The Future Computed, book review: AI and society, through a Microsoft lens
Now is the time to make sure AI systems are fair, reliable, and safe, private and secure, inclusive, transparent, and accountable, says Microsoft. That's because they're the next step in how technology is changing society -- and we all need to decide what we want that to look like.
It's time to wake up and smell the AI coffee, say Microsoft head lawyer Brad Smith and the leader of Microsoft's AI research division Harry Shum, in their introduction to this book on the role of artificial intelligence in society.
They kick off with a look at how much daily life around the world has changed in the last 20 years, and speculating on how much it could change in the next two decades as we get proactive intelligent agents ready to smooth our path through the world.
This could involve buying presents on our behalf and booking dinner reservations for special occasions based on what our agents know about us and our nearest and dearest, or answering emails and creating task lists automatically by listening in on our meetings.
Smith and Shum envision a future that's more convenient, but still very recognisable: gadgets may monitor your vital signs and suggest a checkup when they show something disturbing, but you'll still have an appointment with a real doctor to talk about those measurements. However, it'll be a virtual appointment, your medicine will arrive by drone and your agent will remind you when you should take it. And why will co-workers still be emailing to ask for updates that your intelligent agent can give them by checking the project timeline, when their agent could just look that up for them? Probably because people don't change their habits that quickly.
If this seems like a less ambitious future than some, it might be because, for all the changes technology has made in our lives, much of it had already started to arrive 20 years ago. Email and the web were around in 1998 -- they just weren't ubiquitous. It's also because Microsoft is firmly of the belief that AI is there to support, augment and amplify human abilities, rather than replace them. The examples of what AI is already doing -- marking up CT and MRI scans as a tool to save oncologists time, analysing farm animal activity to spot an anthrax infection in time to contain the outbreak, AI tools in Windows and Office -- are all different kinds of assistants to, rather than replacements for, people.
The Future Computed puts AI in its historical context, starting with Turing and the 1956 Dartmouth College symposium, and pointing out that route planning and search engines, and post office systems that read hand-written addresses, have been using what we now call AI techniques for several years. It also gives non-experts a clear, if somewhat Microsoft-centric, explanation of how AI works and how well it already compares to humans.
The assessment of AI ability sometimes errs on the side of technological optimism: the "work [that] remains to be done to make these innovations applicable to everyday use" isn't just dealing with noisy environment and accents, it's understanding what words mean and what decisions to make based on those words. Or as the authors put it: "Today's AI cannot yet begin to compete with a child's ability to understand and interact with the world using senses such as touch, sight and smell. And AI systems have only the most rudimentary ability to understand human expression, tone, emotion and the subtleties of human interaction."
The coverage of bias in AI is short but salutary: algorithms are designed by humans and use data from the real world, so they can both perpetuate and amplify bias. There's also an issue with putting too much trust in the precise-sounding numbers that AI predictions include: just because 70 percent of people in a group with a 70 percent chance of defaulting on a loan will probably default doesn't make everyone in that group a bad credit risk. The call for testing vision systems in low-light conditions seems uncannily prescient after the self-driving Uber accident in Arizona too.
It's also refreshing to see a title about AI point out that technology has been changing jobs since the 1740s, that automation has been making significant changes to jobs, and that doesn't pretend AI and automation are the same thing -- just that they push us in the same direction. (Incidentally, if you've been wondering why people talk about Industry 4.0, there's a nice explanation of how that's counting the industrial revolution everyone knows about as the first, the rise of industrial processes at the end of the 19th century as the second, and putting computers into homes and businesses as the third.)
You'll learn as much about how previous technology developments changed society as you will about AI -- the way the rise of the motor car helped to cause the depression of the 1920s and 30s, for example, and also the modern consumer credit boom (because taking out financing to pay for a car became common and respectable).
Similarly, AI systems are going to have unexpected results on society that will be both positive and negative, and if we're going to make them "fair, reliable and safe, private and secure, inclusive, transparent and accountable" the way Microsoft calls for, then that means thinking about privacy, regulation, jobs, education and the economy. Will several companies independently using the same algorithms to set prices look like price fixing, even when it isn't?
This broad viewpoint makes the chapter on the impact of AI and automation on jobs, businesses and workers unusually radical for an American software company, diving as it does into politically sensitive areas. Microsoft accompanies ideas about globally distributed workforces with reminders that "countries [may] face nationalist pressures and businesses face more restrictive immigration laws". It also calls for governments and businesses to invest in "modernising the social safety net" to cope with changes to traditional working styles and the rise of on-demand workers in the gig economy. The emphasis is as much on how education and training need to change as it is on the impact on jobs.
You can't really blame Microsoft for lacing the book with details of the tools and products it sells, the philanthropic programs it runs, and the educational experiments with which it's involved. Sometimes these Microsoft-specific examples makes sense, but sometimes they simply remind you that this is a corporate pitch as well as a thought-provoking philosophical essay on how AI will affect society.
Don't pick up The Future Computed if you're looking for a comprehensive survey of what AI can and will achieve. But do read it to remind yourself how much preparation is required for the impact of AI.