Amazon hosted an AI Summit as part of their re:Invent event. Geared towards a more general group of people interested in Artificial Intelligence, the Summit strove to explore forward thinking topics rather than focus on what’s currently being done.
The following two talks especially grabbed my attention:
Compound Commands and Follow-on Conversations with Alexa
To kick off the event, Amazon announced several improvements to Alexa that will make “her” smarter. When you think that they are projecting a market of 75 billion connected devices by 2025, you understand the strategic importance of improvements in this area.
To start off, Alexa will support compound commands. This means that adding bananas, strawberries and apples to your shopping list will finally result in 3 distinct items being logged in.
Then you will be able to have follow-on conversations as Alexa expands her abilities to understand context. So you will be able to ask about something that was related to your previous inquiry without having to repeat the first question again.
Mixed Autonomous Traffic Made Possible by Deep Learning
This was probably the most fascinating talk of the day.
We have been carrying out traffic studies for a long time, with traffic jam models emerging as early as 1935. In 2008, researchers had people drive around a parking lot. They were told to maintain a specific distance between the cars. But humans aren’t good at regulation and couldn’t do it due to distraction or just the sheer inability to maintain constant distances. The researchers then inserted one autonomous car into the mix and reported that that car alone managed to improve the whole traffic pattern as the human-driven cars adjusted to follow the autonomous vehicle. They calculated that a single self driving car would result in 42% energy savings. Today, it is estimated that a mere 10% of autonomous vehicles will have a huge impact on relieving traffic congestion.
The challenge, as always, is to train the model as it’s a very manual process. Fortunately, AI is now able to supersede the human ability to model behavior using Deep Learning, a machine learning technique that teaches computers to learn by example. Amazon has launched a human assisted deep learning tool called SageMaker Ground Truth. However this isn’t good enough.
A more recent AI technique is Reinforcement Learning with Amazon SageMaker RL. Reinforcement learning (RL) is a machine learning technique that aims to train systems to make decisions on their own through a continuous process of receiving rewards and punishments for every action taken. This means we can create interactions that run with no model at all.
At Six Feet Up, some of our clients use regression models that are based on equations. With RL, there will be no need for equations to regulate the interactions. Instead, AI will bridge the gap and fill in the blanks for better forecasting.
Back to autonomous vehicles, the next wave of innovation is striving to use only the surroundings to control traffic. This means that cars will learn through pictures rather than with the state space (speed, position, etc). This is all fascinating.
Other Amazon Announcements You Don't Want to Miss
