On algorithms and their glitches
This is starting to sound like a Philip K. Dick novel:
“People say, ‘Well, what about Facebook – they create and use algorithms and they can change them.’ But that’s not how it works. They set the algorithms off and they learn and change and run themselves. Facebook intervene in their running periodically, but they really don’t control them. And particular programs don’t just run on their own, they call on libraries, deep operating systems and so on …”…
Corporations like Facebook and Google have sold and defended their algorithms on the promise of objectivity, an ability to weigh a set of conditions with mathematical detachment and absence of fuzzy emotion. No wonder such algorithmic decision-making has spread to the granting of loans/ bail/benefits/college places/job interviews and almost anything requiring choice.
We no longer accept the sales pitch for this type of algorithm so meekly. In her 2016 book Weapons of Math Destruction, Cathy O’Neil, a former math prodigy who left Wall Street to teach and write and run the excellent mathbabe blog, demonstrated beyond question that, far from eradicating human biases, algorithms could magnify and entrench them. After all, software is written by overwhelmingly affluent white and Asian men – and it will inevitably reflect their assumptions (Google “racist soap dispenser” to see how this plays out in even mundane real-world situations).
Hey, that’s from the Guardian, so what do you expect? Leftists assume that those “affluent white and Asian men” will of course introduce assumptions that will hurt the other, more righteous races and genders (ignoring the fact that so many people in the computer industry are on the political left, no matter what their race and gender). The central point, however, is a good one, which is that algorithms are not inherently fair and not inherently and always better than what actual human individuals would do.
This is not unfamiliar territory. The machine running amok and creating its own trajectory is a staple of science fiction. That’s where Philip K. Dick comes in (see also this).
More than one hundred years E M Forster, hardly known for science fiction, published a remarkably prescient and completely fascinating novella about the dangers of future technological developments barely dreamt of at the time. It is entitled The Machine Stops, and I would heartily recommend it to anyone interested either in literature or in the critical analysis of technology’s effect on human society.
Anyone who has paid attention to Google and their search engine, knows how it has been manipulated by the left with mass searches to change the ranking of results. I haven’t used Google for a while. Wikipedia has a similar problem but in that case, as at Amazon reviews, it is low paid leftist employees.
Just a sidebar, reading through the linkage, but isn’t this a far cry from today’s polarized political world?
https://www.thenewneo.com/2013/06/21/philip-k-dick-and-me/#comment-619172
IGotBupkis, “Faeces Evenio”, Mr. Holder? on June 23, 2013 at 4:48 am at 4:48 am said:
I did know of this — Robert Heinlein thought very well of him. as Dick describes it: [unfortunately there is no citation, but it sounds very like RAH]
“Several years ago, when I was ill, Heinlein offered his help, anything he could do, and we had never met; he would phone me to cheer me up and see how I was doing. He wanted to buy me an electric typewriter, God bless him–one of the few true gentlemen in this world. I don’t agree with any ideas he puts forth in his writing, but that is neither here nor there. One time when I owed the IRS a lot of money and couldn’t raise it, Heinlein loaned the money to me. I think a great deal of him and his wife; I dedicated a book to them in appreciation. Robert Heinlein is a fine-looking man, very impressive and very military in stance; you can tell he has a military background, even to the haircut. He knows I’m a flipped-out freak and still he helped me and my wife when we were in trouble. That is the best in humanity, there; that is who and what I love.”
They are not so far apart as Dick might think – just in the outward details, now where it counts:
https://www.americanthinker.com/2013/05/philip_k_dick_and_our_predicament.html
“This is what Dick has to offer — something beyond mere politics; a glimpse at what makes us human. The moral law within, the ability to tell good from evil without actually being able to define them. “
It’s well to remember what many of those algorithms were constructed to do.
http://www.market-ticker.org/akcs-www?post=233929
“When Did You Consent?
[NSFW language warning]
When did you consent to your bank “sharing” data on your account with Facebook?
Would you consent? Under what terms?
Are you out of your ****ing mind?
Facebook’s stock is up some 3.5% today on the “news” that it is working on “partnerships” with major banks, which will include access to your account data — including balances.
May I remind you that the major credit-card processing networks — same banks, ultimately folks — have over the last few years, prompted by the “chip” thing in cards, quietly been changing from “Level 1” (that is, amount-of-transaction only) data from merchants to Level 3, which captures the quantity, price and exact items or services purchased?
…
When were you consulted on any of this and given fair disclosure?
