Connected devices and proof

Connected devices and proof

For most of the internet’s history, software has known surprisingly little about what we actually do in the physical world.

We tell an app that we went for a run. We tick a box saying we completed a task. We submit a form saying we attended something. We upload a photo as evidence. In many cases, the system simply has to trust us.

That is beginning to change (or is mid way through).

The devices around us are generating an extraordinary amount of information about our behaviour. Watches know how far we have travelled. Phones know when and where certain actions happened. Cars, payment platforms, connected equipment and health devices all produce their own streams of data. At the same time, AI is getting much better at interpreting messy information that previously required a human to review it. Put those two developments together and an interesting question appears:

What can we build when software can increasingly verify that something actually happened?

That question is what led me to start building a new project. I am keeping the details deliberately vague for now, but the underlying idea is broader than the product itself. It is about turning data into proof.

We already generate the evidence

There is a slightly strange contradiction in modern technology. We generate more data about ourselves than at any point in history, yet much of the software we use still operates as if none of that data exists.

Consider something as simple as completing a goal. Traditionally, if I told someone that I would do something three times this week, they would have two ways of knowing whether I did it - they could watch me do it, or they could take my word for it. Neither scales particularly well. But increasingly, the evidence already exists somewhere. A wearable may have recorded the activity, a phone may have captured the movement, a transaction may have been recorded, a connected device may have logged its use, an application may have generated a timestamped event. The interesting problem is no longer necessarily how to collect the data - it is how to connect the right pieces of data together, interpret them correctly and decide whether they constitute sufficient evidence that something happened.

That distinction matters. A device producing data is useful, but a system being able to turn that data into a reliable yes-or-no answer is considerably more powerful.

From tracking to transactions

Most consumer technology currently uses personal data for tracking.

• Steps.

• Sleep.

• Heart rate.

• Spending.

• Screen time.

• Location.

• Training load.

These products are generally designed to help us understand ourselves. But what if some of those data points could also trigger transactions? Imagine agreeing with someone that if you complete a particular objective, they will reward you. Or setting aside your own money and committing it to an outcome. Or a company rewarding customers for completing something that can be independently verified.

The payment does not have to depend on somebody manually checking whether you kept your word. The proof could come from the technology you were already carrying. That creates a subtle shift. Instead of:

“I say I did it.”

you move towards:

“The available evidence shows that I did it.”

Once that distinction exists, money can sit on the other side of it. A reward can be released, a stake can be returned or transferred, a benefit can be unlocked, an agreement between two people can settle automatically. In other words, data stops being purely descriptive and starts becoming transactional.

A market for verified behaviour

There is potentially a much broader category here than fitness, productivity or any single type of goal. The underlying structure is simple:

An action happens → evidence is generated → software verifies the evidence → something happens as a result.

The final step could be financial, but it does not have to be.

It might unlock access.

Award points.

Trigger insurance benefits.

Release a reward.

Confirm attendance.

Verify completion.

Change the price of something.

The applications become much more interesting as the number of connected devices grows. Wearables are an obvious example because they sit directly on the person, but they are only one source of evidence. Phones are becoming collections of increasingly sophisticated sensors. Vehicles generate telemetry. Payment networks provide transaction data. Etc Etc. The physical and digital worlds are slowly becoming more observable.

Enter AI

Traditional software works best when the rules are neat.

If distance is > than five kilometres, return true.

If a transaction occurred < a particular time, return true.

Real life is rarely that clean. Evidence can be incomplete, contradictory or spread across multiple sources. This is where AI becomes particularly interesting.

Large language models and multimodel AI systems are increasingly capable of interpreting information rather than simply processing predefined fields. That potentially allows verification systems to work with much broader forms of evidence: structured API data might provide one signal, a photograph could provide another, a receipt, message or document could provide another, and location or sensor data could add confidence to all of that. AI could help interpret those signals together and determine whether the evidence supports the claimed event.

We may be moving from software that simply stores our inputs towards software that can make judgements about real-world events.

Trust is probably the real product

The more I think about the project I am building, the more I suspect the underlying opportunity is not really about data. It is about trust.

Money already moves easily across the internet - creating an agreement in software is easy. The difficult part is often determining whether the condition behind that agreement has actually been satisfied.

Did the person turn up?

Did they complete the task?

Did the event actually happen?

Did the required activity take place?

Historically, solving those questions often required another human being.

Connected technology provides evidence, AI can help interpret it, software can then act on the result. That starts to make small agreements economically viable in situations where manual verification would be ridiculous. 

There is also a dystopian version of this future. A world in which every behaviour is measured, scored and financially incentivised would be fairly unpleasant. The goal should not be to turn life into a giant points system.

The experiment

Can the data generated by the technology around us become a useful source of proof? And if it can, what new types of agreements, rewards and incentives become possible?

I do not know yet how far that idea extends.

Perhaps the interesting product is narrow. Perhaps the underlying concept becomes much broader. Perhaps it turns out that people simply do not want software involved in these kinds of agreements at all. That is the useful thing about building it - instead of debating the idea indefinitely, I can create something small, use real technology, put it in front of people and see what happens.

For now, the project stays in stealth.