Applied Artificial Intelligence: An Introduction For Business Leaders

by Mariya Yao

The goal of data mining is to extract patterns and knowledge from large-scale datasets so that they can be reshaped into a more understandable structure for later analysis.

Rule-based expert systems are most effective when applied to automated calculations and logical processes where rules and outcomes are relatively clear.

Machine learning enables computers to learn without being explicitly programmed. It is a field in computer science that builds on top of computational statistics and data mining.

Unsupervised learning occurs when computers are given unstructured rather than labeled data, i.e. no input-output pairs, and asked to discover inherent structures and patterns that lie within the data. One common application of unsupervised learning is clustering, where input data is divided into different groups based on a measure of “similarity."

practice, using simpler AI approaches like older, non-deep-learning machine learning techniques can produce faster and better results than fancy neural nets can. Rather than building custom deep learning solutions, many enterprises opt for Machine Learning as a Service (MLaaS) solutions from Google, Amazon, IBM, Microsoft, or leading AI startups.

Probabilistic programming enables us to create learning systems that make decisions in the face of uncertainty by making inferences from prior knowledge.

Probabilistic programs have been used successfully in applications such as medical imaging, machine perception, financial predictions, and econometric and atmospheric forecasting.

There are four broad categories of ensembling: bagging, boosting, stacking, and bucketing. Bagging entails training the same algorithm on different subsets of the data and includes popular algorithms like random forest. Boosting involves training a sequence of models, where each model prioritizes learning from the examples that the previous model failed on. In stacking, you pool the output of many models. In bucketing, you train multiple models for a given problem and dynamically choose the best one for each specific input.

Systems That Predict are systems that are capable of analyzing data and using it to produce probabilistic predictions.

With AI now used in high-stakes systems to identify terrorists, predict criminal recidivism, and triage medical cases, homogenous thinking in the technology industry has dangerous implications.

“Tools are not meant to make our lives easier,” says Patrick Hebron, author of Machine Learning For Designers, “[t]hey are meant to give us leverage so that we can push harder. Tools lift rocks. People build cathedrals.”(40)

headquartered in San Mateo, California, sponsors projects at the University of San Francisco (USF) where successful students can progress to paid internships and eventual employment after graduation. These partnership programs have become so popular that there are more companies proposing projects than students ready to staff them. Companies thinking about going this route should propose unique projects and clearly articulate the benefits of participation to attract the best students.

Likewise, those that can leverage their proprietary methods to extract maximum value from smaller but more relevant data stores can also excel. Look for partners with access to a lot of data that’s relevant to your domain.

Facebook's true north metric for platform growth is the number of members who connect with ten friends in seven days.(77) Just having a user sign up for an account is insufficient to inspire the engagement rates that Facebook needs to later monetize that user through advertising. Similarly, Slack focuses on teams that have exchanged at least 2,000 messages.(78) Once a team has reached this threshold of usage, they're much more likely to stick around and eventually upgrade to paid plans.

Accuracy, precision, and recall build on these concepts, and they are the most common evaluation metrics for classification tasks, in which a model evaluates some input as belonging or not belonging to a target category.

Changes in data pipelines, data structure, or external conditions all need to be addressed, or they may affect the accuracy of your model.

Since machine learning processes are iterative and improve over time, you can start small and slowly expand your resources.

Anand Rao, Innovation Lead for US Analytics at PwC, and his team use the agile method to run four-week sprints on AI projects, during which they transform an idea into an initial implementation.(82)

In order to support a large number of enterprise-wide machine learning systems, you will need a centralized technology architecture that provides a stable development and deployment environment.

These MLaaS systems, also known as end-to-end machine learning platforms, reduce the time required to push models to production from months to weeks.

Overall, successful MLaaS systems have the following characteristics(89): Algorithm-agnostic. The platform supports numerous machine learning algorithms and innovative combinations of these algorithms. Reusable. Each machine learning algorithm can be reused in other applications. Simple. The system is easy for engineers of varying levels of technical experience to understand and use. Over time, the steps should become fully automated. Centralized knowledge. Information on past experiments, including results, is easily accessible for future reference. Flexible. The platform is capable of handling a variety of data types and learning tasks specific to your company and industry. Reliable and scalable. The platform remains resilient and be able to scale with high volumes of data during both the training and production phases. Intuitive user interface (UI). The system has a simple user interface to allow engineers and even non-technical domain experts to easily manage experiments as well as visualize and compare outputs.

As a rule of thumb, half of your time should be spent on measurements and maintenance rather than on model creation.

Data, digital transformation, and machine intelligence will simply be table stakes for any organization that wants to stay competitive in an increasingly automated world.

To help you navigate through the complex range of AI solutions, we created an Enterprise AI Landscape to track leading companies that have deployed machine learning solutions to streamline their business functions. You can access the guide on our website at appliedaibook.com/resources.

Fortunately, these tasks are well-suited for automation by Robotic Process Automation (RPA), which are software robots programmed to perform a specified sequence of actions. Even better, RPA deployment is relatively fast and low risk, so that problematic robots can quickly be removed without detriment to existing systems. Examples of workflows at which RPAs excel include performing regular diagnostics of your software or hardware, creating and updating accounting records (such as payroll), or automatically generating and delivering periodic reports to the relevant stakeholders.

For companies that do not have the capacity or the desire to tackle data silos on their own, companies like Maana, Alation, and Tamr offer ML-powered data unification and cataloguing services.

Ayasdi and Kyndi leverage different machine learning algorithms to extract patterns and make predictions on your company’s data, while DataSift specializes in sifting through and classifying natural language textual data to track social sentiments.

About 64 percent of consumers now expect real-time responses at any time, and 65 percent say they are likely to switch brands if they receive inconsistent customer service across platforms (online, in-store, phone, text, or via email).