Brighterion’s unique approach to artificial intelligence (AI) and machine learning technology is easily traced to the unique background of founder and CEO Dr. Akli Adjaoute.
Dr. Adjaoute’s family emigrated to France when he was a child. His work ethic was instilled at a young age as he both went to school and worked to help support his family. That dedication paid off as he was admitted to University Pierre & Marie Curie, where in 1988 he earned a PhD in AI. He also served with the French military during his schooling.
AI was still a young sector three decades ago, with many companies making outlandish claims and failing to deliver on them. That was not the case with Cognitix Technologies, the company Dr. Adjaoute founded before he graduated.
“We moved from poor to successful in a year,” Dr. Adjaoute said.
Cognitix developed a powerful tool for the development and deployment of AI applications and quickly amassed a roster of clients in the financial, industrial and military sectors. Dr. Adjaoute also maintained an academic presence, where he has served as the director of cognitive science and artificial intelligence at Ecole Pour l’Informatique et les Techniques Avancees since 1989 (he has also served as an adjunct professor in Artificial Intelligence and Business Intelligence at the University of San Francisco and as scientific director for the Institut Europeen de Valorization de la Recherché).
“I call teaching a hobby,” Dr. Adjaoute said. “Regularly being with 60 top brains in math who were designing AI projects was remarkable for me. I was always challenged by my students and learned more than they did.”
In 2000 Dr. Adjaoute founded Brighterion, where he led the creation of an award-winning AI platform that now secures 75 billion annual transactions for leading organizations and governments in the areas of payments, compliance, financial markets, security and defense, healthcare, Internet of Things and marketing. Among Brighterion’s more than 2,000 global customers are 74 of America’s 100 largest banks. Their technology can successfully handle 62,000 transactions per second.
At the heart of Brighterion’s offerings are Smart Agents, which, instead of trying to anticipate every possible situation a system might face, or looking to pre-trained models, conducts unsupervised learning by tracking and adaptively updating the specific behavior of every entity in real time.
Smart Agents address a few common problems common to standard AI systems, Dr. Adjaoute explained. At their heart AI systems are systems with complex rule sets guiding their actions. That can work for basic situations where you follow a few steps in sequence to produce a predictable outcome but it quickly falls apart as situations become more complex and the potential outcomes more numerous.
Take banking, Dr. Adjaoute said. You may have one account with multiple cardholders, each with their own spending patterns and behaviors.
“You can’t just write in all of the rules to account for all of the possible behaviors,” Dr. Adjaoute said.
Another problem with many AI systems is they only incorporate past behavior, he added. They struggle to anticipate future growth and development.
“They’re not adaptive, there’s no personalization,” Dr. Adjaoute explained. “Neural networks, the only good thing is the name, the technology’s poor. Multi layers? Deep learning? I am against this term. There is no deep learning, just pattern recognition.”
A “deep learning” system may have to look at one million pictures of a motorcycle to consistently know what is while a two year-old can learn it much faster, Dr. Adjaoute said as an example.
Brighterion approaches AI in a unique way, it seems, so I asked Dr. Adjaoute how this different approach evolved. It begins with his experience in the military, he began. Problems are mission critical and mandate optimal solutions. Smart Agents fit right into that philosophy as they produce adaptive solutions to previously unseen situations just like those which troops must anticipate.
Learning in France’s education and business environments also helped, he added. Students are taught to focus on truly unique inventions and not reinvent the wheel. A standard French company deploys a much larger percentage of resources on engineering and research/development whilst a typical American one prioritizes sales and marketing.
Take the multi-person bank account from above. With Brighterion technology each person on the account is assigned a Smart Agent which learns from every transaction, Dr. Adjaoute said. Traditional AI companies rely on data bases, but Brighterion takes a different approach.
“Legacy technologies in machine learning generally rely on databases,” he explained. “A database uses tables to store structured data. First, a database presents a bottleneck when one needs to scale to thousands of transactions per second with low response time. Second, tables cannot store knowledge or behaviors. Artificial intelligence and machine learning systems require storing knowledge and behaviors.
“Smart Agents bring a powerful, distributed file system specifically designed to store knowledge and behaviors. This distributed architecture allows lightning speed response times (below one millisecond) on entry level servers as well as end-to-end encryption and traceability.”
Brighterion also employs distributed architecture, which allows Smart Agents to be deployed in multiple environments and collectively work together, Dr. Adjaoute said. I mentioned that approach lends itself well to the Internet of Things, and he agreed. Each machine can have its own Smart Agent that incorporate information from other machines in the network. No one machine pays a master role.
“The distributed architecture results in better performance using only two or three entry level Unix machines rather than using costly hardware or a mainframe,” Dr. Adjaoute said. “It also makes Brighterion’s solution scalable and resilient to disruption because it has no single point of failure. Intelligent routing (business rules driven) distributes data to the right server. With Brighterion’s distributed architecture, it is very easy to upgrade the hardware configuration by just plugging in a new server.
“For instance, if the number of transactions processed each month is expected to be increased by one billion transaction, one new server could to be added to handle the load. You have to give control to every element you think is extremely important and secure. Our technology fits like a glove for IoT.”
Data quality is another concept Dr Adjaoute finds frustrating. As long as it is clean you can work with it, provided you have the right technology.
And don’t eliminate data outliers, Dr. Adjaoute cautioned.
“Important information can come as ‘noise’ so if systems remove it, it is gone. Many people don’t care about outliers but they should because that is where the future trends are.
“With correlations, find something the opposite of what you might expect.”
