Kai-Fu Lee talks about AI, jobs, and the human heart | Hacker News Reader

Kai-Fu Lee talks about AI, jobs, and the human heart

spectrum.ieee.org

51 pointsNicoJuicy14 comments

"AI's", even in their specialized domain of pattern recognition, still can't distinguish a corgi from a loaf of bread or specially designed white noise. The jobs he listed that are "in danger": truck driving, dish washing, fruit picking both require serious advances in robotics and do not require advances in general AI.

> require serious advances in robotics and do not require advances in general AI

Which if anything makes the replacement of workers in those domains even more tractable in the near future, wouldn't you say?

Dishwashing is an interesting case, because, assuming non-standardized dishes, glasses, and utensils, it requires solving a very general object recognition problem for a robot arm to handle different sizes and shapes of dishes.

Andrew Ng's lab at Stanford was working on it some years ago, but I'm not sure if they solved it more generally:

http://www.andrewng.org/portfolio/stair-stanford-artificial-...

https://www.youtube.com/watch?v=SbipWOeqF1c

Dishwashing can surely be done/automated now? A restaurant standardises a range of dish dimensions, these are mechanically sorted on a 'line' then cleaned according to their specification and surface scanned for error/reclean/escalation. No AI needed, decades old tech suffices.

In large-enough scale food serving operations (i.e. hotels) I imagine it is already automated to a large extent.

It's the domestic dishwashing use case that Andrew Ng's robot was trying to solve.

For smaller restaurants, the capital cost of standardized automation may be significantly higher than the marginal cost of low-wage dishwashing labor, who also can accomplish other tasks around the restaurant, like bussing, prepping, and cleaning, and therefore might not be worth it.

> No AI needed, decades old tech suffices.

This can be said of quite a lot of startup companies these days.

Compare the Zume robot shown in this video https://www.bloomberg.com/news/articles/2018-08-07/softbank-... to any number of videos showing factories making pizza such as https://www.youtube.com/watch?v=jdWftLcvmGM

And more importantly, if Andrew ng didn't figure it out then surely someone else will. The march of progress does not stop. It continues relentlessly.

This is the mindset I see many fail at (less so on here since people here are smarter and more visionary). People see robotics and AI for what it is today and not what it will be in the future. The robotics and AI of today have their issues but in due time those issues will be solved to create something truly impressive.

Truck driving, dish washing, order fullfilment and the myriad of other routine tasks will eventually be fully automated. Humans will be left to continue the march of progress as we should be doing.

Sounds more like a principle of faith than an empirical observation

> I believe with perfect faith in the coming of the Messiah, and, though he tarry, I will wait daily for his coming.

"can't distinguish a corgi from a loaf of bread or specially designed white noise" You are talking about adversarial examples, which are in most cases extremely rare and not detrimental to the typical function of the image recognition system. Even humans have problems identifying some of these adversarial examples. The white noise is particularly troublesome from a technical standpoint. It goes to show that these learned manifolds are not perceptual manifolds.

A road fatality is also an extremely rare event, once in a hundred lifetimes[1].

AI has to deal with extremely rare events to be better than human drivers.

[1] average miles driven per year - 12,000, while a fatality happens every 86 million miles on average.

I always read anything by Kai-Fu Lee out of general interest and nostalgia: many decades ago I was perusing AI books in the San Diego Technical Bookstore and came across his Carnegie Mellon PhD thesis on the Sphinx speech recognition system. I was mostly working as a FORTRAN scientific programmer at that time and that ‘book’ along with the book ‘Mind Inside of Matter’ by Bertram Raphael pivoted my career and life.

In the IEEE article, I agree with his predictions - all very reasonable.

“at the end they had a product [wechat] that was the most innovative social network. It’s so good that Facebook is now copying them.”

Can anybody substantiate some concrete examples of this reverse copying by Facebook?

Has anyone read his book?

I started it and only got part-way through before putting it down. I found nothing insightful in it. It was thought-leader drivel surrounding a core of self-promotion.

Agreed. Here is an excerpt from the WSJ. I couldn't finish even the excerpt!

https://www.wsj.com/articles/the-human-promise-of-the-ai-rev...

Reply on news.ycombinator.com