There’s AI here,
And AI there,
This AI thing is everywhere.
With AI this,
And AI that,
I’m sure they’ll next put AI in my hat

-Dr. Seuss on AI


Graphic done by Mario Arasakumar

By: Mitchell Johnstone

Introduction

Each day, companies are told to innovate or risk becoming obsolete. When executives look at the stock market, they see Amazon, Facebook, and Google – companies which have mastered the implementation of AI. They hear about AI defeating the world champions of chess, GO, and Jeopardy. They see an American Presidential Candidate running on a platform of Universal Basic Income in response to the disruption and job loss AI is expected to induce.

On top of the overwhelming hype and anticipation, there are few who understand AI. Ten different people will have ten different answers when asked to define AI. Not only is it difficult to zoom in on what it is, it's even more difficult to understand how it works. Due to the complexity and confusion, it’s easy to see why 21% of companies say they are experimenting - or 'just dipping their toes in' - and 49%, have no AI plans whatsoever. Planning for something so misunderstood does not get companies very far.

As businesses venture into the unknown, it is the visionary companies who will seize the opportunity to transform through AI. These companies will be able to serve their customers better, while significantly reducing costs. Those who remain frozen in fear, continuing with the status quo, will be left behind. In his book, Built to Last, Jim Collins explores the most visionary companies of the 20th century. Collins says that visionary companies preserve their core beliefs during technological change and ambiguity. The most successful organizations were able to change how those beliefs were applied when the world changed around them. In other words, companies must maintain strong ethics and values during the pursuit of technological transformations. Those with a guiding light can best see through the fog.

Keeping the above in mind, it becomes clear that organizations will need help moving into the future. Not only will they need an AI playbook to provide guidance on this transformation, they also need a world-class leader. Companies will need something I will call an AI Translator. This Translator is someone companies have never required before- even in prior technological revolutions.

The AI Translator is a visionary leader who understands the technical capabilities of AI and the impact of implementation. They understand the tradeoffs between transparency and accuracy and ask questions to uncover risks not recognized by others. They can explain the implications of trade-offs to the data science team and the C-suite executive. Through their understanding of AI, the Translator creates value by encouraging the pursuit of fruitful projects with high business value, while avoiding the public backlash from unintended ethical missteps.

Most of all, the Translator is someone who understands how humans think and feel. They recognize the deficiencies of human rationality and memory and empathize with those who try to improve decision making with the use of AI. They do not, however, allow this empathy to overpower their understanding of the way that computers 'think' and operate. They know the limitations of AI and uncover situations where its application would result in poor outcomes for business and for society.

To understand the importance of the AI Translator, it’s necessary to get on the same page. First, we’ll explore the strengths and weakness of cognition- both in humans and AI. Next, we will uncover why the AI revolution is happening now and why it’s so hard for people to understand the way it works. We will then explore the reason why the Translator is needed in this technological revolution and not in the last. Finally, I’ll describe the ways in which companies are creating the role of Translator now and preparing a new wave of Translators for the future.

How Humans Think

Key to the role of Translator is understanding human’s cognitive capabilities and where AI can improve decision making. Approach a couple, right before their wedding day, and ask them to estimate the chance their marriage will end in divorce. 98% of couples will say there is no chance of divorce when, in reality, the American divorce rate is around 50%. If a surgeon tells a patient their probability of survival is 95%, they receive a much different emotional response than if they communicate the patient has a 5% chance of dying. These are only two examples (of a long list) used to demonstrate the innate irrationality of humans.

Success in business (and success in life for that matter), can be greatly improved when we learn to override our default level of irrationality and set up structures that allow us to make better decisions. Charles Darwin, known for his rationality and clear thinking, built a pros and cons list to decide to marry his wife. In their annual shareholder letters, Warren Buffet and Charles Munger spend their time describing how to avoid making irrational decisions and how this avoidance is the greatest contributor to their success.

In addition to humans' ongoing battle with irrationality, our memories consistently fail us. Psychology Professor Daniel Schacter has proposed a list of 7 unique ways our memories falter:

•          Our memory weakens and we lose memory over time

•          We are preoccupied with distracting issues and don't focus attention on what we need to remember

•          We search for information that we may be desperately trying to retrieve - something we know that we know - but are blocked

•          We assign memory to the wrong source

•          Memories are implanted as a result of leading questions, comments, or suggestions when we try to call up an experience.

•          Our present knowledge influences how we remember our pasts. We often edit or entirely rewrite our previous experiences

•          We recall disturbing events that we would prefer to eliminate from our minds altogether: remembering what we cannot forget, even though we wish we could.

