Artificial Stupidity — Patenting Intelligence

A summer break thought experiment

Google recently submitted multiple patent applications for machine learning methods. These patents are very wide ranging. Cheekily one could say, Google is trying to patent intelligence. The question arrises, what is the narrative here?

An overview of patents Google is applying for:

Dropout, Classification, Parallelising neural networks, Word embeddings, Data augmentation methods for images, Q Learning with deep networks.

The Experiment

From all the patent applications above, one is particularly striking: Classification. This is a very widely used technique in machine learning. With out classification, many intelligent systems would be ineffective. This experiment assumes that access to classification is hindered and only sub-optimal techniques are available.

1. The starting point was the Convolutional neural net to detect facial keypoints by Daniel Nouri. This deep neural network enables us to detect faces in pictures. It´s easy to setup and yields great results.
2. Next, test images had to be selected: Pictures of Google Executives.
3. The neural net was tweaked and extended until it correctly recognised all faces in the images. The goal: Classify faces by position, age and race.
4. For the sake of this experiment, I assumed access to standard classification techniques is hindered. Due to summer-break time-constrains, I could not explore alternative approaches and had to use manual classification. Predictably this produced sub-optimal results, that are not worthy of publishing here.
5. Finally, the detected faces were programmatically replaced with an alternative: Donkey heads. This stands in a grand tradition of caricature commentary using animals heads, going back to the 18th century.

Experiment Results

Comments on Statements

Statement: We should blame the US patent laws for allowing patents on well-known abstract ideas; Google is just exercising its legal rights.

Comment: Simply because something is legally sound, does not make it right. It is widely acknowledged, that the patent system is broken. Patent-trolls are systemically exploiting this flawed system.

Statement: Google has previously used patents in a defensive manor, they will do the same with intelligence patent, nothing to worry about. If they don´t do it, some other company will.

Comment: Google is not a victim defending itself, but a primary driver of such behaviour. As a anonymous poster on Redit says:

Google has just started an arms race, which could do significantly damage to academic research in machine learning. Now it’s likely that other companies using machine learning will rush to patent every research idea that was developed in part by their employees. We have all been in a prisoner’s dilemma situation, and Google just defected. Now researchers will guard their ideas much more combatively, given that it’s now fair game to patent these ideas, and big money is at stake. These new patents cast doubt on any new company based on machine learning and make it harder to attract investments — now there is the looming threat that Google has rights to their IP and can sue or at least use the patents as a bargaining chip in case of a legal dispute.

Statement: Google: “We’re just patenting it so no one else does. This protects everyone.”

Comment: We need to call on Google to release these patents into the public domain. I don’t mean release open licenses like Creative Commons, any license can be rescinded. Non-commercial doesn’t cut it, because these methods are already in commercial use. They must be fully open and free.

Statement: Unless you are a patent attorney, you are not qualified to decipher the word soup that makes up a typical patent. (reddit comment)

Comment: By this logic, only nuclear physicist are qualified to comment on atomic bombs. As there are very little efforts from with in the legal/corporate establishment to explain / contextualise all of this, its up to mere mortals to try their best. Every public discussion/learning process starts in mediocracy. Its up to the experts to help educate the public.

Final thoughts

I have great respect for the researchers and engineers at Google. The decisions made by their leaders though, are questionable at times. Attempting to patent basic building blocks of intelligent systems sets a dangerous precedent. It is a direct negation of the open/free culture Google is promoting. Its up to the machine learning community and public at large to help decision makers make the right choices.

Finally, here is a donkey head picture of me.

Get in touch here: https://twitter.com/samim | http://samim.io

Sign up for the Newsletter for more experiments like this!