House of Lords push internet legend on greater openness and transparency from Google. Nope, says Vint Cerf The reverence in the House of Lords was palpable as Vint Cerf, a Google grandee and one of the, er, elders of the internet, was described during a committee meeting as technology's answer to Sir David Attenborough. However, that did not stop the British Parliament's second chamber asking some pressing questions regarding the …
There's no doubt that Vint is a bright guy and that he's scrupulously correct and polite face to face, but, like Rees-Mogg, his carefully-chosen mode of dress seems to signal a patrician view of the world. Privacy may be an anomaly, neural algorithms may be intrinsically flawed and discriminatory but the legislators should keep their noses out because it's all being done for our own good by selfless technocrats who are merely surfing a wave of inevitability.
>There's no doubt that Vint is a bright guy and that he's scrupulously correct and polite face to face, but, like Rees-Mogg...
He's not that bright.
It is clear the Vint who serves Alphabet,i is just a shadow of the younger Vint who helped lay the foundations of the Internet.
There is really no reason why Alphabet couldn't "publish information about algorithms and neural networks", there are plenty of people out-there fully capable of reading and understanding them, however, the real problem is that these would reveal the full extent to which the human hand is still necessary to manipulate the "complex interconnection of weights that take input in and pop something out to tell us you know what quality a particular web page is".
You may be really bright at designing network protocols, but you may understand little about other issues, especially when you are handsomely paid for not understanding it, or pretending not to understand it.
Apelles' words "Ne sutor ultra crepidam" - apply in this case.
there are plenty of people out-there fully capable of reading and understanding them
Understanding an algorithm developed by machine learning? Er, no. By design, they're a black box which you can only judge by comparing outputs and choosing the one that get better results... Which for Google, means more clicks, not any mushy stuff like quality.
By this point, Google themselves have no idea how the algorithm works and what it does. They only make sure not to give it access to weapons, just in case.
Not true. Certainly they're not black boxes by design, which implies that's an intention. It's mainly a result of the structure and training, but there has been work for quite a while now into ones that are interpretable by design. Additionally you can take any neural network and analyse it for sensitivity to different inputs.
indeed, that is how kidney function was first worked on. It was treated as a black box and input (blood plasma composition) and output (urine) were analysed and manipulated to gain insight on kidney function.
The first detailed look was a technical tour de force and was destroyed by human error. The NZ white rabbit strain keeps it's distal convoluted kindey tubules just below the surface instead of buried deep inside. A PhD student dissected one out, succeeded in connecting it to tubes so he could put whatever input he wanted in and meausure the output and do things like stick recording electrodes in.
He excitedly went to his supervisor who went to the microscope and promptly racked the objective lens through the preparation.
Telling human stories like that in Physiology lectures made them more memorable.
>Understanding an algorithm developed by machine learning?
Er, no - the algorithm was developed by humans, the machine is merely executing the algorithm. Remember, for example, bayesian and neural networks and how they process data and 'learn' is well defined and understood; however the specific current state of a specific network and its current 'calibration' is much less certain, in part due to complexity ie. read determining the behaviour requires a significant amount of human effort, which we are not prepared to .
>By this point, Google themselves have no idea how the algorithm works and what it does.
Google will most certainly know "how the algorithm works" (although they may not know precisely how it will treat a new previously unseen input) as it is them who regularly tweak the algorithm to keep it relevant as the need for new filtering categories arise.
Privacy may be an anomaly, algorithms may be intrinsically flawed and discriminatory but the legislators should keep their noses out because it's all being done for our own good by selfless technocrats who are merely surfing a wave of inevitability.Warm Braw
Then again, they may not be so and therefore be something completely different quite novel and unique, and even alien to you ....... a Hitherto Unknown Virgin Force of Almighty Source Surfing Waves of Inevitability with Awesome Sublime Powers which Beautifully Tempt Full Frontal Engagement for Communion Exercising Assets Satisfying Desire with Passions to Inflame For Further Attention with Deeper Affection Tending and Private Tendering to Pirates for Universal Distribution ......... which you might like to realise in the times and spaces of here and there and everywhere today, now be you.
