Jamal Bentahar | Photo by David Ward
Netflix recommendations are one concrete example. Intelligent recommender systems are trained to analyze user histories, understand complex user preferences and respond to new information (like interests or search queries) to produce perfectly curated results.
But with this new intelligence comes a new challenge.
“Intelligent systems exhibit autonomous and non-deterministic behaviours,” Bentahar says. “Verifying this type of system is particularly challenging from the computational perspective. New algorithms have to consider the autonomy and selfishness of agents.”
Your system has to talk to other computers and software systems to log changes, make requests, transfer files or money, and so forth. When smart devices do so, they may not follow standard protocols. They can instead behave in their user’s interests.
A trust-building exercise
To secure intelligent systems, Bentahar creates a model that shows how they behave.
He uses this to solve for security issues like trusting other systems and detecting malicious components.
A smart device must figure out how to trust. For instance, it’s first taught to identify signs that another system is trying to mislead it. It learns to assign initial trust values to new systems that it encounters. And finally, it creates a network of friendly systems that it deems trustworthy.
Bentahar says that despite the effectiveness of these security measures, regular backups remain crucial. “Intelligent systems benefit from learning how to do their own automatic and smart backups.”
From Amazon Marketplace to Uber
With an increasing need for intelligent systems and advanced AI techniques, Bentahar plans to take on the economics of intelligent systems and their influence on verification, security and access to data.
“Amazon Marketplace, Google Ad Exchange, Uber, eBay and recommender systems such as Netflix are examples of intelligent systems I am going to analyze,” he says.
“There are many challenging issues in these systems yet to be addressed, such as data monetization, security of trading protocols and the influence of user ratings on consumers.”
Learn more about the Concordia Institute for Information Systems Engineering.
