The artificial intelligence (AI) hype isn't going away any time soon. So as a digital professional, your job is to figure out what AI is really about and how you can use it to your advantage. But where do you start if you're not a machine learning programmer or an AI expert?
A good place to start is "Applied Artificial Intelligence – A Handbook for Business Leaders" by Mariya Yao, Adelyn Zhou and Marlene Jia. The authors wrote a practical guide for business leaders who want to use machine intelligence to improve their organizations and communities. It's a great read not only for business leaders, but also for digital and marketing professionals.
Why a Book on AI?
The authors believe AI will eventually follow the ubiquitous role of the internet, email and mobile devices in organizations and our society. But as the authors say: “Sadly, market hype about AI has made it trendy to brag about using AI without any understanding or experience.” They warn readers against thinking that by buying AI “your organization is probably not going to turn into Google overnight.” Instead the authors recommend a course of long-term thinking, working hard on your information architecture and governance, and executive commitment. "Applied AI" shares examples of AI bragging versus true AI examples we can learn from.
The book consists of three parts which bring readers from a clear explanation of what AI is, to how to do it, and concludes with AI best practices. You could go directly to Part III to find some good examples, but reading the first part of the book won't hurt and Part II has some great insights I'll share with you in this review.
- Part I: What Business Leaders Need to Know — Gives an essential education in the state of AI today.
- Part II: How to Develop an Enterprise AI Strategy — Tells you about the steps to implement successful AI projects.
- Part III: AI For Enterprise Functions — Highlights popular AI application or common business functions.
Part I: Why AI Is Weak, and Why That's OK For Now
The book begins with a comparison of AI versus Artificial General Intelligence and some modern AI techniques. You learn why the current AI is still “weak” or “narrow” and designed for one specific task. Yes, it can defeat humans in specific tasks, but will fail when it needs to solve tasks outside of the program's original parameters.
To help business executives comprehend the functional differences between the different AI approaches, the authors designed the Machine Intelligence Continuum (MIC), which divides AI initiatives into seven levels. Going from the lowest level: Systems That Act, to Systems That Predict based on data analytics, to Systems That Learn without being explicitly programmed to do so. There are already some examples of Systems That Create, and Systems That Relate.
The highest AI levels are Systems That Master. They are capable of constructing abstract concepts and strategic plans from sparse data. The authors claim our current AI hasn't reached this level yet: “Humans are Systems That Master, current AI programs are not.” The final level is Systems That Evolve, exhibiting superhuman intelligence and capabilities like dynamically changing their own design and architecture to adapt to changing conditions in their environment. This is the level that makes singularists very excited and AI gloomers very worried. Until today we can enjoy this AI level in SciFi and apocalyptic movies and literature. But it's still currently a fantasy.
The promise of AI for human benefit
The book provides ample best practices and examples of applied AI, with the choice of those examples being a selling point of the book in my opinion. The authors' examples go beyond driving advertising clicks, streamlining sales and boosting corporate profits. Instead the authors share examples of AI being used to fight social injustice, solve pressing community problems, and improve the quality of life for everyone.
The authors not only talk about the benefit of AI in medical diagnosis, but also about a micro-finance case to prevent rural farmers in India from committing suicide and UNICEF's U-Report that enables young people in developing countries to report social injustice in their communities. Both cases use data to change policies and take actions in real time which literally saves lives.
The challenges of bias, discrimination and even malicious AI
The authors find the lack of diversity in the technology industry alarming. They see data and technology as human interventions, ideally designed to reflect and advance human values. To quote the authors: “As our creations (of technology and IA) grow exponentially more powerful and their footprint even larger on our society, we need to be increasingly mindful of the need to build them to be robust against adverse and unintended consequences.”
Yao, Zhou and Jia warn against blindly trusting the output of automated systems without vetting the accuracy of both the input data and the decision making process itself. They give examples of where it went really wrong and where the biases of technology creators “trickled down” to their creations. We hear about examples like this almost every week and recently some large tech companies, including IBM and Apple, as well as government and NGOs have begun to speak out about this problem.
AI was never intended to be evil. But the authors are very clear that intentions and actions do not necessarily align: "the probability of bad people taking advantage of intelligent automation or evil purposes is 100 percent.” Let that message sink in for a moment. Think not only of cyber attacks and hijacking autonomous weapons and vehicles, but also illegal surveillance, propaganda, deception and social manipulation. The authors create a sense of urgency to get our policies and governance — and protection — in place before it's too late.
A safe and ethical AI. Fix it!
The authors end the first section by pleading for a safe and ethical AI. AI must be designed to avoid (un)intentionally harming human society. And following Isaac Asimov's "Three Laws of Robotics" doesn't cut it — it says a lot about people's technology ignorance when they refer to a SciFi book for our ethical guidance ....
So it's good to hear there are many initiatives out there whose goal is to shut down "rogue AI." The authors also stress the importance of widespread education in technology, AI and ethics to democratize decision making on AI so it doesn't rest in the hands of an elite few.
3 Principles to avoid bad AI
As a solution, the authors share three principles of collaborative design that will prevent from creating harmful AI. The first principle is to build user-friendly products to collect better data or AI. This is no platitude: if the products collecting data to power AI systems hamper the user interaction then the chance of getting incomplete, incorrect and compromised data is very high. As a consequence the decision making based on this wrong data will be false. Imagine the impact of that for your organization, your community, or you as a person.
