Artificial intelligence systems and their development are becoming increasingly important. “Smart” devices, self-driving cars, and robots are all gaining more attention. We are now at the point where people – blue-collar and white-collar alike – are worried about losing their jobs to robots.
Like it or not, however, AI is the future – for the industry, for enterprise, and for society.
Of course, all of this will mean new headaches for the CIO, the IT administrator and the network engineer alike as the ongoing evolution of AI places increasing demands on the networking industry.
Forgetting, for a moment, how one successfully programs and achieves artificial intelligence, there are very real data processing and network concerns that have to be addressed. It’s one thing to program artificial intelligence – but how do you build it? How do you house it? How do you process it?
What does the AI-enabling network of tomorrow look like?
AI networks will use the cloud
The cloud – known for its ability to offer agility and added storage – is already enabling AI to some extent. IBM, for instance, maintains Watson in the cloud partly for accessibility issues and partly because of the sheer amounts of data that Watson must effectively contend with.
For artificially intelligent robots, however, the cloud is a necessity. Onboarding everything a mobile robot will ever need is impractical for reasons of power requirements, operational duration, and cost.
The cloud has other benefits for robots, too. “Cloud robotics allow robots to take advantage of the rapid increase in data transfer rates to offload tasks without hard real-time requirements,” reports robotics project team RoboEarth.
Indeed, artificially intelligent robots are already using the cloud. A 2013 UC-Berkeley project outlined how robots could better grasp objects through cloud enablement. Companies such as Gostai offer customers cloud access to a wealth of complex robotic actions, “including … advanced vision and speech algorithms.” And last year’s DARPA Robotics Challenge – which tested robots in simulated disaster scenarios – used a cloud infrastructure over a VPN connection for added resiliency.
AI networks will store lots of data
Rob High, VP and CTO of IBM Watson, told students and technologists during his keynote address at MIT’s annual Tech Conference earlier this year that we are presently experiencing an “information explosion.” Identifying the Internet of Things as a significant contributor, High noted that, at present, the world produces approximately four exabytes of new data each day.
“We really do need computers to help us,” High argued.
Specifically, said High, we need cognitive computing – an advanced form of machine learning that is vital to AI.
“Cognitive systems have to learn … because that is what it takes to deal with all the variations we deal with as human beings[,]” said High. “You can’t sit down and write all the rules of language and ever feel like you’ve completed the task. Our language is far too diverse and varied for that.”
But even cognitive AI itself presents is own data-management problems, requiring advanced interactive solutions.
“Deep learning typically requires a significantly larger data set for training,” reported High, who said the technology is “changing the paradigm of training to one of ‘give us all your training data’ … to ‘now let’s put it in this system of interaction’ … so it can kind of learn on the job, if you will.”
AI networks will be fast
“To read all of the [medical research] data being published on a weekly basis would require about 160 hours of reading a week,” said High, who went on to point out that the average doctor spends only about five hours per week reading medical research. (For those of you who don’t feel like doing the math, a week contains 168 hours.)
As a counterpoint, High related Watson’s early days as a Jeopardy contestant in 2011.
“We had about 200 … pages of literature that Watson had to read at the [time] the question was being asked,” said High. “It had about three seconds.”
Watson went on to beat two of Jeopardy’s biggest champions ever.
The challenges – and the necessary speed – of AI systems like Watson are yet more compounded today because our information is becoming exceedingly difficult to process. Unstructured datasets (e.g., video recordings) are inherently problematic to search and analyze. Structured data, too, can be difficult to break down and process mathematically because they contain “human forms of expression.”
These forms of human expression – including tonality and body language – inform human intelligence in real time and allow us to react.
“I will intuitively and subconsciously react to [your cues] and try to adjust what I’m saying,” said High of human intelligence and communication. “We all do this.”
Accordingly, faster connections and powerful processing will be imperative for IT because AI systems must be able to interpret our data quickly – and act on it.
Joe Stanganelli, Principal of Beacon Hill Law, is a Boston-based attorney, business consultant, writer, speaker, and bridge player. Follow him on Twitter at @JoeStanganelli.