Thales, recognizing that there is a perception problem with artificial intelligence (AI), has launched at the Paris Air Show the concept of TrUE AI—for Transparent, Understandable and Ethical.
According to Thales Group chief technology officer Marko Erman, Thales started developing AI solutions around 30 years ago and has embraced it “from many angles.” But the company felt increasingly that a “framework” is required to help put AI in context and prevent fear—the typical imagery being of Skynet in the "Terminator" films, and movies involving androids. “AI functions very differently to the human brain, actually,” he noted.
Building trust with customers is a key aspect of TrUE AI and Erman said how a company processes and secures data is key. In April, Erman said, Thales acquired cybersecurity specialist Gemalto to help take care of security.
For applying AI to data, he said it was important to ensure “biased” data was not being used, and to make how a result/decision was reached transparent would help the user/customer understand the validity of the result. “You have to really understand what the machine does compared to what you asked for,” said Erman.
With understanding, he gives the example of a future pilot trusting an AI assistant blindly—or where “the AI says turn left and the pilot would rather go straight on. It presents a dilemma” and the danger is the pilot not understanding what is behind the AI’s “advice.”
Thales is already using AI on preventative maintenance—“the first application.” Where the AI predicts a failure, Thales wants to be able to explain the reasoning.
It also acquired a company called Psibernetix that has developed algorithms for use in a flight simulator. They are genetic algorithms using decision trees and “fuzzy” logic—replicating pilot decisions using pilot knowledge. “The AI is able to do it much faster than a human,” said Erman. “And Psibernetix already slowed it down by 200 times but still AI won—and the algorithm was run on a very cheap PC.”
Thales is working to offer its aerospace customers “a toolbox of ‘algos’ that are understandable…some of the algos may be certificated sooner than others.”
“Then there is ethics,” said Erman. “We don’t believe AI is a threat to humanity…it is still a machine and is staying a machine. It’s unable to develop its own plan—but could be misused or malfunction.”
In addition, he said, “Our AI needs to respect all the regulations and United Nations rules on what’s good and what’s not—discrimination, etcetera.” Also, Thales will ensure the user can break into the loop if things do go awry.
Finally, sustainability needs to be a core consideration, said Erman, given that the rate of energy-consumption growth driven by the internet means by 2040 it will consume more power than everything else in the world. To that end, Thales is developing its own low-power chip technology and “neural chips using quantum physics that operate at room temperature—based on ‘spintronics.’”
These are in nanocubes that are much smaller than conventional transistors and have radio frequency connections to each other. “This is in the lab at the moment,” said Erman. “We can improve power consumption by a least two orders of magnitude.”