The second principle is to prioritize domain expertise and business value over algorithms. A solution — even a perfect one — for a problem nobody needs to be solved has no value. You need a thoughtful UX design, domain expertise and business acumen to make the difference with AI.
The third principle is to empower human designers with machine intelligence. I think Patrick Hebron, founder of machine intelligence design at Adobe, put it best: “Tools lift rocks. People build cathedrals.” AI can simplify design tools without limiting our creativity or removing our control. Personally I have high expectations on this use of AI. Artists and creatives will likely lead the way in exploring things unimaginable now in their use of AI.
Part II: The Importance of Strategy
Part two dives into how to develop an enterprise AI strategy. With the question “Do You Have a Central Technology Infrastructure and Team?” the authors get to the heart of the challenges with AI and everything digital. Without a strategic foundation, the siloed business units will just implement single use AI solutions — and no APIs will fix that. Laying this foundation is hard and long term work. You need C-suite sponsorship of the necessary ongoing collaboration between all departments and business units.
But let's say you have the technical architecture, governance and collaboration in place. AI will likely give you all kinds of outcomes that you didn't expect, and that demands you change your way of doing business. Are you willing to follow up on the adagio of “data-driven-decision-making”?
AI talent is scarce
Finding AI talent is a challenge in itself. According to Bloomberg, only about 22,000 Ph.D-level computer scientists around the world have the required experience to build cutting-edge AI systems. And 54 percent of all deep learning specialists already work for six technology companies. Fresh Ph.D.s and others with limited work experience are regularly paid between $300,000 to $500,000 a year or more in salary and company stock. In other words: good luck with getting AI talent before Google, Amazon, Facebook and a handful of other technology leaders.
To help you win this battle, the authors first explain what the different AI related job titles are, then describe the ideal characteristics of an AI expert and conclude with some recruitment strategies.
Data is not reality, and AI is not a silver bullet
The authors share an extremely important lesson: data is not reality, it's a human invention. So data alone is never a ground truth. Here the authors present some commonly seen mistakes with data, such as having undefined goals, and errors in definition, measurements, capturing, processing, sampling and inferencing.
The error of all errors is of course the unknown error. There are unknown unknowns between your representation of reality, in the form of data, and reality itself. However, in the example of a consumer using a digital product you could know something about her motivation (the unknown) through qualitative research.
Yao, Zhou and Jia also warn against believing AI can solve anything. Even perfectly-implemented algorithms will fail without the right data. Here they introduce accuracy, recall and precision to non-technical business leaders. They take great effort in explaining these rather complex concepts, and they do it well. It is essential for every one who deals with AI to understand these concepts.
Technical debt: The 'boring' part of AI
The final chapter of Part II explains how AI can start small and grow over time through iterations. The idea of “Technical Debt” is introduced here. Writing code is just a tiny fraction of the AI work. Most of the time is consumed by harvesting, and cleaning data and maintaining existing models. That is hard and “boring” work! Many AI experts hate doing this kind of work. They prefer coding new cool AI solutions and ignore the maintenance of their previous AI projects.
The authors raise once again the importance of a centralized technology architecture. Companies like Google, Facebook, Uber and Airbnb have created what they call an internal Machine Learning as a Service (MLaaS) which reduces the time required to push learning models to production from months to weeks. Can you compete with that?
Part III: AI for Enterprise Functions
The third part of the book covers the applications of AI in businesses. It starts with a chapter on obstacles and opportunities. Businesses should deploy existing AI techniques to optimize common business functions. The authors claim the market offers multiple AI products that are targeted at inefficiencies in virtually every enterprise function. They are very good at streamlining processes and taking over rote tasks such as triggering a workflow.
The authors provide an Enterprise AI Landscape as a resource on the accompanying website for the book which gives a sense of the scope of products available now.
AI is (not) like teenage sex
The final chapter of the book starts with an online meme: “AI is like teenage sex. Everyone talks about it. Nobody knows how to do it. But everyone thinks everyone else is doing it, so they claim to do it too.” There is certainly some truth in this. But Part III provides us with ample examples of applied AI in several business areas including in Finance and Accounting, Legal and Compliance, Business Intelligence and Analytics, Records Maintenance, Human Resources, Software Development, Marketing, Sales, and Customer support.
'Applied Artificial Intelligence': A Valuable Book for Digital Professionals
"Applied Artificial Intelligence" provides a valuable explanation of AI and the impact on organizations and our lives in general. But the specific value for digital professionals is the book talks a lot about AI from our perspective. For example, the value of data hygiene, governance, breaking data silos, data quality, digital maturity. All topics I talk about as a consultant daily. This book really helped me to explain AI and the information part of it to my clients.
My main criticism of the book is the structure: I think it could have used some additional work. Dividing the books into three parts was a good idea. But the authors regularly repeat themselves across the book. Part II — 'Strategy' — is particularly chaotic in its structure. The authors clearly had a lot of information to share and I sense their struggle to fit everything in. In my opinion, the many ROI explanations are the weakest part of the book, both in structure and content.
More Information on 'Applied AI'
You can find more information on the book and additional resources on AI here. The 112 end notes also give the reader a lot of additional information and URLs to web sources.