For new readers and those who request to be“好友 good friends” please read my 公告栏 first
The sunday New York times this morning has a very worthreading article on artificial intelligence (AI). Since NYT is blocked in China, I reproduce this article  below which contains nothing politically controversialor offensive to Chinese government. My previous post of this same article somehow became frozen and un-readable. Attempts to repair or delete it was also not successful. Hence I am reposting it here. Don't waste time trying to read it.

Artificial Intelligence Is Stuck. Here’s How to Move It Forward.

Gray Matter

By GARY MARCUS JULY 29, 2017

Photo Credit JunCen

Artificial Intelligence is colossally hypedthese days, but the dirty little secret is that it still has a long, long wayto go. Sure, A.I. systems have mastered an array of games, from chess and Go to“Jeopardy” and poker, but the technology continues to struggle in the realworld. Robots fall over while opening doors, prototype driverless carsfrequently need human intervention, and nobody has yet designed a machine thatcan read reliably at the level of a sixth grader, let alone a college student.Computers that can educate themselves — a mark of true intelligence — remain adream.

Even the trendy technique of “deep learning,”which uses artificial neural networks to discern complex statisticalcorrelations in huge amounts of data, often comes up short. Some of the bestimage-recognition systems, for example, can successfully distinguish dogbreeds, yet remain capable of major blunders, like mistaking a simple patternof yellow and black stripes for a school bus. Such systems can neithercomprehend what is going on in complex visual scenes (“Who is chasing whom andwhy?”) nor follow simple instructions (“Read this story and summarize what itmeans”).

Although the field of A.I. is exploding withmicrodiscoveries, progress toward the robustness and flexibility of humancognition remains elusive. Not long ago, for example, while sitting with me ina cafe, my 3-year-old daughter spontaneously realized that she could climb outof her chair in a new way: backward, by sliding through the gap between theback and the seat of the chair. My daughter had never seen anyone elsedisembark in quite this way; she invented it on her own — and without thebenefit of trial and error, or the need for terabytes of labeled data.

Presumably, my daughter relied on an implicittheory of how her body moves, along with an implicit theory of physics — howone complex object travels through the aperture of another. I challenge anyrobot to do the same. A.I. systems tend to be passive vessels, dredging throughdata in search of statistical correlations; humans are active engines fordiscovering how things work.

To get computers to think like humans, we need anew A.I. paradigm, one that places “top down” and “bottom up” knowledge onequal footing. Bottom-up knowledge is the kind of raw information we getdirectly from our senses, like patterns of light falling on our retina.Top-down knowledge comprises cognitive models of the world and how it works.

Deep learning is very good at bottom-upknowledge, like discerning which patterns of pixels correspond to goldenretrievers as opposed to Labradors. But it is no use when it comes to top-downknowledge. If my daughter sees her reflection in a bowl of water, she knows theimage is illusory; she knows she is not actually in the bowl. To adeep-learning system, though, there is no difference between the reflection andthe real thing, because the system lacks a theory of the world and how itworks. Integrating that sort of knowledge of the world may be the next greathurdle in A.I., a prerequisite to grander projects like using A.I. to advancemedicine and scientific understanding.

I fear, however, that neither of our two currentapproaches to funding A.I. research — small research labs in the academy andsignificantly larger labs in private industry — is poised to succeed. I saythis as someone who has experience with both models, having worked on A.I. bothas an academic researcher and as the founder of a start-up company, GeometricIntelligence, which was recently acquired by Uber.

Academic labs are too small. Take thedevelopment of automated machine reading, which is a key to building any trulyintelligent system. Too many separate components are needed for any one lab totackle the problem. A full solution will incorporate advances in naturallanguage processing (e.g., parsing sentences into words and phrases), knowledgerepresentation (e.g., integrating the content of sentences with other sourcesof knowledge) and inference (reconstructing what is implied but not written).Each of those problems represents a lifetime of work for any single universitylab.

Corporate labs like those of Google and Facebookhave the resources to tackle big questions, but in a world of quarterly reportsand bottom lines, they tend to concentrate on narrow problems like optimizingadvertisement placement or automatically screening videos for offensivecontent. There is nothing wrong with such research, but it is unlikely to leadto major breakthroughs. Even Google Translate, which pulls off the neat trickof approximating translations by statistically associating sentences across languages,doesn’t understand a word of what it is translating.

I look with envy at my peers in high-energyphysics, and in particular at CERN, the European Organization for NuclearResearch, a huge, international collaboration, with thousands of scientists andbillions of dollars of funding. They pursue ambitious, tightly defined projects(like using the Large Hadron Collider to discover the Higgs boson) and sharetheir results with the world, rather than restricting them to a single countryor corporation. Even the largest “open” efforts at A.I., like OpenAI, which hasabout 50 staff members and is sponsored in part by Elon Musk, is tiny bycomparison.

An international A.I. mission focused onteaching machines to read could genuinely change the world for the better — themore so if it made A.I. a public good, rather than the property of a privilegedfew.

Gary Marcus is a professor of psychology and neural science atNew York University.

Follow The NewYork Times Opinion section on Facebook and Twitter (@NYTopinion), and

For those of you who can jump the firewall, here is thedirect URL for the Times article.https://www.nytimes.com/2017/07/29/opinion/sunday/artificial-intelligence-is-stuck-heres-how-to-move-it-forward.html