Reid Blackman - Ethical Machines: Your Concise Guide to Totally Unbiased, Transparent, and Respectful AI

If you told me you were going to a conference on AI ethics, and that at a cocktail reception, you walked up to a group of attendees already discussing the topic, I could confidently tell you what was about to happen.

First, youre going to hear a lot of familiar phrases and buzzwords. Youll hear Accountability. Transparency. Explainability. Fairness. Surveillance. Governance. Trustworthy. Responsible. Stakeholders. Framework. Someones going to say black box at some point.

Then, therell be hand-wringing over the threats posed by AI gone wild. Biased data sets! Unexplainable algorithms! Invasions of privacy! Self-driving cars killing people!

Finally, the group will land on a healthy skepticism. You cant really define AI ethics, you might hear. And You cant plan for everything. Of course, also, Its just peoples own personal view of whats right and wrong. Or, more practically, Look, how do you even operationalize ethical principles? One person is going to say something about KPIs.

At the end of all this, there will be shrugs. Thats because, for the most part, people are raising issues, the underpinnings of which are not (well) understood, and then, after theyve put together a conference in which they sling buzzwords back and forth at each other, they declare AI ethics essential, of course, but also really, really difficult.

But not you. Having gotten to this point in the book, you can see those underpinnings. You see whats at issue, from both a business and an ethical perspective. And now that you understand it, you understand that AI ethics isnt so difficult after all.

If a colleague tells you, We need to say something about AI ethics, you know what a meaningful document looks like and whats superficial PR.

If your colleague tells you, This AI ethics is for the AI folks. Its technical. Tell them to get on it, you know how impoverished an understanding that is.

Maybe a company comes to you and says, We have your solution for responsible AI or, less ambitiously, We have your solution for bias in AI. That piece of software, you now see, cant solve all those problems by itself.

You know that and so much more about, for example, the role of people in creating appropriate metrics for fairness. And about when you need to know what the AIs are doing and when it doesnt matter so much. You know there are levels of privacy, and its not just about anonymity. You know how to build real AI ethics statements, and that just saying you value AI ethics doesnt mean your employees will take AI ethics seriously. You know you need Structure to make that happen. Most of all, you know software alone can neither handle the substantive ethical issues nor effect the kind of organizational change you need to systematically and comprehensively identify and mitigate AIs ethical risks. In short, provided the software can do its job, you see the ecosystem in which the software needs to get embedded.

You see the AI ethics landscape. Now, you can navigate it.

And now that youre here, Im going to let you in on a secret. Two, actually.

The first secret is that theres another book in this book. Pretend are a single book and delete the word AI whenever I said AI ethics. What youll get, with some minor tweaks here and there, is a book on how to articulate and operationalize the ethical values of your organization. I dont care if youre developing AI, putting microchips in peoples hands, or just selling bottled coffee: the way to create, scale, and maintain an ethically sound organization is already contained in those pages. If your aim is not only to create ethically sound AI, but to create an organization that takes ethical standards seriously, go back and reread those chapters with that in mind.

The second secret is that this book is about AI ethics, but its not only about AI ethics. Its about the value of ethical inquiry. Its about the power and importance of philosophical investigation.