Latest posts by JP Emelie Marcos (see all)
- Four Predictive Analytics Challenges Facing Site Reliability Engineering Teams - August 24, 2017
- Monitoring Microservices at SignifAI with AI and Machine Learning - June 27, 2017
- Monitorama PDX 2017 Recap - May 31, 2017
As you know, we are super excited by everything that is going on in the World of A.I. and Machine Intelligence. Self-driving cars may very well be the first autonomous machines we will rely on at a massive scale, they are still a few years off but A.I. has already begun to transform industries and fields – including technical operations – in ways we may not even realize. We constantly strive to keep up with all the news and content that is out there and it is definitely hard to do given the sheer volume of information made available on a daily basis.
We have curated the following video presentations by four A.I. experts who focus on different aspects of the technology. The talks address both the current state of A.I. and where it’s headed. You will hear an explanation about the rapid rise of A.I. technology as well as what’s standing in the way of it becoming universally transformative.
A.I.: The New Electricity
Coursera co-founder and Stanford adjunct professor Andrew Ng is one of the leading thinkers in the A.I. industry. In Artificial Intelligence is the New Electricity, Ng explains that A.I. is having a significant impact on the way we live, work and play. He points to how 100 years ago electricity was a key driver in the industrial revolution – from enabling refrigeration, to transforming transportation. Today, A.I. holds that same promise. A.I. can determine if a loan should be approved, when your take-out order will arrive at your door, and how likely a consumer is to click on an app.
But while A.I. is useful for determining input to response mapping, it’s still relatively limited when compared to human intelligence. Still, Ng says progress is much faster when A.I. tries to automate things humans can do, especially with less than one second of thought. And if a human can’t do something – like predict the stock market – it’s highly unlikely A.I. will be able to.
Deep Learning Deep Dive
In The Nuts and Bolts of Deep Learning, Ng further explores the notion of Deep Learning relative to A.I.. While the fundamentals of Deep Learning have been around for decades, the concept has just recently begun to take off, due to the amount of data available today and the ability to synthesize that data.
Today, end-to-end Deep Learning can generate much more than numbers – such as speech recognition and facial recognition – but it relies on a ton of data. For example, a self-driving car would need data on the location of pedestrians, other cars, and barriers, the speed of traffic, the curve and trajectory of the road, etc. in order to compute the steering direction. Ng recommends that A.I. teams and machine learning teams work together to tune the algorithms for their needs.
The Common Sense Problem
Dileep George, co-founder of Vicarious, one of the leading A.I. firms in the U.S., takes another look at A.I.. In Artificial Intelligence at Work, he reviews how we’ve solved hardware problems but still don’t have the software to power a Rosie from the Jetsons-type robot to cook and clean. Alexa and Siri may be headed in the right direction, but we still have a long way to go. The challenge, he says, is that A.I. still hasn’t solved the common sense problem.
Today, A.I. can break CAPTCHAS, but to create systems with common sense, we need to build models that simulate the real world. For instance, when an iRobot vacuum gets stuck in a corner, it has no way of realizing it’s gone over the same patch of carpet a dozen times. What’s missing is the common sense reasoning humans provide. That’s the next big hurdle for A.I., according to George.
Taking the Leap from Deep Learning to Unsupervised Learning
Computer scientist Yoshua Bengio echoed his peers’ sentiments with a talk he gave at Stanford University: “Foundations and Challenges of Deep Learning”. Bengio believes five key ingredients will help us evolve from machine learning to true artificial intelligence: massive amounts of data, bigger and more flexible models, more computing power, computationally efficient inference and powerful priors that defeat the curse of dimensionality. In other words, if we don’t assume things about the world, it’s actually impossible to learn about it.
He believes the goal in A.I. is to move beyond pattern recognition to more complicated tasks including reasoning and planning and reinforcement learning. The biggest challenge ahead of us now, he said, is unsupervised learning. For instance, humans can learn from large quantities of unlabeled data and make assumptions based on what we already know, but machines aren’t there yet. When machines make mistakes they’re using the wrong cues to provide the answers. In order to make machines smarter, we need to ensure those models reflect how the world actually works.
In Shyam Sankar’s TED Talk, The Rise of Human Computer Cooperation, the data-mining innovator talks about the need to go beyond algorithms to solve the world’s biggest problems. Instead, he says, we need to identify the right symbiotic relationship between computation and human creativity. While humans are incredibly creative and good at non-linear thinking and iterative hypotheses, we’re not good at scale, computation, and volume, where technology thrives.
Human Computer Symbiosis (AKA augmented reality) lets humans and machines cooperate in making complex decisions, Sankar says. Computers don’t detect novel patterns and behaviors, humans do. So by designing humans into the process, we can impose intuition and reasoning on data, which is critical for catching terrorists or identifying hidden trends.
While Ng, George, Bengio and Sankar all have slightly differing opinions and areas of focus, they all agree that there’s tremendous value in A.I., and that it’s changing the future.
Are there any videos that get you excited about A.I.? Please share in the comments below.