There is no doubt that humans are irrational and have poor memories. Our cognitive mistakes lead us to make decisions in business and life with great consequences. With this in mind, we can see why AI is an attractice option to improve the basis of human decision making. Although AI is effective at improving some of our cognitive faults, a Translator must also understand where AI can be damaging.

How Computers Think

Because AI is so hard to define, let's start with what it is not. AI won't console you when you fall and skin your knee, it won't walk you down the aisle on your wedding day, and AI wouldn't make a very good drinking buddy (although, I’ve never tested this theory). AI and computers have trouble duplicating the emotions that are key to the human experience.  Emotions are difficult to duplicate because researchers do not have a strong grasp on what makes a conscious mind or how humans experience reality. The French philosopher Descartes said, "I think, therefore I am." Computers can mimic human cognition, but given their lack of emotional replication, that does not make them innately human. In the age of AI, we may instead say, "I feel, therefore I am."

Computers and AI also struggle with creativity and ambiguity. If an AI encounters a new experience, it has trouble updating and responding. Think back to the last time you asked Siri a ridiculous question and received no response; this is because Siri has not been 'trained' to answer a particular question. Google's AlphaGo can beat the best GO player in the world, but ask it where the nearest grocery store is, and you'll get nothing in return. When AI has been trained on specific tasks, they handle these with extreme accuracy and speed. This training does not allow the AI to generalize and leaves it unable to display the dynamic thinking available in humans.

By understanding that AI cannot duplicate emotions, and how it struggles with creativity and improvisation, we narrow in on a definition of what it can do. AI is a machine’s ability to mimic the cognitive processes that we typically associate with humans. This can include conversation, object recognition, and movement through robotics. The key in the preceding definition is "mimics." Even though AI may appear to duplicate human intelligence on the surface, the strengths, weaknesses, and processes of human cognition are completely different.

Why Now & How it Works

The algorithms and concepts used to power AI have been around since the 1960s. Since its inception, there have been several proclamations about how its use will lead to world domination and extermination of humans (think Terminator or 2001: A Space Odyssey). The reason AI has re-emerged now is twofold: computing power and data. As the world turns digital, the collection of data has become prevalent across all industries. Facebook knows who your friends are, Google knows what questions you ask, Amazon knows what you buy. The world continues amassing more and more data - in fact, Forbes has stated that 90% of the world’s data has been created in the last two years alone. It's all this data that allows AI algorithms to learn.

With so much data, computers needed to become more powerful in order to process it all. In 1965, Gordon Moore, CEO of Intel, wrote a paper predicting the doubling of computer power each year. This has since become known as Moore’s Law and has held true- even speeding up to a doubling every 18 months. Data, when combined with large increases to computer processing power, allows AI algorithms to run and scale.

We can see why management is overwhelmed with the volume of data and complex mathematical concepts at work. On top of this complexity, there are several disciplines of AI to be familiar with. For most business applications today, the key area is Machine Learning. Machine Learning is the ability of computers to recognize patterns in order to decide, sense and learn without a human programming code for them to do so.

There are 3 types of Machine Learning:

•       Supervised Learning uses a large training set of labelled data that clearly identifies the desired output to make inferences on new data.

•       Unsupervised Learning allows the system to search for patterns itself in large amounts of data without giving it guidance on the desired output.

•       Reinforcement Learning is where the system is given only a sequence of actions and is rewarded and scored on how well it performs- like a game.  This type of Machine Learning makes adjustments and receives feedback on these adjustments to determine if they had a positive or negative effect on the outcome. This process leads to an optimal sequence that arrives at the best result.

We now understand that the topic of AI spans many disciplines and it can be difficult for management to grasp. Now we'll look at why the AI revolution is unique and how differences between the AI revolution and that of the internet boom reveal the need for an AI Translator.

Why the AI Revolution is Unique

In the early 2000s, the world was buzzing with excitement and software companies were gaining popularity with investors. The internet was new, and money was being enthusiastically invested in tech companies. After the 'Tech Bubble' burst in 2001, the companies that successfully understood the internet and harnessed its power became the largest companies in the world.

When we look at the current AI revolution, we can pick out several similarities from that of the previous technology boom. Reminiscent of the 2000's, Investors are buying  the hype with $9 billion in venture capital investment flowing into the AI space. Just as there was confusion about software, there is widespread misunderstanding about AI’s capabilities due to the mathematical and statistical complexity of the underlying algorithms. The internet has transformed the world over the past 25 years and these simliraties make us aware that AI is about to do the same. Although there are several similarities, it would be a mistake to say this revolution is the same. The differences below point to the area's companies must address if they are to successfully harness the opportunities AI enables.

