REBOOT: Getting to AI We Can Trust

As Marcus and Davis put it, the questions that people are focusing on now aren’t necessarily the right ones:

  [Many] agonize about dangers from a kind of AI that thus far is imaginary... grappl[ing] with the prospect of superintelligence taking over the world, despite the fact that current-day robots struggle to turn doorknobs. ... It’s as if people in the fourteenth century were worrying about traffic accidents, when good hygiene might have been a whole lot more helpful.

What we have for now are basically digital idiot savants: software that can, for example, read bank checks or tag photos or play board games at world champion levels, but does little else. Paraphrasing a line from investor Peter Thiel about wanting flying cars and instead getting 140 characters, we wanted Rosie the Robot, ready at moment’s notice to change our kids’ diapers and whip up dinner, and instead we got Roomba, a hockey puck-shaped vacuum cleaner with wheels.

What’s missing from AI today — and likely to stay missing, until and unless the field takes a fresh approach— is broad (or “general”) intelligence. AI needs to be able to deal not only with specific situations for which there is an enormous amount of cheaply-obtained relevant data, but also problems that are novel, and variations that have not been seen before.

As long as the dominant approach is focused on bigger and bigger sets of data the field may be stuck playing Whack-a-Mole indefinitely, finding short term data patches for particular problems without ever really addressing the underlying flaws that make these problems so common.

To move forward, they argue, the field will need to refocus its goals:

A short-term obsession with the narrow AI and easily-achievable “low-hanging fruit” of Big Data has distracted too much attention away from a longer-term and much more challenging problem that AI needs to solve if it is to progress: the problem of how to endow machines with a deeper understanding of the world. 

Without that deeper understanding, we will never get to truly trustworthy AI.

Why do we need deeper understanding, and what would that look like? In Montreal, Dr. Marcus will outline three key problems the field of AI must address if we are to get to machines that we can genuinely trust, and include ample time for discussion. 

We are thrilled to host this important reflection on progress in AI so far and where it needs to go next.

ABOUT GARY MARCUS

Gary Marcus, scientist, bestselling author, and entrepreneur was CEO and Founder of the machine learning startup Geometric Intelligence, recently acquired by Uber, is known for his provocative and bold claims in artificial intelligence, neuroscience, and cognitive science. Trained by Steven Pinker, he received his PhD at MIT at age 23.

His professional research, published in leading journals such as Science, Nature, Cognition, and Artificial Intelligence, has focused on the foundations of cognition in humans, animals and machines, 

spanning fields from developmental psychology to neuroscience to genetics to artificial intelligence. A special interest has been on the challenge of endowing machines with common sense.

He is also well-known for his writing for for the general public, including frequent essays and op-eds for The New Yorker and The New York Times. His books include The Algebraic Mind, The Birth of the Mind, Kluge: The Haphazard Evolution of the Human Mind, The New York Times Bestseller, Guitar Zero and The Future of the Brain: Essays By The World's Leading Neuroscientists.  

In a 2012 essay for The New Yorker, he was perhaps the first person to publicly criticize deep learning, drawing on arguments he developed in his 2001 technical book The Algebraic Mind. More recently, in a 2018 arXiv article, Deep Learning: A Critical Appraisal, he asked whether deep learning might be “approaching a wall.” The challenges he laid out there were covered everywhere from The New York Times to Wired to The Financial Times and Wall Street Journal.  

In another provocative 2012 essay, Moral Machines, Marcus was the first to adapt “trolley problems” to driverless cars, anticipating much recent research on the ethics of AI. His next book, REBOOT: Getting to AI We Can Trust, to be published by Pantheon in Fall 2019, co-authored with Ernie Davis, calls for a fundamental rethinking of current approaches to artificial intelligence.