Google is committed not to put its artificial intelligence technologies at the service of weapons

The American company has developed seven major principles that it promises to apply in its development of artificial intelligence technologies.

Seven main principles, lying on paper: for Google, it is a first. On Thursday, June 7, the US company unveiled the rules that will now frame how it will continue to develop its artificial intelligence (AI) technologies. She also inscribed in the same document, to which Le Monde had access, what she pledged herself not to do , and it is undoubtedly in this part that the most notable elements reside.
Thus, Google undertakes not to contribute to developing "weapons or other technologies whose main purpose is to cause , or facilitate directly, injury . " The issue, particularly sensitive, has put Google in trouble in recent weeks, after the revelation in March of a partnership with the Pentagon .

The Maven project, in which Google was involved, was running artificial intelligence programs to analyze drone images. This partnership has generated heated debates internally, as well as the resignation of twelve employees and a petition signed by more than 4,000 employees demanding a "clear policy that neither Google nor its partners will ever manufacture war technologies . " Google had defended itself by assuring that these technologies were intended for "non-offensive use". Before finally give Friday 1 st June, announcing that it would end in March 2019, with this contract.

"We will continue to work with the army"
Has this affair precipitated the publication of these great principles? On the side of Google, in any case, it claims to study the issue for months, even years. And it is said that under these new safeguards, the agreement with the Pentagon could no longer take place today.

However, the company writes in black and white:

"While we were developing AI not to be used in weapons, we will continue to work with governments and the military in many other areas. This includes cybersecurity, training, recruitment, veteran health and relief. "
Google also agrees not to develop AI technologies "that collect or use data for surveillance purposes violating international human rights principles , " or "whose objective is against the broad principles of law. international " .

This is vague enough that Google retains a wide margin of maneuver, but its very existence could potentially complicate his life. These rules give its detractors, or dissatisfied employees, more concrete arguments on which to rely in the event of disagreement.

Seven main principles
Google has therefore defined what it committed not to do. But he also established "seven great principles to guide [his] work," says the company. Artificial intelligence should thus, according to Google:

"Benefit society. " The company says: " We will consider a wide range of economic and social factors, and we will conduct tests to determine if the potential benefits substantially exceed the risks and predictable bad sides. "

"Avoid creating and reinforcing unfair bias. " Currently, many AI systems based on human data, tend to reproduce racist, sexist or other of their models .

"Be designed and tested for safety" , to "avoid unexpected results that may pose risks" .

"To be accountable. " One of the great problems of AI programs, including those based on the deep learning methods , is that their designers are not able to explain their decisions. This question is the subject of research, especially at Google.

"Incorporate principles of respect for privacy. " A number of AI technologies require to operate , large masses of data, which raises questions about respect for personal data.

"To maintain the high standards of scientific excellence. " In research and development methods, but also in publishing the results of its scientific work, says Google.

"To be made available to others for uses consistent with these principles. " Google says well it " reserves the right to prevent or put to use a term of [its] technology " if it finds " incompatible uses with these principles " .

Companies so far chilly
How to ensure that these principles will be respected? Google talks about internal auditing processes that already apply to cybersecurity or privacy issues, for example. But for the moment, no ethics committee in sight.

So far, the Web giants like Google had been relatively cautious about the idea of engraving in marble principles of this type, which could harm innovation in a key sector evolves very quickly. These actors have often returned the ball to the civil society and the governments, more according to them, than private companies to pronounce on big ethical and societal stakes.

Faced with the considerable progress of artificial intelligence in recent years, and the concerns they may have raised , some initiatives have emerged, such as the Partnership on AI , which has since 2016 GAFAM (Google, Apple, Facebook, Amazon , Microsoft) and about 50 companies, NGOs and research centers around these issues.

Google is one of the most advanced companies in the field of artificial intelligence. It owns DeepMind, a London-based company it bought in 2014, which managed two years later to create a program that could beat the champions in the game of go - a performance that was not expected. before one or two decades.

The publication of these rules can improve the image of Google, damaged by its partnership with the Pentagon, and position itself as a leader on ethical issues, taking the lead in the face of competition. It may also be intended to reassure those who had been moved that Google's early motto, "Do not be evil," is significantly diminished in the new Code of Conduct. of the company released in the spring. Finally, these seven principles may also be intended to alleviate internal discontent, whereas in Silicon Valley, early idealistic hopes have been blackened, scandal after scandal, by doubts