Three more hints from the future

Today I’d like to continue my practice of sharing news stories that seem especially future-significant.  It’s part of horizon scanning.  (Check this article for an introduction.)

For each story I’ll include a quick description, some supporting links, and a brisk yet cautious discussion of why it might matter for the future.

ITEM: two neural networks outperformed humans on a reading comprehension quiz.  Two different pieces of software written by Alibaba and Microsoft, respectively, exceeded homo sapiens on the Stanford Question Answering Dataset.

The model developed by Alibaba’s Institute of Data Science of Technologies scored 82.44, edging past the 82.304 that rival humans achieved…Microsoft achieved a similar feat, scoring 82.650 on the same test, but those results were finalized a day after Alibaba’s, the company said. 

Why this might matter: this is one area where AIs can contribute value in the economy and to humans.  Think about the many ways reading comprehension occurs in the world beyond academia.  One possibility is, as Newsweek rather excitedly proclaims, he automation of many human-performed jobs.

It’s also a reminder that China is working hard to be an AI powerhouse.  So is Microsoft, but we don’t talk about them if we can help it.

Moreover, it’s another venue where machines may exceed humans.  At what point will we start reacting?  And how will we: with resignation, violence, the development of new capacities, a focus on what humans still do best?

Caveats: I’m not sure what value of “human” they tested against.  Was it some sense of median or average person’s reading comprehension?  And what did the machine output look like?  The latter might need work before exiting the lab.

ITEM: while international students have started avoiding American universities, they are heading to Canadian academia in greater numbers.  “[T]he percentage of international students is on the rise in both Canadian colleges and universities.”

International enrolments as a percentage of total enrolments, Canada, 2006-07 to 2015-16

“International enrolments as a percentage of total enrolments, Canada, 2006-07 to 2015-16”

Why this might matter: for one thing, attracting international students is one way to grow overall enrollment, especially in advanced (i.e., aging) nations, like Canada.  “In the university sector, roughly two thirds of all student growth since 2009/10 has been from international students”.

For another, it reminds us that certain nations, such as Canada, are competing in a global marketplace for those students willing to travel abroad for post-secondary education.  Put another way, Canada is successfully competing with the USA in this market.  It’s possible this is a trend that will continue for at least the next three years of the Trump administration, and possibly beyond.  Will American colleges and universities manage to overcome that political problem, or will they continue to fall further behind?

Caveats: none right now.

ITEM: the Russian military stated that two of its Syrian bases were attacked by drone swarms.  The attackers weren’t known, and didn’t accomplish much damage.  Yet it might just be “the first announced use of a swarm of drones in a military action”.

According to Asia Times,

It appears the mission of the swarming drones was three-fold: it was to show the Russians that their bases are vulnerable to attack even if the terrorists are far off (the attack was launched about 50 km away originating in Idlib according to reports and the Russians have now destroyed a stockpile of drones there); that the Russian aircraft and missiles were vulnerable to a drone strike; and finally that the bomblets could be used to terrorize ground crews and military personnel on the Russian bases.

One key detail: these were very, very cheap drones, DIY affairs, possibly bought for cheap on the black market:

The drones themselves are simple.  They use a small commercial gasoline two stroke engine that might be found in a weed whacker or used to power a bicycle. Structurally the drones are made out of wooden spars and styrofoam “boards” that are tied into the wooden structure with glue and plastic wrap.

The drone itself is launched from some sort of simple rail platform and guided by two piece[s] of wood on the drone with cutouts to protect the drone’s aerodynamic quality.  The drones carry either eight or ten bomblets, each stuffed with the explosive PETN (pentaerythritol tetranitrate)…

Another detail: the Russian army’s counter-drone electronic warfare team brought down some of the drones.  This is apparently now a thing.

One small detail: the Russian Ministry of Defense also issued a statement about this on Facebook.

Why this might matter: my readers might be thinking of the recent slaughterbot video, and see the Russian announcement as one step along the path to realizing that new stage in the history of human violence.

There are also multiple geopolitical aspects, which I don’t have time to get into here: Russia’s continued presence in Syria, the Ukrainian connection, blaming the US, etc.

Caveats: it’s not clear why the Russian military made this statement.  Are they actually spooked, and are airing the story to raise awareness?  Is this a hoax or maskirovka, aimed at some other goal, like Ukraine or the US?

Moreover, the “swarms” aren’t very large.  One attack featured ten drones, while another a mere three.

So, three stories from this week.  Three potential pointers to the future.

This entry was posted in horizon scanning. Bookmark the permalink.

3 Responses to Three more hints from the future

  1. Mike Sellers says:

    Re: the reading comprehension test. I dug a bit, and found the original article on the dataset and measures used here (“SQuAD: 100,000+ Questions for Machine Comprehension of Text” – https://aclweb.org/anthology/D16-1264). It’s a poorly written and cited article, which surprises me. They discuss but never define their scoring methodology, the “F1 Score” (see this paper for an actual definition — note this isn’t cited by the earlier paper, and this one actually makes a case against it: https://adamyedidia.files.wordpress.com/2014/11/f_score.pdf). The SQuAD paper says that the human score is “86.8% F1 based on interannotator agreement.” As this score (and the quiz) is based on a combination of precision and recall of Wikipedia articles, that’s presumably what “interannotator agreement” refers to. But, while this new score is potentially significant, this seems once again like an abstruse accomplishment being dressed up to mean something more than is justified — i.e., that “AI can now comprehend articles better than humans.”

    As an indicator for the future, it’s a good one — jobs that depend on sixth-grade reading comprehension are not in themselves safe from automation — but as always, liberal grains of salt must be applied.

    On the third item, drone swarms, this seems to me to be like the first time a primitive musket was used in combat; unremarkable in its immediate effect, but inescapable in terms of what it foreshadowed.

    • Bryan Alexander says:

      Drones: agreed.
      Offhand, I’m expecting a lot of creativity and development for this first stage, while drones are above a certain size – say, ready human visibility (down to one inch). Especially once 3d making accelerates.
      The next stage, when drones are smaller? Then things get *very* interesting.

      Reading comprehension: Good analysis. We’ll need to keep an eye out for followup research, either by the authors or others.

      Thank you for digging in, Mike!

  2. Bob Millern says:

    I remember building reading comprehension in high school with that darn color-coded card box. “Go back and do another purple!”. Those were nasty. I, for one, welcome AI’s to do my reading for me.

Leave a Reply

Your email address will not be published. Required fields are marked *