The Age Of Autonomy Is Coming
Originally published in ReadWrite, Jan 19, 2016
All of the recent buzz about autonomous vehicles and humanoid robots—a conversation driven by big players ranging from Google to Toyota—obscures one of the most historically important events that’s unfolding right in front of us. The Age of Autonomy is arriving, quietly and widely, as truly autonomous devices move onto the scene to supplant those that are merely automatic.
And we are not just talking about vehicles. Fireworks in the sky won’t mark this evolution. Instead, buyers of devices ranging from small kitchen appliances to hugely expensive industrial systems will simply notice that things seem to work a little differently, a little better, than they did before.
Beyond The Internet Of Things
However quietly it begins to arrive, The Age of Autonomy marks a giant step beyond the world of connected devices and the Internet of Things. With 50 billion devices and a trillion sensors coming on line, simple practicality would suggest autonomy will become a necessity. In fact, autonomy is almost here already. Every device and even some of the things we don’t generally regard as devices are essentially becoming robots of one type or another. And with that, autonomy is starting to appear.
As we take our first steps in the Age of Autonomy, it’s important to understand the difference between autonomous devices and automatic devices. Autonomous devices make reasoned decisions based on their knowledge of the situation, the goals that have been established for them, probabilistic beliefs, and possibly even domain expertise.
An automatic device or system, by contrast, is made up of some number of simple reflex-like rules. If it’s 8 a.m., the automatic sprinkler system comes on—provided the grass isn’t already wet from an overnight rain.
A truly autonomous sprinkler system operates more like a knowledgeable landscape gardener. Current conditions play a role—it’s going to be warm today, it didn’t rain yesterday, the evapotranspiration rate indicates the grass needs 0.3 inches of water. An autonomous system might tap into a calendar to know that a wedding is planned on the lawn in five days, and would adjust the water application appropriately to ensure that the grass is green but dry when the bride begins her walk.
Devices that are autonomous have a high-level appreciation of context both near and far, and they are sensitive to the context as they evaluate possible actions. A good way to think about an autonomous device is that it has the capacity to choose its actions without the involvement of another device or operator.
Consider big data. Even if large amounts of data are generated, the system that uses big data isn’t autonomous if someone needs to think about the data and push the proverbial button to put the data to use. Devices that simply deliver actionable insights or actionable intelligence and leave the human to make the decisions and execute the appropriate actions are intelligent assistants, not autonomous systems.
A Bias Towards Action
An autonomous system is distinguished by its capability to choose and perform an action at any time. When it chooses to perform an action, it operates with an awareness of the context of the current situation, including its own state, the state of other devices, or both. Or to put it more simply, an autonomous device is rationally adaptive. It can assess its performance and improve, all without being revised by a human actor. It learns.
Simple systems that are based purely on rules are assessed equally simply: Does the system trigger a logically fixed action when a condition is met?
For example: Does the set of rules provided by the human actor for a sprinkler system specify that the water will run for the 20 minutes required by the lawn rather than a wasteful 20 hours? And when the condition to turn on the sprinklers is met—it’s now 8 a.m., it didn’t rain yesterday—do the sprinklers run as they are programmed?
For autonomous systems, the questions are these: Did the system act in a way that was good, given some set of conditions and goals? Was the action that the system selected as good or better than previous actions taken under similar conditions?
To use our example of the sprinkler system again, we would ask if the autonomous system made a good decision about the amount of water to apply, based on the established standard of a healthy lawn that meets the needs of the people who use it. Then we would ask if the system made better choices than it did a week ago, as it used data from sensors in the soil as well as suggestions from human landscape maintenance specialists to fine-tune its decisions.
Like an automatic system, the autonomous system still accepts sensor and human inputs. But instead of the reflex-like responses to inputs that are the only way an automatic system can react, an autonomous system balances goals with reactive or responsive aspects when making its action choices.
The level of sophistication, of course, will vary from one device to another. Some devices will be fully autonomous as they operate without direct human control or oversight. Others will be will be semiautonomous as they remain subordinate to some level of human authority.
The Machine Cooperative
It’s a small step from the development of autonomous devices to the creation of autonomous systems in which autonomous devices coordinate their activity, and possibly even cooperate with one another. Whether they cooperate or merely coordinate their activity, autonomous devices in a system will need to be aware of what’s happening with other devices and the relationship between the state of the other devices and the possible actions of the other devices. The structure that links autonomous devices together into systems will become one of the important pieces of the technological infrastructure.
The Age of Autonomy will become an age in which all of us, humankind and its devices alike, learn together to make decisions that are good today—and even better tomorrow.
Photo by Laurence Simon