From January to April 2019 I am involved in a seminar course on the Socio-Cultural and Political Implications of Artificial Intelligence (ARTS490). This article is part of a series of essays written for the seminar on AI’s implications for society.

I. Zones

People want to know how machines work. The algorithms that make decisions for us are confusing to some and downright dangerous to others. But of course, not everyone begins their education in AI with the same level of information or learns the same way. Instead, we create a variety of explanatory material that meets people where they are at.

Vernor Vinge writes a series of novels where the Milky Way galaxy is divided into concentric volumes called zones of thought. Each zone has fundamental differences in its physical laws that affect how both biological and artificial intelligence can develop. These differences make it difficult for technology from higher zones to continue functioning in the lower zones. This seems like a fitting analogy for our methods of explanation when it comes to artificial intelligence. It is characterized by the alignment of a level of understanding with ones method of interaction.

I see the three zones of understanding in the realm of artificial intelligence:

1. Motivation Zone is where people are inspired to understand the ways different intelligences work and how they are affected by them. Often taking the form of art, business, or news. Engaging with the motivation zone is an act of consumption.

2. Exploration Zone is where people begin to understand the mechanisms of AI by creating experiments, modifying the machine learning system, and observing the change to the outcome of the experiment. Exploring is defined by feedback. The zone is responsive to an individual’s inputs.

3. Creation Zone is where people can create new forms of understanding artificial intelligence. Interaction here is defined by production.

Each one provides access to tools and materials which the earlier zones would find difficult to take advantage of. Moving through all three zones provides a more complete picture of what artificial intelligence is.

Unfortunately it is more often the case that these zones are viewed as completely separate. Some computer scientists skip directly into the creation zone, assuming their previous experience with algorithms, data structures, and programming will carry them to a place of understanding. Some business owners find the motivation for their work without exploring the concepts they claim to understand. Many people critique others’ motivations in the development of artificial intelligence without having any idea how the development actually occurs.

The rest of this article is for those who have found themselves sequestered in one zone without having had a chance to experience the others. Compare my conceptions of AI understanding with your own. See if there are areas you’d like to explore. Suggest extensions or reimaginings of these zones if you find them inadequate.

II. Motivation

Motivation means interpreting.

Inside the tech industry it can feel like artificial intelligence is the catch-phrase on everyone’s lips. Outside of it, the perception of and experience with it can be very different. Helping people interpret the current state of the industry through their experience is critical to motivating further research and development.

Motivation to learn more about a subject can come in many forms. Art is often a motivator that people consume without interacting with it. Or more accurately the interaction happens inside ones mind. I’ve found art so inspiring or thought-provoking at times that I have dropped current projects to learn about new things.

One of my early inspirations to learn more about computers was the 2007 film Live Free or Die Hard which featured a group of activists exploiting software and hardware flaws to shut down the United States’ energy systems. Fear-inducing media is quite common in the AI space as well. In as much as they inspire conversation or further research, they are successful motivators for improving understanding.

Another form of motivation can come from projects with more narrow intended audiences. Here are some great examples of projects that help motivate specific groups of people to understand what AI is and how it might affect them:

• Social. Mimi Onuoha and Diana Nucera wrote A People’s Guide to AI to help society learn what AI is and how it might affect our lives right now.

• Business. Venture capital firm Andreesen Horowitz released an AI Playbook that helps organizations answer the question “What can I do with AI in my own product or company?”

• Art. Another Mimi Onuoha creation is an art piece, The Library of Missing Datasets v2.0, which draws attention to the under-representation of black people in American datasets.

III. Exploration

Exploration means feedback.

Most content that we consume only flows one way. We watch a youtube video or read a book, and the only way to interact with the media is by thinking about it or discussing it. The medium itself does not change in response to thoughts. When people enter the exploration zone, this all changes. Explanations become interactive, and content responds to manipulation in ways that deepen understanding.

Bret Victor sums this idea up well in this essay on explorable explanations. This is an emerging area of work in the field of artificial intelligence. It takes time to both come up to construct these explorations in a way that is instructive.

Often these take the form of interactive models, where parameters can be changed on the fly and the corresponding change in output can be observed. One of the first that I remember encountering was the Tensorflow Playground. When it was first released my entire office stopped work to play with the parameters and see if we could match the data by constructing our own neural networks.

Here are some interactive explorations of machine learning concepts from around the internet:

• Training Models. The Distill journal is dedicated to publishing in-depth explorations of AI concepts. A great article start with is Why Momentum Really Works about learning rates in gradient descent.

• High-Dimensional Input Data. Experiments with Google is a collection of AI experiments that can be manipulated by the user. The Embedding Projector is a great example of one that helps with understanding high-dimensionality input data.

• Biological Inspiration. Nicky Case introduces some concepts of neuroscience in the Neurotic Neurons explanation of Hebbian and Anti-Hebbian learning.

• Generation. The Met created an exploration called Gen Studio that uses generative adversarial networks on art found in their galary to generate new pieces. It helps a user explore the output of different models.

IV. Creation

Creation means participation.

To enter the zone of creation is to be a part of making all the things I just linked to above. In creation we harness lower level tools that might not be accessible to everyone in the motivation and exploration zones.

Developing new models, datasets, tasks, and applications in the field of artificial intelligence is at the core of moving the field forward. The creation process generally has a longer feedback loop than exploration. It takes time to develop an idea to the point where it can be shared and others can give input to it or it can be tested in new contexts.

The output of the creation zone can be found in a few common places:

• Code. Public code seems to be mostly found on Github

• Research Papers. In machine learning they are all generally found on arxiv first. OpenReview is also a great place to visit in order to see the discussions around proposed or accepted journal and conference papers.

• The Other Zones. Part of improving the applications of artificial intelligence is helping people understand it. Constructing tools for exploration and motivating media is a significant achievement of the creation zone.