Videos are back, after a couple of days of production bugs. This one is on algorithms. You know, the things that deliver entirely reliable forecasts of human behaviour. Or, maybe not.
Thanks for reading this post.
You can share this post on social media of your choice by clicking these icons:
You can subscribe to this blog's daily email here.
And if you would like to support this blog you can, here:
The film “Sully” really brings this issue to the fore….a tale of man versus machine . In the showdown scene ofbthe film, as the accident enquiry reaches its finale Sully proves the algorithms were wrong as they took no account of the human factor. Great film and a true story.
Agreed
You know that Sully was a film, right?
In real life, the algorithms proved that the flight crew made exactly the right choice and the NTSB had nothing but praise for them….
So not sure your point makes any sense at all – other than to say that it is often the prejudices we humans have that cause the problems, not the algos.
Of course I know Scully was a film
One he did not disown
And you miss all the other points
Why is that?
I’m afraid that neither of you followed the film plot if that’s what you think.
The flight simulation ran perfectly. The point was that in the initial runnings of the simulation the pilots (i) knew what was about the happen and (ii) knew that both engines could not be re-started and than a glide diversion was immediately necessary. As Sully pointed out, that was wholly unrealistic. It was the human input during the flight simulations that was at fault, not the algorithms of the flight simulators, which accurately predicted that once the inevitable pilot decision time was allowed for, a crash was inevitable if a diversion to alternate airports was attempted.
The algorithms were right, the humans operating it were wrong.
The lesson to be learned from this is that if you don’t understand what you are watching, you will come away with the wrong conclusion.
The algorithms were wrong
They assumed human behaviour that could not occur in real life
Please don’t make absurd observations unless you want to prove my point
And yes, I did understand the plot
It appears you did not
Semms you understand very little in this case.
In REAL LIFE – not the dramatised film version, the NTSB weren’t out to get Sullenberger and no blame was placed on the flight crew at all. So the algorithms you are talking about which “got it wrong” are just a product of the film itself. And even in the film the reason the algorithms got it “wrong” were because they assumed unrealistic parameters.
The real world simulated tests, both by human pilots AND by algorithms, neither were able to safely land the place at an airport in real world conditions – which were simulated. The human and algorithm tests both showed exactly the same thing. I quote from the real world NTSB report:
“Simulation flights were run to determine whether the accident flight could have landed
successfully at LGA or TEB following the bird strike. The simulations demonstrated that, to
accomplish a successful flight to either airport, the airplane would have to have been turned
toward the airport immediately after the bird strike. The immediate turn did not reflect or account
for real-world considerations, such as the time delay required to recognize the extent of the
engine thrust loss and decide on a course of action. ”
https://www.ntsb.gov/investigations/AccidentReports/Reports/AAR1003.pdf
But if you think that a film is good evidence for your argument, please, go ahead and make a complete fool of yourself.
Ok
But you prove yourself to be a complete troll in the process
Is such aggression your normal modus operandi?
I hope not for your sake
It also makes me doubt you, considerably
We must avoid the trap of “Algo Bad, Human Good”.
Algorithms are fantastic when employed appropriately – when I use a pocket calculator to do some sums it uses an algorithm to deliver its answer…. and I trust it completely. I do not trust my own arithmetic to the same extent.
Having said that, I agree, we must avoid the more dangerous trap (the one you allude to) of taking a tool that is successful in one field (doing sums) and assuming that it can deliver answers about human behaviour…. the evidence is that algos can’t do that very well.
However, I fear that if we “algo bash” too aggressively there are problems. The analysis of drug trial data uses algorithms that delivers robust conclusions. I would not want humans to decide effectiveness on a “hunch” (even if we still need those “hunches” to suggest lines of enquiry).
I agree
Of course I use algorithms
The point is that they must be used intelligently
On one of the rare occasions when my father might have been right on any such issue, his lament for the slide rule was appropriate. When using it, he suggested, you had to have some idea what the answer might be to make sure you got the scale right
When using algorithms too many people have blind faith
Not only judgement about the future but about our history.
I had a neighbour on my street who was a Professor of Financial mathematics.
He had convinced himself that his models were true and that derivatives could reduce risk.
I argued that the long term rate of return on capital was about 2 per cent and anybody that told you otherwise was probably inducing you to take risks with your money.
All his models failed when “normal” credit worthiness assessments went out of the window; when the banks were given an effective state guarantee; and credit risks could be effectively palmed off onto others.
Algorithms are only as good as the assumptions they use.
They are not reality.
that is not to say that advances in science cannot take place but anything that is human, social and riddled with uncertainty needs to be approached with great caution.
Agreed
I’d highly recommend Cathy O’Neill’s Weapons of Maths Destruction for an intelligent analysis of the strengths and weaknesses of algorithms. They have been around for longer than most people realise.
Ultimately they are a way of encoding human logic, decisions and choices. As we saw with A levels, people try to hide behind the argument that it was the computer that made the choice but with algorithms, they reflect personal (and political) choices.
AI is a whole different ball game where potentially the decisions are hard to explain, though they may be for good or ill. With both algorithms and AI, what we need is transparency and auditability. Cue argument about who should do those audits. Quis custodiet…
Hi Richard,
Off topic and not of general interest, but what you have to say is important and interesting and you may not realise that the Moire patterning from that shirt fabric you are wearing can be quite distracting, especially when you change stance slightly. Exactly how obvious it is probably depends on the viewing device and zoom level, but at my default settings it isn’t something that can be ignored. IIRC this cropped up in the early days of colour TV: with digital there are, allegedly, algorithms to mitigate the effect but I’d guess the [sartorial] solution of wearing something plain would be more reliably effective.
Best regards
Odd
We tried this, and concluded it was not an issue
I actually got some plain shirts but really hate them…
I discussed with Mark, but he is overwhelmed with other issues right now
We will need to work through what is recireded….
Sorry
So if algorithms are usually wrong, what does that say about the whole climate change industry, which relies on lots of different algorithms to tell us what temperatures are and make predictions about global warming?
If the algos are basically useless, why should trust them and hose who tell us we should totally change our economies based on them?
Did I say they were useless?
Or did I say we have to use them with care? Especially when modelling human behaviour, to which you are not referring
It’s continually tedious to find deliberate false extrapolation in the comments of those who appear here for the first time