back to article Today in bullsh*t AI PR: Computers learn to read as well as humans (no)

Researchers from Microsoft and Chinese cyber-souk Alibaba separately claimed this week that their artificially intelligent software is as good as, if not better than, humans at understanding the written word. Journos fell over themselves to breathlessly report that, for instance, "ROBOTS CAN NOW READ BETTER THAN HUMANS, …

Silver badge

What is the name of the southern border?

Would the machine have still returned that answer? It would have to understand language to know that "southern" is logistically equivalent to "to the south" in this context, even when re-ordered to proceed the noun. That's how our human brains work.

If we all talk like robots, should it not be easy for the robots to understand us?

0
0

Another , more promising, approach

There are other AI natural language processing approaches that are not based on machine learning type pattern matching. I am much taken with

http://www.cortical.io/

It's still arguable if the program understands anything. But the deeper we dig into "intelligence" the more that seems to be so for human, IMO.

0
0
Gold badge
Unhappy

There are other AI NLP approaches..not based on machine learning type pattern matching.

True.

Cortico is not one of them.

"Semantic folding" is very much based on chewing on a shed load of data to construct an N dimension vector and look for words that are "close" together in some sense. They are large binary sparse matrices, allowing much compression and the ability to run comparison using the sort of Boolean operators common in most MPU instruction sets, but they are absolutely in the "Throw a metric f**k load of text at it and something will come out" school. :-(

Once I saw that I immediately thought of the Binary Neural Net work in the facial recognition system developed by London University called WIZARD. It's also got features in common with speech recognition approaches using "Time Warping" to cope with words spoken at different speeds (something else most humans can cope with up to a point).

9 pages into Cortico's 59 page White Paper and my BS meter is redlining like a Geiger counter in the engine compartment of a Cold War era Soviet nuclear submarine. Lots of verbiage, little insight. Why does the word "Autonomy" keep flitting through my mind as I keep reading?

I smell a BFR.*

*Big F**king Rat.

0
0

Let's face it,

we've all done a helldesk session or two, we've all met those users. Whatever the merit of current "AI", it can't be worse at understanding and executing a lavishly illustrated step-by-step guide on how to set up out-of-office replies.

2
0
Silver badge
Pint

Human Brain versus A.I.

The Human Brain contains quite a few "hardware co-processors"; it's certainly not 100% Neural Nets for higher order "learning". Any Biological Brain Boffin could probably give you a two-page list of the ones that they've found so far.

Most A.I. researchers don't seem to have hoisted aboard this basic factoid. Which is why they're always about 40 years further behind than they naively believe.

0
0
Bronze badge
FAIL

It would get interesting

if AI could suggest a sensible answer to the question Do you beat your wife only on Friday?.

2
0
Gold badge
Unhappy

"Do you beat your wife only on Friday?."

Indeed.

Actually as that's a query the easy answer is to check the listeners internal model and answer it.

That is an NLP problem, and illustrates that the questions "What does a sentence mean?" reduces to "Queries or updates on the listeners internal model of whatever it's about," even if the listeners internal model is "WTF are you babbling about." No model, no capacity to make a model, means the question (or any sentence) would be meaning-less.*

Recognizing the question is actually a very tricky attack on your personal behaviour is rather more of a general AI question and (I'd suggest) much harder.

*This should be pretty much SOP for NLP, but I've yet to find someone actually state it directly. I'd love a reference. I'm guessing Winograd did so, somewhere, but I've not had time to hit his thesis.

0
0

Page:

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Forums

Biting the hand that feeds IT © 1998–2018