shows that have the most talk about trust.

Quantico: In which one member of an FBI training class may be a terrorist

Lucifer: In which the protagonist is actually Satan and really wishes you would trust him (in his defense, he doesn’t lie)

White Collar: In which a con-man joins an FBI team to help catch other forgers and ne’er-do-wells

How to Get Away with Murder: A class covers up all kinds of things, like murder, and they all seem to have a lot of secrets

Other shows with lots of mentions of different forms of trust include The Americans, The X-Files, Person of Interest, Battlestar Galactica and Hannibal. That is, if there’s a conspiracy of aliens or robots, if your neighbor might be a spy or your psychiatrist might be a psychopath, you’re going to have a lot of discussions about trust. For those of us involved in AI, note the relevance of not only Battlestar Galactica’s Cylons but the artificial intelligence surveillance system (and its associated humans) in Person of Interest.

The TV shows that use trust and its related forms the most are those where there are real problems with justifying trust. To return to the aspects highlighted by Carolyn McLeod, these are situations where there’s real risk and vulnerability. The highest rates of trust don’t really have to do with problems of competence or commitment — those are characteristics of trust breakdowns in comedies. Rather, characters in these shows are focused on second-guessing motivations.

Quick conclusion

In real life, technologists are deploying systems that affect justice, health, housing, education and a host of other not-funny-when-it-goes-wrong areas. The stakes are higher and so our attention and thoughtfulness needs to be higher, too.

It’s easy to build AI systems that do not build trust: that will always happen when you take massive amounts of data about people and then try to optimize for some goal without considering how the data, the goal, and the whole system will affect people. In the next few weeks, I’ll be writing about trust in politics and media, but in the meantime please comment and question! You can also check out this deck on an ethical framework and practical steps and overview of bias by my colleague, Kathryn Hume.

Tyler Schnoebelen (@TSchnoebelen) is principal product manager at integrate.ai. Prior to joining integrate, Tyler ran product management at Machine Zone and before that, founded an NLP company, Idibon. He holds a PhD in linguistics from Stanford and a BA in English from Yale. Tyler’s insights on language have been featured in places like the New York Times, the Boston Globe, Time, The Atlantic, NPR, and CNN. He’s also a tiny character in a movie about emoji and a novel about fairies.