A book for those worried about AI
The robots are coming! The robots are coming!
It is hard to avoid stories about the potential of artificial intelligence to change the way we live today.
According to a Pew survey, 72% of us are “worried” rather than “enthusiastic” about the potential developments in automation.
Count me among the worried, though I am not worried about artificial intelligence coming for my job so much as the unthinking embrace of technology as a “solution” for flawed human behavior. It seems as though there’s a story every day about the failure of a so-called intelligent algorithm, including recent news of the Amazon facial recognition technology, “Rekognition,” “matching” 28 pictures of New Englandarea pro athletes with a database of police mugshots.
Oops.
Except sometimes it’s more than an “oops.” Denmark is currently reviewing more than 10,000 court cases that may have been wrongly decided because of software bugs in cellphone-tracking technology.
To understand how worried I should truly be about artificial intelligence and the robot invasion, I recently dug into “Artificial Intelligence: A Guide for Thinking Humans” by Melanie Mitchell (Farrar, Straus and Giroux, $28).
Mitchell holds a Ph.D. in computer science and is an artificial-intelligence researcher who has worked alongside some of the pioneers of the field, including Douglas Hofstadter, author of the seminal “Gödel, Escher, Bach: An Eternal Golden Braid.” That book established many of the ground rules for how we discuss issues of computer intelligence and automation.
In short, Mitchell knows what she’s talking about. Even better, she’s a clear, cogent and interesting writer, which is vital, because even with the significant effort Mitchell puts in to translate the technology to a layperson, full understanding (at least for me) required the occasional timeout for contemplation or re-reading.
What we learn is that artificial intelligence is far too broad a term to be truly useful. We’re talking about many different technologies under one label, and Mitchell helpfully walks us through the development and current status of those technologies.
Her early chapter on teaching algorithms to do something as seemingly simple as identifying individual numerals illustrates a consistent theme: Researchers have made huge progress when it comes to machine learning, and yet we’re still far from achieving anything like Ray Kurzweil’s “singularity,” where technology achieves a kind of selfawareness that upends human civilization.
The Terminator franchise is an illustration of this potential dark side, where Skynet — originally designed to protect people — realizes that humankind is a threat to its own existence and therefore must be eliminated.
Mitchell is not particularly worried about singularity coming true, and as a reader, I trust her expertise. If anything, AI seems to have been oversold. The book makes a case that we’re much further from self-driving cars than the popular hype would have us believe.
While “Artificial Intelligence: A Thinking Guide for Humans” has significantly improved my knowledge when it comes to automation technology, the greater benefit is that it has also enhanced my appreciation for the complexity and ineffability of human cognition.
Algorithms can now compose music that we may call “art,” but as Mitchell points out, those algorithms will always be incapable of appreciating that art. The things we may see as human flaws or foibles are actually an important aspect of what makes us intelligent in ways computers and algorithms can’t match.
Every week, this column is a kind of challenge to Amazon’s “customers have also read” algorithm. Often, I do not even know why I think a particular book is a good match for a reader, even as I know deep down that it is a best choice — if only because it involved a human exchange artificial intelligence will never match.