When was a sign posted at the checkout that every single line item in your cart was being sent, in detail, to both Visa and your bank and that they may do with that data as they please, including selling it to your health insurance firm, car insurance company and, quite-possibly, your employer?
Never.
So now you go into a restaurant and instead of the bank knowing you charged $45.02 on your card they now know you bought a $20 steak and $18 worth of beer — and the rest was tip.
They also now know if those beers were $6 each — you drank three — or if you got them on happy hour for $2 and you drank nine of them!
This is extremely valuable information; among other things if you ever buy alcohol far enough away from your home that you could not have reasonably walked you are instantly assumed to drive drunk by your car insurance company and pay another $200 a year — or more — for your car insurance.
Think I’m kidding?
Nope.
This is happening right now, today.
Now Facebook wants the same thing, indirectly, so they can exploit it and they’re going to get it.
When did you consent to any of this from either end?
Never.
…
Has any of this been curtailed or this data been restricted or deemed to belong to you, anywhere, or has any executive ever been indicted or prosecuted even if you get directly screwed by some “glitch”, such as Wells Fargo’s repeated and outrageous hosing of customers?
Nope, never.”
https://www.the-tls.co.uk/articles/public/ridiculously-complicated-algorithms/
Hard to excerpt from that one, but it’s much like Neo said only in more detail.
And there is that question of just what, exactly, these self-teaching algorithms are learning?
https://reason.com/blog/2018/07/03/facebook-algorithm-flags-removes-declara
Facebook Algorithm Flags, Removes Declaration of Independence Text as Hate Speech
The social media site has a difficult time telling the difference between white nationalist ravings and the writing of Thomas Jefferson.
Christian Britschgi|Jul. 3, 2018 4:50 pm
> The social media site has a difficult time telling the difference between white nationalist ravings and the writing of Thomas Jefferson.
Yeah, there’s no political bias here.
I didn’t know that Neo was a PKD fan from a young age. I think I caught his bug with the films Blade Runner and Total Recall and then the Am. Thinker article. The film bus. has by now taken up many of his stories, including the recent anthology “Electric Dreams” over at Amazon.
______
Whew! I read the whole Guardian article, and while it wasn’t terrible, the author made no attempt to limit his scope. Self modifying code, neural networks, and genetic-like evolutionary model development. These are all very different things and pretty old. What scares me generally is the idea that trained neural nets, which are decidedly non-deterministic and not perfectly predictable, would be unleashed in areas where we really should have 100% predictability.
The autonomous car thing, which may or may not contain neural networks, also really bothers me. I took interest in the walking pedestrian with bicycle who was killed in AZ when it happened. I read a few articles and watched the onboard forward looking camera footage. I think I can say that if I was driving that vehicle, as a normal driver, in that situation, I wouldn’t have come close to striking the pedestrian.
The impact happened only 20 yds. or so from a street lamp. While the pedestrian was a little hard to see early in the camera footage, this was a function of the exceedingly bright low-beam (short range) head lamps and/or high camera contrast. Then we have this quote from the Guardian article,
“We now know that Herzberg, the pedestrian killed by an automated Uber car in Arizona, died because the algorithms wavered in correctly categorizing her.”
No. She died because some software engineer decided it was OK for the vehicle to strike unknown objects, in lieu of having false danger alerts resulting in emergency braking. Put the person in charge on trial I say. Maybe that will get their attention.
Not on point, but I think that it’s important enough news that people’s attention should be focused on it.
Judicial Watch sent a FOIA request to the DOJ asking for all transcripts of the Hearings that the FISA Court held on the issue of granting DOJ the Warrant to spy on Carter Page–an initial request and then three 3 month long renewals.
And we were told that the Court would require the DOJ to re-justify the renewals by citing new information to justify them.
Judicial Watch reports today that the reply of the DOJ is that, they could not supply any transcripts, because no hearings were ever held.
So, it appears that the DOJ just handed over the warrant request and the FISA Court–supposedly established to make sure that such extraordinary power to intrude on citizen’s liberty and privacy was very strictly controlled and monitored, just rubber stamped the request; took all the allegations in the warrant request as the complete and unimpeachable truth.
If this report by Judicial Watch and from the DOJ is true, then, FISA Court itself is a complete sham.
Supposedly knowledgeable commentators have said that these FISA Applications are usually comprised of many pages–I believe that the figure I heard was 30 pages or more–of information/evidence presented to justify issuing a FISA Warrant.
Has anyone seen and heard Carter Page on TV–Page a U.S. citizen, Naval Academy graduate and former Naval officer who, I note, has not to date actually been charged with anything, nor arrested?