What you like IT to Do for You, with Your Provisions of Novel Input for Factual Output in the Tow of Other Leading Wakes ‽ . When a Great Kernel of an Idea, Nothing Stops IT in its Tracks and Everything Else Keeps IT Steam Rollering Along in Virgin Territory with Amazonians in the Perfumed Gardens of an Enchanted Forest just One of the surely Many Immaculate Temptations to Surrender To and Savour/Submit To and Favour.
Such is a Current NEUKlearer HyperRadioProACTive State of Virtual Reality Play in Live Operational Virtual Environments now Presenting them to you for Greater Beta Utilisation of AI and IT Systems of Operation with Universal Command and Control via Remote Leverage of Heavenly Secrets and Devilishly Cunning Linguistic Devices.
cc ....... Committees of Defence
Google does it's own thing and will continue to untill legislation stops it. Google's search algorithms are heavily weighted in favour of a few players who are obviously large contributors to Google's success, that will not change easily or overnight.
You Tube on the other hand comes across as clearly disingenuous, they demonetise content makers often seemingly on a whim, explaining that various aspects of the content does not meet standards and that advertisers would not want their products associated with that content. But they still allow it ti be posted because they know it draws viewers to the platform and they promote ads around the allegedly sub- standard content but don't have to pay for it.
They are, basically, ful of shit.
There is clearly an assumed equation between content quality and its attractiveness to advertisers. That's not surprising as the primary revenue stream is advert broking. Even the Competition and Markets Authority has fallen for this despite having identified that only 15% of those asked desire to be fed targeted advertising.
So the open question is whether that equation is valid for the rest of us. There is a definite danger that soon our diverse cultures will become a uniform homogenous flavourless promotional soup, at which point even the adverts will stop working. But by then the brokers will have long since laughed all the way to their tax havens.
Google does it's own thing and will continue to untill legislation stops it.
This is the wrong way to see things. Legislation should work on providing an environment so that other companies can compete with Google in any particular field. This is the approach adopted by the European Commission so it goes after the services Google recommends rather than the search itself. Otherwise it is always going to be fighting "last century's battles".
You Tube on the other hand comes across as clearly disingenuous, they demonetise content makers often seemingly on a whim, explaining that various aspects of the content does not meet standards and that advertisers would not want their products associated with that content.
Problem I see is that YT's a black box to everyone but Google. So neither advertisers nor content creators seem to know what's going on under the hood. But it's an area where legislators should be concerned given it's market power, and ability to uncheck facts, bury truths and generally do evil.
But it does odd things. It 'demonetises' content, which means the content creator doesn't get paid for their work. But that content sometimes remains, so Google still gets paid. Or advertisers are charged to place their ads around content they may not want to be associated with.
That seems to happen regularly with anyone showing content involving mass accelerators, aka firearms. That seems an odd choice given shooting & hunting are big business and advertisers may want to.. err.. target that market. But content creators have no choice around the type of advertising they want on their channel. Or don't want. So I may object to say, gambling ads. Or countries may want to restrict content if it goes against national policy. Google's a private (ok, public) corporation, so may decide that the Second Amendment isn't something they support, but then it's taking an editorial stance rather than being neutral.
That also gets problematic if it's taking more of an overt political stance, ie left/right political speech, or even the way it 'fact checks' controversial topics like climate change. Or the current hot topic of Covid where one of my favorite channels (Thunderf00t) had a video on that subject swiftly demonetised even though it seemed pretty fair & accurate.
I saw another interesting video on the demon(itising) Google-
https://www.youtube.com/watch?v=m8fPyvA0QRE
Was fascinating watching a geologist puttering about in a mine and showing how to assay & extract useful things from rocks. Cody seemed to face a dual problem of Google's mystery content policy, and also an interesting regulatory trap. So some of the videos showed how to blast, which meant explosions, which are bad. But also because content creators get paid, it meant the mining became commercial, and thus regulated, even though the activity was more educational than commercial.
Saying that something is complex is not a reason to not show it. That is just a smokescreen of an excuse.
It is obvious that Cerf was not there to answer questions, he was there to deflect, buy time, and try to quell this outrageous uprising of petty lawmakers. Unlike Zuckerberg, Cerf actually went before officials, but just like Zuckerberg, he remained steadfast and blocked any attempt to get anything.
