The most crucial real-world problem today, one that concerns every large and small company, is cybersecurity. In the future, security will be a battle between artificial intelligences.
In Tel Aviv I met Dr. Eli David. His work focuses on deep learning, also known as deep neural nets. During the past few years, deep learning has achieved 20 to 30 percent improvement in most benchmarks: computer vision, speech recognition, and text comprehension. This is the greatest leap in performance in the history of AI and computer science. The development is being driven by the growing performance of graphic processing units (GPUs) and the improvement of algorithms.
What is unique about deep learning is that it learns by itself – a little bit like a child. But to do this, neural nets need data. A much larger amount of data. But the more data you have, the better the results.
The higher the number of neurons, the higher the level of intelligence
The Brazilian neuroscientist Suzana Herculano-Houzel invented a method to count the number of neurons in the brain. “Her research suggests that the level of intelligence correlates with the number of neurons in the cerebral cortex. The higher the number, the higher the intelligence.” So when we talk about neurons it is mostly about quantity.
In the human cerebral cortex there are 16 billion of these neurons. Each neuron is on average connected to 10,000 other neurons. So, the total number of connections (synapses) in our cerebral cortex is about 160 trillion. The largest neural nets at the moment have about a few billion connections. As the number of neurons in neural networks increases, it is likely that these networks will reach the human level of intelligence. As soon as artificial intelligence reaches the human intelligence level, it will quickly be able to surpass it. You can find more information here.
A chimpanzee looking at us humans
To illustrate what the rise of a superintelligence could mean, Eli gave me an example. The difference between a monkey and a human seems to be just the number of neurons. In the future an AGI might look at us like we are looking at a chimpanzee. Will we then look at an AGI the way chimpanzees look at us? Will AGIs understand us as something as insiginificant as a chimpanzee? In the end we don’t know – yet, says Eli.
Cyberwarfare is the real warfare of the future
Eli is working on deep learning system in the field of cybersecurity. He said that one of the most crucial real-world problems today, one that concerns every large and small company, is cybersecurity. More than a million new malware threats are created every single day and sophisticated attacks are continously crippling entire companies – or even nations – by targeting ciritical infrastructure. In a digitalized world the potential damages increase exponentially.
At Deep Instinct, they have developed a neural network for exactly this issue. The big potential of deep learning for security solutions is that all data is viewed. Given the complexity of today’s malware, this is a great advantage. So far there is no malware based on deep learning, but there will be very soon. This means that in the future security will be a battle between artificial intelligences.
Eli and his team trained their system in the laboratory with their own malware. Then the system started learning and getting better and better. Eli’s vision is a deep-learning based security for everything – every endpoint, every traffic stream, every device.