Why Artificial Intelligence Might Owe You Something

5 min read

The ideas summarized below derive from an op-ed by Dr. Miguel Luengo-Oroz which first appeared (in Spanish) on 7 January 2019 in El Pais Retina. The topics discussed represent the views of our chief data scientist, which we wanted to open for further discussion. The article was edited by Felicia Vacarelu, Communication Specialist with Global Pulse. Photocredit: @JavierCentelles. 

The borders between the digital and the physical world have been crumbling for some time. We live in an inter-connected world where our cells don’t just link us to other people, but to the news, weather, or to billions of smart objects. In many cases, access to data has become as essential as food. Our lives are influenced — for the better or for the worse, or just for optimization purposes like finding you a cab —  by decisions made by algorithms. 

Many of these algorithms are fed by data that we produce on a daily basis just by going about our lives. Big data and the exponential increase of computing capacity have led to the explosion of artificial intelligence (AI). We are now at an inflection point around its use, and in order for us as a society to decide what it is we really want to do with it, we first need to break it down. 

The AI  Factory of Masks

In the last few years, a new generation of machine learning algorithms, called “deep learning”  has revolutionized the AI field. It broke the leaderboards when it learned how to distinguish cats from dogs and how to translate languages. It mastered Go, the Chinese board game from scratch, and with no human help beyond being told the rules. Deep learning, however, is far from being magic, neither is it intelligent on its own. What it is, is more like a factory of masks. 

Imagine a mix of 1000 women and men lined up in a factory waiting to be fitted for plaster masks. An artisan mask maker applies wet mold over the face of the first person in line, who’s a young woman, and the mold captures her characteristics. Second in line is a young man and the mask maker takes another fresh mold and applies it to him. For the third person in line, another woman, the mask maker uses the first mold, which still wet, now carries the characteristics of both faces of women it was applied to. And for all 1000 people, the mask maker continues to apply the two wet molds: one for women, and the other for men. Once done, he lets the plaster masks dry out. 

The next day, a new AI system made of those masks will allow the factory’s doors to open automatically only if the first “mask fits,” that is only the women will be able to get in without using a key.  

This is how deep neural networks work. The artisan mask maker uses the tools (algorithms) to fabricate (train) the mask (model) from the data. AI for facial recognition is nothing more than wet mold that captures common features found in pixels of photos of a particular class: man / woman; black/ white; cat/dog.

How to make a mask

The Ingredients

First, the 1000 people on which the wet mold was applied, aka, for an AI system, the data used to train it. The more data the algorithm can train on, the better it can create categories.

Second, the mask maker whose tools are the algorithms. Today, many algorithms are open source and available for free, but the experts (mask makers) who know how to use them are in short supply, mostly because training in data science is still not widely available. 

Third, the masks themselves, which are the models created by the algorithms. Once created, they can be widely applied – to automatically open the doors of other factories, but also to unlock a smartphone, to automate the process of buying groceries in a supermarket from a cashier-free counter or for mass surveillance.

Why Artificial Intelligence Owes You Something

These masks – the AI models that are being created – once produced, can be easily used and maintained at very low cost. The business model tech companies expect to apply, is to get paid every time a mask is used. 

Doctors in the public healthcare system spend for example hours inserting information into forms — which creates the same thing the molds applied to the 1000 employees do – training data. So the question arises, who does the mask belong to? The company that provides the IT services? The healthcare system? The doctor? Or the patient? 

Let’s now bring 1 million bartenders to get fitted for a mask at our mask factory. This time, we are applying the wet mold to create a mask that captures the movements of a bartender serving beer; this is a dynamic mask, something resembling a robot. The result of applying the wet mold to the million bartenders is a mask (aka the model) which can serve beer. In fact, the bartenders are no longer needed – their knowledge has been codified in a mask and they are now superfluous and too costly. 

Whoever is able to compile enough data to create a mask will be able to replicate and use it almost for free. This is in the core of the current arms race both for governments and private sector – whichever arrives there first, will be able to exploit these masks with nearly infinite margins – and in a platform economy where the winner might take all.

In this equation, where does the doctor, the bartender, or the employee fall? Should a doctor be compensated everytime an AI diagnoses a new case based on their findings? Or should professional bartenders be paid every time a mask trained with their data is used for serving a cocktail?

My idea goes beyond a one-time payment system for the use of our data. I propose that those whose actions are used to train models (create masks) should be compensated each time a mask is used. What would that system look like? Is it something similar to royalties or authors’ rights? Is it a tax on the use of AI models?

Miguel Luengo-Oroz

In any case, the solution will depend on the context. I strongly believe that the real long term challenge for AI, isn’t the development of trustworthy AI – which is a necessary, but not sufficient condition. It is finding the right way to redistribute the wealth that AI will bring in a way that ensures no one is left behind. 

The race to produce masks which might render current human activities and jobs irrelevant has already started, and it has done so with nothing in it for the people being used to create them. Can we find another way?

Did you enjoy this blog post? Share it with your networks!

News, thoughts and ideas about big data and AI, data privacy and ethics from across the Pulse Lab Network. Read more on the blog.

Pulse Lab Kampala

Dialogue is the Key- Shaping AI for Africa

At UNGP, we believe in dialogue. This is why we participated in the Conference on the State of Artificial Intelligence in Africa (COSAA) held in March 2023. The conference was

Scroll to Top