Hi there, we are sensor.link!

We believe that machine learning & AI have the power to change the world for the better, especially when used in a responsible and accountable way.

Our service enables blockchain-backed data authenticity for companies that deploy inferential systems in public spaces.

Applying machine learning inappropriately can potentially lead to simplistic, inaccurate, unaccountable, un-auditable decision systems getting deployed in serious situations and negatively affecting people’s lives. [...] It’s not a hypothetical risk, it’s a pressing concern.
— Francois Chollet (Google AI researcher and creator of the Keras deep learning library)


sensor.link is born of a idea to create an unbiased, unfalsifiable and independent security registry for sensor data and associated services that can be interrogated and cited as reference in case of disputes between legal entities that use software and machine learning based systems to make critical decisions.

Machine learning, in the broadest sense, works by taking a vast training data set as input and producing an algorithm that allows us to make inferences about data points that bear some resemblance to the training data. However when such an algorithm is presented with data that it cannot correctly classify such systems necessarily fail because it is impossible to infer the correct result from the data that is presented to the system: this is the limitation of inductive reasoning.

What is induction exactly? Induction is going from plenty of particulars to the general. It is very handy, since the general takes much less room in one’s memory than a collection of particulars. The effect of such compression is the reduction of detected randomness.
— Nassim Nichoals Taleb - The Black Swan

Automated decision-making systems grow ever more complex with layers of hardware and software interacting perfectly to produce results that facilitate many common tasks in our everyday lives.

However when such systems fail and it is necessary to pin-point why things can get very complicated because these systems may rely on many different subsystems to produce data points that are combined by the final decisional algorithm.

Imagine for instance a basic model of an autonomous vehicle:


In this case using a combination of RGB cameras, LIDAR and depth sensors we reach a total of 7 different vendors producing systems that either collect or operate on sensor data to participate in reaching a decision at any given point in time.

A problem arises when such a system fails and it is necessary to establish responsibility.

To reverse engineer the scenario which led to failure, we need to take the data produced by the sensors in the moments prior to the event and run them through the same inferential systems as used at the time of failure.

This is where the problem lies: we are presented with a system where at least 7 different participants may have a vested interest in misrepresenting sensor data and / or ML solutions in a post-hoc analysis to demonstrate that they are not responsible for the failure.




sensor.link leverages the immutability of blockchain databases to establish a common standard that can be implemented by all developers of decisional algorithms and sensor manufacturers to guarantee and integrity of the sensor data and algorithms employed in their systems.



Supported by:


As seen on:  


Name *