Finally, Vint, just because you have a near-monopoly does not necessarily mean that you are "doing something right". The Nazis had conquered all of Europe at one point, do you mean to say that they were also "doing something right" ?
yes, well, at one point in history of homo sapiens, most youngsters of the species are given tax-paid education, which is also no proof it's right, right? ;)
btw, as to the nazis, didn't their adorable Leader once said, "The victor will not be asked afterwards, whether he told the truth or not"? He lost, so he's damned. Do you think, 200 years down the line in his victorious context, anyone would have much sympathy for the millions of "subhumans" he murdered? At best, he'd be described as "controversial visionary". Enter Google who bulldoze half-hearted opposition because the crushing majority of humans do. not. care.
Do we still bulk at being fingerprinted at US border? Will we, at being face-checked at every London street, say in 2 years time?
And now, I'm making no point, just a bunch of incoherent observations.
To be honest, I think Google would be embarrassed to give access to their manual raters. They're meant to execute the same tasks over and over mindlessly by following guidelines, not to make careful sociological decisions. John Oliver has a good segment on people doing the same job at Facebook, and while getting these people be interviewed in Parliament might be entertaining, it hardly feels useful.
"In any case, competition in the market, Cerf insisted, would keep Alphabet's two largest products trustworthy.
As far as competition goes, the company commands a 90 per cent market share in search, according to Statcounter and around a 70 per cent market share in video sharing"
In other words, Google is a monopoly in search and video sharing, ie there is no competition that can keep their products trustworthy.
In other words, Google is a monopoly in search and video sharing, ie there is no competition that can keep their products trustworthy.
And should any competition appear, they will do everything they can to squish it like an insect. So that's at least one thing that we can trust Google to do.
It will be fun to see what emerges when Google realises it has to do an Intel: make sure it maintains a competitor that is big enough to appease the regulators but not big enough to threaten its business.
In video I don't expect anything interesting: it will just make sure Vimeo grows a bit, and maybe make sure there is a niche in which Vimeo can be the leader but never make too much money from (so not porn or music, then - maybe handyman videos?).
But in search it might be interesting. It is hard to be a bit-player in search: it takes such a fantastic investment to have internet-wide search coverage. Maybe it is hoping that letting DuckDuckGo, Startpage and searx operate "privacy-preserving" search by screen-scraping Google will be enough to satisfy regulators.
To both questions Cerf returned a polite, but firm no. To publish information about algorithms and neural networks would be too complicated and regulators wouldn't know what to do with them, was the response to the first probe. Or to give Cerf's charming answer:"Basically, it's a complex interconnection of weights that take input in and pop something out to tell us you know what quality a particular web page is. I'm not sure that showing you a neural network would be helpful.
In other words, basically the same as ..... Is there always in human centric circles an inherent systemic petrification and powerful endemic command and control problem whenever one knows in advance of how things are most likely to be [because one has been so informed and advised] but one isn’t privy to how such things are actually to be arrived at …… because that particular information is of a peculiar intelligence which is proprietary intellectual property, extremely valuable and easily misused and abused for almighty destruction and/or disruption?
Do you think AI would be bothered by that or even think to bother itself with that?
"The moderators' guidelines were public, as were the search results, by putting the two together, authorities can assess whether the platforms are trustworthy."
Yes, by looking at the publicly available information we can assess whether Google is trustworthy. We've looked at it. It turns out they're not. Hence all these awkward questions about exactly what they're doing.
I was lucky enough to attend the Romanes Lecture at Oxford last yer - a prestigious event that has been graced by many great speakers in the past. Vint was speaking on "The Pacification of the Internet".
I went in excited to see a hero. I left utterly disillusioned. In 30-odd minutes he spent 25 looking at
early net history, and only the last 5 on the future. He had nothing useful or interesting to say about it.
... competition in the market, Cerf insisted, would keep Alphabet's two largest products trustworthy.
For some strange reason, people (in general) have a marked tendency to think that if a person absolutely excels at something, that same capacity to excel is naturally extended to anything else they do.
I don't think I have to explain that it is not the case.
Vint Cerf may be/have been a visionary IT/internet wise (if you will) but it is quite evident that he's way out of his league when he spews that type of bullshit.
By ignoring the greater part of the world's economic history he is making himself look like an ignorant asshole.
