Knowing how to improve: AI, machine learning, neural networks

Many of the ways that we're already interacting with smarter technology, often without knowing or even perceiving it, are made possible by some form of artificial intelligence, machine learning, or neural networks.

?As a result, user behavior and expectations continue to be dramatically reshaped in step with the available technology.

?That each of these terms are often used interchangeably does not help in our understanding of how they add something special to a user experience. As I say elsewhere, one definition of a great product is that it makes users’ lives easier in some way. AI, machine learning and neural networks can all do so - in various ways.

?The power of all three is their capacity to not only accelerate the development process, but more importantly, expand the realm of what’s possible, changing how we use our devices.

?The ability for devices first to be trained, then to learn, and then to improve, at the programming level, without any human intervention. This alone is awe-inspiring, but then the resulting ability for the technology to recognize an event or a command or a desired end-state and perform an action accordingly without having to explicitly program that action, continues to reshape the entire technology landscape.

?AI, machine learning and neural networks, in all their iterations, are sufficiently advanced and integrated into products. In the next couple years, there will be a fix of refinements and major new capabilities generated. So, I think it makes a great deal of sense to understand these three different disciplines, how they often work together, and some of their applications within devices that are already in the marketplace.

Not automation

Let's start with AI. Artificial intelligence is the ability of a device to do tasks that would be typically done by humans. What AI is not, is automation. An example of automation would be to walk into a room and have a sensor turn on a light. The experience becomes artificial intelligence when it has the ability to understand when to turn the light on and when to turn it off.

The light bulb moment is going to get more intelligent in the future…

AI is the ability to discern and take advantage of a scenario as a system takes over a task – to understand that if I walk into my kitchen at midnight, the light should be turned on softly, or that my laptop understands from my movements that I'm engaged in front of a screen, not just present. The software built into the device has the ability to do more than simply automate a task.

The difference is that discernment comes from thinking about what is intended, and the intent.

Self-learning

Machine learning takes things one step further: now, the algorithm itself becomes better. So, this is the part where the intelligence starts to predict and determine what's going on and then takes an action that creates a beneficial result.

In the case of machine learning, the machine monitors results, and then changes its actions over time, becoming better at predicting outcomes, but is not explicitly programmed to do so.

The machine learns and modifies. It has the ability, outcomes, or data, and can determine when to store information to make those decisions in the future.

A good example is the Nintendo Wii remote, which learns the movement patterns of individuals.

By doing so, it predicts the outcome of that movement, again without input from the designer or user. The machine itself has hysteresis and feedback, so that it becomes better.

Self-improvement

Now, on to neural networks which are needed for high-powered capabilities. The difference here is in the programming. Away from neural networks, programming is essentially linear – if this, do that. It assumes a step-by-step kind of a logical outcome.

A neural network has parallel processing and assumes a number of different outcomes. Like the human brain, neural networks can learn different paths to a desired outcome and can start to determine which are the best, fastest or most-efficient paths to take. Neural networks can also grow?and update as they learn.

A good example here is programming a machine to recognize an object. You start with a large number of data points that define and recognize a face or a dog. Once the object is recognized and understood for what it is, the neural network will continue to add more data points, so that it builds out the neural network. If, as in this example, the neural network is accumulating data points, it will know to add new rows of data, as well as adding data to existing rows.

So, in summary:

●???????artificial intelligence is the ability to recognize and discern

●???????machine learning provides a method to allow the device, as it learns, to predict outcomes

●???????neural networks have many parallel paths, expands the model, and learns by trial and error.

In Lenovo, our trackpads are a great example of early machine learning, particularly in how we predicted trajectories, placement and the motion of the cursor across the screen to write a better feedback loop. That's why you can point using the trackpad or track point accurately, with a much better response.

Other areas where we’ve been able to incorporate these advanced technologies are computer vision and battery life, both functions that can enhance the user experience when they can learn and respond more proactively to user behaviors. In the case of Lenovo Device Intelligence Plus, these three technologies are being deployed to predict likely device failures before they occur, all by analyzing certain statistics and data.

I believe it’s important and useful for everyone, not just programmers or device designers, to have at least a basic understanding of what makes AI, machine learning and neural networks different. Their current and potential impacts on the user experience are too great. Understanding the context of where each discipline plays helps us see where the future might be and helps deconstruct many of the myths that have emerged about them.

As author Rob Toews points out in this recent article in Forbes, artificial intelligence is fundamentally unlike human intelligence, is not simply a “less evolved” form of human intelligence, nor will it become a more powerful version of human intelligence.

What AI is, along with machine learning and neural networks, is all about knowing how to improve.