AT&T uses artificial intelligence to tell if things are going wrong in its network. Soon, AI may know it before it happens.
The carrier says it’s been using AI for decades in areas like call-center automation but developed it for each use as they came along. Now AT&T is pouring its AI smarts into a one platform that can be used with multiple applications.
“I can’t just keep doing this once at a time. We need a foundation,” said Mazin Gilbert, assistant vice president of the company’s Inventive Sciences division, in an interview last week at the AT&T Shape conference in San Francisco.
That foundation is about two million lines of the code that powers AT&T’s Domain 2.0 software-defined network, which the carrier built so it could roll out new services more quickly and efficiently. Along with its own AI code, much of which is open source, the company is using open-source components from partners including universities and third-party vendors.
AT&T is trying to make the same AI leap as many other companies with a lot of data. Why just react to events if you have enough data points to make predictions and improve them over time?
AI began more than 20 years ago at AT&T, but it didn’t go too far at the time. The machine-learning systems would capture data, analyze it, figure out what was amiss, and generate a trouble ticket to a someone who could act on the problem.
Now the software can identify the problem, like a suspicious activity on a virtual machine, and then take action based on company policies. For example, the software might direct traffic away from that VM and later restart it.
The next step will be to make the software smart enough to predict whether a system is about to fail. By analyzing lots of data about the network over time, AI would forecast possible failures and set a certainty level, like 80 percent, then take action.
The real magic will happen when the AI has done this many times and keeps getting better at making predictions. It could even modify policies over time, Gilbert said.
One area where AI is playing a growing role at AT&T is in security tools, such as the Threat Intellect software platform the company announced last week. AT&T predicted the system would speed up its deployment of security protections by more than 95 percent. Threat Intellect can process a full day’s worth of activity for its security customers -- 5 billion security events – in 10 minutes, the company said.
IoT is another area where the carrier can bring its AI power to bear, Gilbert said. For example, if a wearable sensor detected a change in a patient’s blood chemistry, AI might be able to determine what was going on and take action, he said.
AI can handle about 80 percent of network management and AT&T is working on raising that by a few percentage points, Gilbert said. But while it’s great at quickly analyzing data, finding patterns and refining its conclusions over time, only people can deal with all the varied considerations that go into higher level decision-making, he said. “There’s no way we’re going to be able to automate 100 percent of our network.”