AI Book Club: Moral AI And How We Get There by Jana Schaich Borg, Walter Sinnott-Armstrong & Vincent Conitzer
Moral AI And How We Get There is a hefty read, but beautifully written, and one of the most thought provoking books on AI I’ve read so far. It is the brain child of a philosopher, data scientist and computer scientist, as they consider the ethics of using AI to help surgeons pick the best candidate for organ transplants. Recommend for anyone still unpacking the ethics behind AI use in the modern world, especially from the perspectives of ethics, the law, business and philosophy.
But authors Jana Schaich Borg, Walter Sinnott-Armstrong & Vincent Conitzer left me with more questions than answers. Questions like:
What is AI? (Surprisingly, there are so many different ideas and thoughts here.)
How does AI interact with the basic moral values of safety (e.g. self driving cars, medical diagnoses), equality (e.g. not adopting gender or race bias that’s present in training data), privacy (e.g. can sensitive training data accidentally get leaked), freedom (e.g. can a government track and stop people from moving around, but also AI to help the blind navigate the world freely), transparency (e.g. does the AI have to explain how it came up with something), and deception (e.g. deepfakes or AI hallucinations).
Can AI respect privacy? (I learnt some very surprising and scary hacks around reverse engineering private information from data.)
Who is responsible when things go wrong, such as a self-driving car running someone over? What are the differences between moral obligations, legal responsibilities
How do we impart human morality onto AI?
You’ll notice that these are all questions, because frankly this book just raised more questions than answers in my head. Luckily, the authors don’t leave us intellectually stranded. In their final chapter, they wrap things up with five practical ways to turn good ethical intentions into real action:
Scale moral AI technical tools, so they’re easy check points baked into the AI development and deployment at scale.
Disseminate practices that empower moral AI practices, which is easier said than done when businesses are generally designed around stakeholder/shareholder management and profits. Other issues mentioned include how agile-lean methodologies might not easily fit AI ethics into their workflow, and figuring out who is responsible for AI ethics.
Provide career-long training opportunities in moral systems thinking, similar to how doctors and lawyers need to re-train at regular intervals to keep up to speed with changing industries.
Engage civic participation throughout the life cycle, although in an earlier chapter they discuss the balance of bringing in laypeople but also recognising that those with less experience or knowledge may have skewed context.
Deploy agile public policy, because smaller businesses without the budget to payroll an ethics or AI officer would find government best practice guidelines and policies helpful to not just survive but also thrive. Because strong AI use is good for economies.
Grab a copy of Moral Ai via Penguin Australia.