I would think that anyone, once having observed very low key, laid-back Mr. Page, would find the charge that this guy was a real threat, a very savvy spy, the agent of a hostile power, to be pretty ridiculous and, thus, they would require a boatload of pretty unimpeachable evidence to convince them that Page was, indeed, the agent of a hostile foreign power.
We have been assured all along that–since the virtually global, 24/7 monitoring allowed by the FISA court is so intrusive–the FISA court applies the highest level of intense and searching judicial scrutiny to Warrant requests.
Could it be that the FISA court judges had not one question about any element of such a large mass of information that essentially accused Carter Page of treason to the United States, that the evidence against Page was so strong that they didn’t even need to conduct a Hearing to examine it, and that they just glanced at the paperwork, assumed it was all true, and immediately signed off on it, and then signed off, all over again, on each of three, three month extensions?
We know that one judge on the FISA court, Judge Rudolph Contreras, recused himself from the General Flynn case.
Which judge of the 11 judges on the FISA court signed off on the initial Warrant request and, then, on all the subsequent extension requests?
Was it just one judge, a couple of judges, or were all of them involved?
We don’t know.
That information has never been released.
Whose signatures are on the paperwork?
I still think the Flynn case will go the Ted Stevens route. Dismissed and prosecutors sanctioned.
To actually comment on the original point.
When the programs get so complex that a human–no matter how well educated and intelligent–can’t hold it in his mind, comprehend it, understand how it will work, what it’s capabilities truly are, and what it will, in the end, produce, we are already lost.
My impression is that we are already at the threshold–or perhaps beyond–of this happening.
Snow on Pine on August 31, 2018 at 5:15 pm at 5:15 pm said:
Not on point, but I think that it’s important enough news that people’s attention should be focused on it.
* * *
Whoa Nellie! as my gramma might have said.
Looking forward to the news cycle tomorrow.
More on algorithms: what they are, how they are “trained,” and why they lean left.
https://pjmedia.com/trending/algorithms-the-ultimate-excuse/
Algorithms are not dynamic. The paths they follow are. The paths follow the input with control[ler] biases.
Either The Guardian denies Catastrophic Anthropogenic Global Warming, or believe they can abort the baby and have her too. Both are possible, but only the latter is a known control[ler] bias, and therefore probable. In fact, given the control[ler] bias for the former, it is not only improbable, but rather an absurd outcome of the algorithm.
how they are “trained,” and why they lean left
Indoctrination (e.g. input, feedback) and contol[ler] (e.g. exclusion, inclusion, weighting) bias. The training regime is much like the press in this regard.
Not buying the narrative they “run themselves”…
Back to my comments above on the FISA court not holding any hearings to examine the evidence before OKing a FISA Warrant to spy on Carter Page.
Out today on the Gatewaypundit is a story by Jim Hoft, citing an apparently astute Twitter participant named ImperatorRex, who points out that there is a law (50 USC 1805) which would have allowed then Obama Administration AG Loretta Lynch to issue a FISA warrant on her own, if there was an “Emergency” justifying such an action.
The blogger then goes on to point to various statements and actions that have all already been reported–by Obama, the Obama Administration, Hillary, certain Strzok emails, etc.–that he highlights to show how the appearance of such an “Emergency” situation, to justify Lynch’s issuance of a FISA warrant, could have been manufactured.
Once her FISA Warrant was issued by Lynch, according to law she had to notify the FISA court within 7 days, and then, says ImperatorRex, it looks like the Chief FISA judge just signed off on that Warrant–thus no hearing necessary.
see https://www.thegatewaypundit.com/2018/09/report-there-was-no-fisc-hearing-for-carter-page-warrant-because-obama-admin-bypassed-court-using-story-they-knew-was-false/
Snow – thanks for the updates; I haven’t seen anything on PowerLine or my other “regular” spots, but Gateway is often ahead of the pack.
On Algorithms, ran into this amusing piece today.
https://outline.com/6ps979
WIRED ›
The Fasinatng… Fascinating History of Autocorrect
GIDEON LEWIS-KRAUS JULY 22, 2014
Actually, the history of computer nerds runs libertarian, not left-liberal. The COMPANIES run left, and the current gen, as with all of this one, runs left.
That, of course, requires people working in the industry from a variety of backgrounds.
Pretty sure the technical term for this is “a great big load of hooey”.
It’s called “very bad QA”. As a QA person, whenever I look at a web page, one thing I ALWAYS check is how the screen font and colors look. I always consider color blindness
I am not colorblind
It does not matter, I don’t need a black co-worker to test this. I only need a clue how todo my job.