Black Secrets and Grey Secrets

They say that everyone has dark secrets.

But how dark?

For the purpose of this article I am going to define two levels of darkness: black and grey.

A black secret is a secret that you would never tell to anyone, unless it is someone actually directly involved in the secret, ie a co-conspirator (such as the person you are having an affair with, assuming that the other person even knows that it is an affair …).

A grey secret is where you don’t freely hand the information out, and you might be keeping it back even from everyone that you know well, but, there are other parties that you might sometimes reveal it to.

A black secret is normally black because the downside of having the secret get out is just too great.

For a grey secret, there is some downside if the secret gets out, but at the same time there are situations where you judge it worth the risk because there may be some benefit from revealing it to another party.

What we don’t know about secrets, in general.

The thing about secrets is that we know all about our own secrets, and we might know something about the secrets that other people close to us have, but, we don’t really know much about the general level of secrecy in society in general. How many people have more and darker secrets than us? And what sort of secrets are they, usually?

There was a time when people seriously believed that no one else masturbated. Everyone thought it was their own special secret.

Nowadays such a belief seems rather silly. Even though masturbation is still not a subject for polite conversation.

That’s one example of something that wasn’t known scientifically about an aspect of human behaviour, even though a majority of adults did it.

Grey Secrets in the Online World

The development of the internet has enabled new ways for people to choose to make their grey secrets available to third parties.

Two common ways this can happen are:

Modern AI is a new development, and there are two different ways that AI enables both the release and the use (or abuse) of grey secrets: using grey secrets as training data, and users’ maximisation of context.

Grey Secrets as AI Training Data

Online services can use existing grey secrets that they already have access to as training data for AI models.

They might do this because users have (perhaps unknowingly) opted in to allow this usage, or the online service might just do it anyway.

If an online service directly uses your personal data that they are meant to be keeping secret for you, then there is a reasonable probability that they will be found out, and if any online service repeatedly or systematically abused their users’ data in this way they would almost certainly get found out.

But if such data is used to train AI, then it will be less obvious that such abuse has happened, even if the trained model is made available as a service that anyone can use (paid or otherwise).

Context Maximisation

The second thing about AI and grey secrets is that AI itself is a thing that people will want to reveal their secrets to.

We all know that AI is sometimes really good at quickly providing good answers to questions that we may have about many different things. AI is not able to solve all our problems, and sometimes it makes up wrong answers, so we still have to check if we want to be careful (sometimes “checking” is just “ask a different AI”).

But, often enough AI is good enough that for any problem we want to solve that we are struggling with, it’s usually worth asking AI for help.

And this applies even for those problems that involve our dark secrets.

For a black secret, you won’t want to tell that to an AI, at least not to an online commercial AI service (but you might consider going local, if you are rich enough to own a computer powerful enough to run a good-enough local model).

But for a grey secret, then quite likely you will at least consider telling AI about it.

For most people, confessing stuff to an AI is no more or less risky than any other online interaction involving their grey secrets. You always know that the company involved might abuse their access to your personal information, but as far as you know they don’t, and unless you represent some special kind of target, there is no particular reason to believe that they would specifically abuse your data unless they were abusing everyone’s data, in which case you would probably (eventually) hear about it in the news.

The extra thing about revealing grey secrets to AI is not just that you might reveal secrets relating to the problem you are trying to solve. Anyone who uses AI eventually learns the importance of providing adequate context to AI about whatever problem you want it to solve. A priori, the AI only knows about things that it read (so-to-speak) on the public internet and other sources of training data not specifically related to you personally. If you want AI to do a good job of solving your specific problem, you have to tell it all the relevant information relating to your specific situation.

In other words, you are not only motivated to tell AI about your grey secrets – you are motivated to tell AI as much as possible about those secrets in the hope that AI will do the best possible job thinking of possible solutions to your problems.

As a result, modern AI services are gaining access to private information en-masse of a type that has never quite been made available in such a manner.

And they have the means to process all that information to feed into new AI models, which, as a result, will “know” things about the general human condition which currently no one else knows about.

The Future…

So that’s the future that may be coming: commercial AI models, and the owners of those AI models, will know more about what it is like to be human than anyone has ever known before. And they may or may not make that information available for the rest of us to use.