Google is offering resources to the US Department of Defense. UU For a "pilot project" to analyze drones sequences using artificial intelligence. The collaboration was reported for the first time in a Gizmodo story, which noted that some Google employees had been "outraged" by the news after it was shared on an internal mailing list last week.
The details regarding what Google is really providing to the Department of Defense are unclear, but we know that the work is part of the Maven Project, a Pentagon research initiative to develop a computer vision that can better understand the video. Such technology could have a variety of uses in the military, including helping to monitor and track targets using drones and provide the brain for static CCTV cameras in camps and military bases. Technology of this type can be used to recognize basic objects, such as cars and people, but has problems analyzing complex scenes.
Work on Project Maven began last April, and in July, the head of the team overseeing the initiative, Colonel Drew Cukor, said the Pentagon was looking for "business partners" to help develop its technology. Google, it seems, is one of those partners.
"The technology shows images for human review, and is only for non-offensive uses."
In a press release, a Google spokesperson confirmed that the company was providing the Department of Defense with access to its TensorFlow software to help build object recognition algorithms. "For a long time we have worked with government agencies to provide technological solutions," the spokesperson said. "This specific project is a pilot with the Department of Defense, to provide open source TensorFlow APIs that can aid in the recognition of objects in unclassified data, the technology shows images for human review, and is only for non-offensive uses."
TensorFlow is a piece of free software that has been an axis in Google's AI strategy for years. It offers machine learning engineers a framework for sorting data and training their algorithms and is widely used throughout the industry for a variety of purposes.
If the Pentagon is using TensorFlow, it's not a surprise, but a familiar source with the matter confirmed to The Verge that Google is also actively helping to configure the technology company's software for use in Project Maven. The same source pointed out that the confidentiality agreements between Google and the Department of Defense means that the details of the number of personnel involved and the depth of this help can not be revealed. Is Google simply guiding DoD researchers through the TensorFlow manual, or are they paying the company's engineers to develop algorithms using drone video? It's not clear.
However, the use of artificial intelligence for these types of tasks raises a number of ethical issues (in addition to those already associated with the use of unmanned aircraft from the United States for surveillance and selective killing abroad) . Although the army of the EE. UU He has insisted that he will only deploy AI as a human intelligence partner, computer-assisted decision making could still encourage mistakes. Human drone operators could become too dependent on faulty computer analysis, and without proper supervision, these errors could go unnoticed.
For Google and its employees, there are additional dilemmas. Are the company's engineers happy that the software they are making can be used by the US military? UU Someday? What happens with the shareholders of Google? In the past, the company has been extremely cautious of any link with military research. When he bought a number of robotics companies in 2013, for example, he did everything possible to distance himself from the company's current DARPA contracts.
In this case, the company says it has yet to make decisions about how and when it is appropriate for its artificial intelligence technology to be used for military research. A spokesperson told The Verge: "We are actively discussing this important issue internally and with others as we continue to develop policies and safeguards around the development and use of our machine learning technologies."