How to speak AI

Artificial intelligence methods are based on mathematics, but the words of AI are also remarkably important. AI speak can sound like a foreign language. Like any language, speaking it fluently starts with command of basic AI words and phrases, jargon that is commonly used but often difficult to interpret consistently. For an introductory AI vocabulary lesson we welcome Dr. Julie Rosen, Leidos Vice President, Chief Scientist, and Technical Fellow Chair.

Q: It seems like the explosion of AI has also led to the explosion of AI buzzwords. Why is it so important to clarify the meanings of these words?

Dr. Rosen: When trends emerge or expand quickly, confusion often arises. The rapid maturation of AI has led to certain parts of the lexicon becoming unclear and confusing. We need strong working definitions to base not only technical conversations, but also broader ones involving funding sponsors and consumers of AI findings. It’s important to speak the same language because as technology leaders, if we aren’t clear, people will be misinformed. As far as business goes, if it appears that we don’t understand the nuances of AI speak, our customers will think we’re not expert in the field.
 

Q: Let’s start with the basics. How do you define artificial intelligence?

Dr. Rosen: The concept of AI is as old as computing, and its definition has changed over time. In terms of modern technology, AI is the broad domain of digital devices programmed to mimic human behaviors and intelligence. In the past decade or so, computing technology has advanced enough to allow for greater computational investigation into the gobs of data at our disposal, which is mostly what we mean today when we refer to AI. Within the field, we’re trying to learn what these data are telling us, and we’re using new methods of interrogating the data that are more advanced than ever before. So today’s AI is mostly about exploring new ways of saying ‘OK data, tell me what you know.’

Q: AI and machine learning (ML) are often used together. Are the two synonymous?

Dr. Rosen: Not quite. ML is a subset of AI, but there are other flavors of AI as well, including rules-based expert systems and knowledge graphs, and it’s important not to conflate the two. Methods of AI can be thought of like a Russian doll set. AI is on the outside, then ML, then deep learning. A true data scientist should know the full range of models and methods to consider for a given purpose.

Think of ML as the reasoning engine that consumes data, applies algorithmic logic to recognize patterns in the data, and sends the analytic results to a human or another computing component. General ML methods work with a layer of observed data (input), a layer of analytic output (e.g., forecasts, matches), and a hidden (or latent) layer where the attributes are connected to each other, and to the input and output layers. If the data are appropriately curated for like-comparisons, then ML techniques have the ability to learn new patterns by operating with a few initial rules, which mature algorithmically as data are processed over time. The beauty of ML is it allows us to characterize situational reasoning without the need for a full complement of programmed rules and instructions, which are dangerously brittle in real-time decision making with uncertain or missing data.

Q: What do you mean by deep learning?

Dr. Rosen: Deep learning (DL) is a machine learning technique, a set of algorithms that model phenomena measured through complex data. The etymology of “deep” in deep learning refers to the depth of the model. Consider huge, complex data sets like the human genome, social media data, or satellite imagery. When you’re trying to forecast outcomes or recognize patterns within the data, each measurement associates with many attributes, also called features. In these cases we need more advanced models with more layers to learn features among the patterns in the data sets. When you’re working with unstructured text and imagery, for example, these features become deeply interrelated, and the multiple layers of the DL network can start to resemble a big, hairy fur ball of connections. Now you’ve got to start digging deeper into the possible connections to determine what patterns you can infer from them. Such digging can get computationally intense, which why a new kind of chip, called GPUs, are in demand to train deep learning models.
 

Q: Speaking of networks—what’s a neural network?

Dr. Rosen: Think of a neural network model as a computer that mimics how the human brain works electrically. The brain functions through a complex set of neurons, its basic working units, which modelers call nodes. Like our brains, neural net models are highly interconnected. They can be hundreds of layers deep and very wide. The brain has all these synapses, which modelers call edges or links, between nerve endings in the various cells in the brain. You want to be able to follow the connections among all those cells. In math-land we call this a multiply connected network. So if you can explore deep into the brain, maybe you can find a network path or multiple paths that help you connect one nerve ending to another. Or in the case of mathematical modeling done with ML algorithms, you’re looking for a path of similarity or a path of feature closeness. But to make these discoveries, the network itself has to be highly multiply connected. It has to be very deep, hence DL modeling, to answer very large and complicated non-categorical kinds of questions.

Q: You mentioned data scientist, which is a major career path in AI. Walk us through the difference between the terms data scientist, data engineer, and data analyst.

Dr. Rosen: First off, let me say that these terms are greatly overlapping in academia and in the general marketplace. At Leidos, we use the term data scientist to refer to researchers who create data models and algorithmic methods to analyze big, highly complex data sets in order to find patterns. Often these patterns are not yet prescribed and too difficult to detect with traditional rules and static statistical methods. These professionals typically operate at the basic R&D end of the maturity spectrum, proving the accuracy and efficacy of the AI model or method.

At the opposite end of the technological maturity curve are Leidos data analysts, who employ these proven models, methods, and tools to investigate mission-specific data in support of decision making. Data analysts’ mission is to impact the bottom line. As such, it’s very important these professionals know their consumer’s domain and what questions to ask the data.

On a Venn diagram, the Leidos data engineer overlaps with data scientist (to get the models and methods matured to scale and deployed in an end-to-end pipeline) and data analyst (to make the pipeline usable and maintainable). Data engineers have similar skills to generalized system engineers and solution architects. They start off with specs from a customer and work with third-party vendors to integrate commodity tools with custom-developed and optimized models and methods. Data engineers have the additional charge of understanding the impacts of design options to ingest and curate these big, dynamic, fast-arriving, diverse, unclean, mission-specific data sets. The data engineer is charged with making the data scientist’s accuracy-proven models run faster, at the scale of the incoming data, and output the analytic findings through visualization, explanation, or communication element.