Unintended Consequences

Traditional IT projects were “set it and forget it.”  If you implemented a robust system, you could spend a minimal amount of money maintaining it, depreciate it over a period of time, and decide when and how to enhance, or decommission it.  Because AI systems continue to learn, organizations must be aware of the need for a feedback loop to ensure the systems continue to work as expected and avoid unintended consequences.

Brand New Risks

Historically, conversations about IT delivery projects centered around timeline, scope and budget.  When implementing AI, companies must make decisions regarding the transparency of the algorithms versus the value that can be created for the business. Companies must weigh the risks of designing systems that perform well but lack the explainability required to determine if their outputs align with company values.

A New Approach

AI is uniquely disruptive because it has so many downstream effects – both intended and unintended. Without a manager who can adequately understand the risks, businesses can get themselves into trouble. Earlier, when defining what AI is, we recognized that AI lacked the human ability of connection and emotion. AI is only as good as the data it has been trained on and only weighs factors provided by human creators. “Garbage in, garbage out” means something different in the context of AI.  If you train an AI algorithm using biased data, you get a biased algorithm. The algorithm does not have an emotional framework to alert it when it’s doing something wrong.

Without a deeper understanding of how the technology works, it’s easy for AI deployment to take an unintended turn.  You can’t control what an AI algorithm will learn, but you can be aware of how it learns to prevent unintended consequences. Without an AI Translator, companies risk profits, customer satisfaction, losing control over their business model and, ultimately, creating a society that no longer supports equal opportunity for all.  Let’s look at some examples to help illuminate what I mean.

The AI Translator & Ethics

Many of the recent advancements in AI involve algorithms with very little explainability and transparency. Imagine a bank deploying AI to approve customer loans. They want the best algorithm with the most accurate predictions to achieve maximum profits with minimal loan losses. Without a leader who understands the functioning of the algorithms, and the tradeoff between predictive accuracy and transparency of predictions, the bank risks unknowingly rejecting specific groups of customers for loans.

Imagine being a researcher for MIT's Media Lab, one of the most advanced technology research labs in the world. You are experimenting with new facial recognition software that uses AI. The software is relatively new and is having difficulty picking up on your facial features. You ignore this as a technological error, but then realize the algorithm has no difficulty picking up the faces of your white colleagues. It turns out, the algorithm was trained on a dataset that included almost all white faces. Without an AI Translator, data science teams will pursue the task given to them. They are incentivized to build the most accurate model with the data they have and are not charged with considering the ethical implications. An AI Translator can help break down barriers and create AI that can be used equally by all.

Now consider a third example from 2015. Amazon wanted to improve their hiring practices, so they developed an algorithm that used data from employees hired in the past and trained an AI algorithm to identify high-quality candidates. Because they did not have the oversight of an AI Translator, the algorithm spread biases that originated from previous hiring practices. The algorithm did not identify and hire candidates in a gender-neutral manner because Amazon had hired mostly male software developers in the past.

We can see that numerous ethical issues arise in AI projects. As data scientists build solutions, well-articulated values and good intentions can get lost in the data and mathematics. Neither the motivations, nor the expertise, of the data science team involves translating the company vision and values into an AI system. This is where the AI Translator can create incredible value for companies attempting to pursue innovation.

Translators in Action

Scotiabank is creating a framework to help companies build towards ethical AI solutions. Large firms like Deloitte, Accenture, SAS, and others have set up entire departments to act as the AI Translator. They provide companies with management consultants for business expertise and data scientists to build useful solutions. Their organizational structure is an acknowledgment of the importance of the Translator and lays the groundwork for ethical deployment.

Business schools have also begun to recognize the importance of the Translator. Smith School of Business at Queen’s University has launched the first Master of Management in Artificial Intelligence, with the University of Toronto launching shortly after. This is in direct response to the needs of organizations struggling through their AI transformations and of those who recognize the importance of ethical implementation. The graduates of these programs will be equipped with the management and technical skills needed to play the role of AI Translator immediately.

In the last technological revolution, Steve Jobs combined science and technology with his adept business sense and built the most successful company of the 21st century. The CEOs of the future will need to be equipped with the skills of an AI Translator to pair algorithms and strategy with rock-solid ethics. Perhaps the next visionary company is one with an AI Translator at the helm; one who leads to a fair society for all.

Read more about AI topics for managment at www.abacusaisolutions.com