How to Implement Artificial Intelligence: An Introductory Overview

 

A Crash Course in Geek Speak

“Companies like calling their technologies AI. It sounds better, it’s more futuristic, but it’s not AI: it’s actually data analytics.” —Alan Smeaton, Founder, The Insight Centre for Data Analytics

Artificial Intelligence Vocabulary Words you Should Adopt Immediately

Thanks to the entertainment industry and intermittent waves of media hype, the term artificial intelligence (AI) conjures up a lot of drama and misperception. Academics divide AI into two categories: artificial general intelligence (AGI), which refers to the hypothetical ability of a machine to perform intellectually as a human, and narrow AI, which refers to machines solving more targeted problems or reasoning tasks. The possibilities for AGI are mind-blowing, and we believe they will be here sooner than many predict. But rather than waking up one day and noticing, “Hey, we have AGI now,” it will be more of a gradual transition from millions of narrow AI applications that drive progress to a world where machine intelligence is irrevocably intertwined in everything we do. And that is when the super evolution will have occurred.

For practical purposes, the term operational AI is used to describe an intelligent system designed for real-world, commercial-scale applications (i.e., it is integrated with the routine usage of a business). For example, Molecula falls within the operational AI category because it helps organizations leverage big data and AI to achieve short-term business goals within a long-term AI strategy. 

In October of 2020 Gartner reported that nearly half of CIOs say they now employ AI or intend to within the next 12 months. There’s no doubt that developing AI capabilities will continue to become a more pressing responsibility for IT leaders wishing to remain competitive in their industries. 

Machine learning is another term that is often heard alongside AI. Machine learning (ML) is a subfield of AI. ML uses statistical models that automate the process of extracting knowledge and identifying patterns from data to enable better decision-making. What makes ML particularly powerful is its ability to evolve and improve upon itself without a predetermined set of rules.

Neural networks are algorithms with instructions that make decisions on bits of data. With each decision, a weight is given to a particular outcome. Neural nets can process a variety of input, such as images, videos, files, and databases.

Deep learning is a part of machine learning that uses neural networks to compute results. Deep refers to the number of layers in the neural network.

A visual illustration of the relationship between analytics

A visual illustration of the relationship between analytics, AI, ML, neural networks, and deep learning helps clarify terms that are often incorrectly used interchangeably when discussing advanced analytics and data science.

 

The Artificial Intelligence Implementation Lifecycle

While every project will have unique needs, the general process will be the same for most AI initiatives. However, before you approach your initiative with a project-mindset, keep in mind that you will be investing in a technology that has the potential to improve many aspects of your business. As you make technical decisions, consider how your investments can be leveraged and reused across your organization, now and in the future. 

The AI implementation lifecycle

Every project is unique, but most will follow the basic steps outlined in the AI implementation lifecycle above.

 

The Artificial Intelligence Dream Team

Whether you are outsourcing your AI initiatives, building them in-house, or a combination of both, the execution team is a critical part of your success. Technical skills and interpersonal skills are of course important, but when it comes to the rapidly changing nature of the AI and big data space, attitude and culture are equally critical. 

A functional AI team is made up of members who can see the big picture, understand that their roles and responsibilities can be fluid, and aren’t resistant to change. 

If your company isn’t already digitally native, you may find that you need to advocate in favor of AI across the organization. Setting up information sessions or informal training groups can go a long way. AI is not something that should be bestowed upon one department. There is much to gain from centralizing your AI strategy so that everyone in the organization understands and can participate in how AI could improve the business.

 

The following roles are considered the nuts and bolts of a successful AI team:

Screen Shot 2021 11 09 at 9.25.42 PM

Business Expert


This person is a subject matter expert in your particular industry and has deep knowledge of your business goals, your budgets, and how success will be measured.

Screen Shot 2021 11 09 at 9.25.42 PM

Data Engineer


The data engineer is in charge of collecting, managing, and providing access to all the data your organization requires. They often have a background in IT or software engineering. They are experts at distributed systems, building data pipelines, and preparing, monitoring, and securing data for analysis. 

Data engineers are responsible for responding to information requests submitted by data scientists. This is a very manual and labor-intensive part of their job. It can also be repetitive as they often have to provide similar data across multiple jobs, but since each request requires so much processing, the work is not reusable. 

The engineers are also responsible for keeping data secure. This can be challenging when data scientists perform data modeling on personal notebook computers, as is often done, making it difficult to track and secure sensitive data.

For many reasons, which we won’t get into here, there is currently a shortage of data engineers. When hiring, be realistic with expectations and seek the advice of an HR professional who specializes in tech, if needed. We’ve seen job posts asking for 10 years of big data experience—which is a bit laughable since big data as we know it today didn’t exist 10 years ago. The most important thing is to hire data specialists who have an aptitude and attitude that align with continuous learning and new technology adoption.

Key Skills Needed: Programming, database systems, mathematics, big data architecture, ETL system design, security best practices, monitoring
Key Technologies Used: Java, Python, Scala, Hadoop, SQL, NoSQL, AWS, Spark, Hive, Kafka

“Big Data” job titles

Screen Shot 2021 11 09 at 9.25.42 PM

Data Scientist


The data scientist is responsible for understanding the business goals, identifying valuable data sources, and training and refining models to discover trends and make predictions that are aligned with the business goals. They are also responsible for communicating results to the team.

They also often track down, clean, and transform data, which is a stereotypical sore spot in the industry and provides folly for regular social media memes. Data scientists don’t always have formal training in data prep or don’t enjoy that aspect of the job. The trouble is that data prep frequently consumes the majority of their time. This is partially due to the nascent nature of the data team dynamics in many organizations and partially due to the shortage of data engineers.  

Many of the data team roles are presently being defined and refined. A general rule of thumb is to hire two to three data engineers for every data scientist, but the ratio may need to increase with more complex data situations.

Key Skills Needed: Statistics, mathematics (linear algebra, calculus), data modeling, programming, storytelling, business communication
Key Technologies Used: SQL, Python, R, TensorFlow

Screen Shot 2021 11 09 at 9.25.42 PM

AI Engineer


The AI engineer (also referred to as an ML engineer) sits at the intersection between data engineering and data science, developing working software that makes use of the data, as well as automating predictive models. They ensure production readiness, build a deployable version of production data to feed the models built by data scientists, and integrate these models with the end product.

The AI engineer is relatively close to the data engineer in nature, but with a specific focus on building reliable, scalable production data pipelines to feed models in production, and then maintaining them long term.

Key Skills Needed: Programming, mathematics, statistics, big data architecture, algorithms and frameworks, communication, problem-solving 
Key Technologies Used: Java, Python, Scala, Hadoop, NoSQL, SQL, Python, R

 

The intersection of these four roles is where the magic happens. Focusing efforts on improving efficiencies within the AI ROI area will ensure the best operational AI outcomes.

Beyond the roles outlined above, there are many other players in a successful AI initiative, depending on the specific job at hand. Keep in mind, the rapidly changing nature of the industry means many people end up wearing a variety of hats and there can be confusion about who owns which responsibilities. A data scientist at one company may have significantly different duties than at another company. Communication is the best antidote to these problems, as is ensuring your AI initiatives are adequately funded and prioritized in the organization. 

Additional roles in the AI process might include: AI Architect, Business Analyst, Software Engineer, Software Developer, Full Stack Developer, Project Managers, Legal Experts, and Ethicists.

The AI Dream Team

The goal is to have an equally strong team in the business, data science, and engineering disciplines. Good communication among these functions will ensure that projects are technically optimized with the big picture in mind. The business team benefits from understanding the technology issues, and the technical teams benefit from understanding the motivations behind business goals. The cross-pollination of ideas leads to more efficient and creative solutions.

Data Team Skills Chart

A data team might be composed of several specialists who have overlapping skill sets. Each role typically requires a different level of expertise in key areas. Keep in mind that knowing “how” to do something is not the same as being an expert practitioner in that area.

Data Team Roles in the AI Process

Each of the above roles plays an important part in the data engineering and AI processes. They will typically overlap at different stages throughout a project’s lifecycle.

Download our complete Feature-First Field Guide to read more. 

 

Download Field Guide