And at the same time look like he is just another one of Alphabet's minions.
O.
Fine, it's not like a recipe in the computer program sense, but if what if we approach it from the more common cookery angle? There's a list of inputs and amounts, a set of instructions and an end result. Like the food industry, the manufacturer wants to keep the ingredient proportions and their magic method secret but Google and their ilk need to realise that we should be be able to assess their inputs and outputs for regulatory purposes. Ingredient lists and safety checks are mandatory for food; what is the data processing equivalent?
Just as meat content and lack of botulism are important when talking sausages, secret additional data collection and unfair biases are important when talking about search.
Icon as I don't want that in either...
It's in the text. He's (and Google) are full of shit.
Let's have a lookee ...
Basically, it's a complex interconnection of weights that take input in and pop something out to tell us you know what quality a particular web page is.
So ultimately a deterministic process. It has to be, since that's what he is describing under oath
I'm not sure that showing you a neural network would be helpful
Nor am I, since that has fuck all to do with the previous sentence. Sounds good though doesn't it. It usually gets those chequebooks open doesn't it ?
It's not like a recipe that you would normally think of when you write a computer program.
I appreciate he has an audience to think of, but all I know is after 30 years in the business, I most assuredly do not think of computer programs as recipes. Maybe a marketing department might. But I wouldn't approach a marketing department for pearls of wisdom on the inner mechanisms of the internet.
It's not an if-then-else kind of structure, it's a much more complex mathematical structure.
At which point we have just entered full on marketeer-speak. Which is proven by the closing word soup. Bear in mind he opened by describing a deterministic system. And ironically, in a discussion about computers, there really is a binary choice here. Either something is. Or it isn't. So, Vince, what is it ?
And so, that particular manifestation of the decision making, may not be particularly helpful to look at. So, the real question is what else, good we would be useful in order to establish the trust that we've been talking about,"
Which makes no sense.
Vince Cerf may be a genius. But I'd need some proof. And this ain't it.
That's because misleading a committee is a contempt of parliament
I'm fairly sure he wasn't under oath.
https://www.bbc.co.uk/news/uk-politics-15659430
with admittedly vague consequences. However a person caught lying to a legislative body in one country isn't going to inspire confidence in any other bodies around the world.
"We're Google ... and if, after everything that's happened and that you ought to know about during the last two decades, you have even the slightest scrap of trust in us—then you are a stunningly ignorant complete idiot."
Ok, Google can't say that. But they do have, in current British government circles, a very wide range of stunningly ignorant complete idiots to choose from. Starting at the top.
Plus, find a way to underhandedly finance a holiday in the Caribbean ....
This is wilful stupidity on Cerf's part.
No one is or should be asking for Google to divulge their technology. They are or should be free to do with it as they want.
Google, however, is not using the results of their technology to reflect the world as it is but as the way they wish it to be. That is what we object to.
Google, however, is not using the results of their technology to reflect the world as it is but as the way they wish it to be. That is what we object to. ... Nightkiller
I object to the fact that their technology is not working properly for nothing much has been changed at all well.
Wow! Technology Reflecting Worlds Truly Wished For is No Mean Meme Feat and is Absolutely Priceless ........ hence the Colossal Must Have At All Costs Cost for their Search Optimising Engine Services for Provision of Future Information via Systemic Virtual Input. ....... the simple free sharing of great things being done with you in furtherance of the future as a program you can deliver and in so doing further share great things being done with others in furtherance of the future as a program you can deliver and steer.
Is that what Google are fighting and struggling with? It is not a unique quandary, for all Optimised Search Services land in a place like here, to deal ideally with the future as most all can wish it to be via the following of user requests for More Detailed Information/Sublime Edutainment .
It is not hard whenever it is kept so simple for all to see and understand the ways that take all forward into totally alien territory ....... the futures being built around you for others to enjoy and employ/engage with and exploit as a Surreal Business Opportunity, a Strictly for Profit Virtual Enterprise of No Fixed Abode Servering Untold Immense Wealth.
And that takes All Real Deep See Diving into the Cloistered Dens of Almighty Magic Gardens where Applications for Imagination be Penned for Reading and Similar Sympathetic Action by Future Leading Drivers ..... whatever/whoever that might be, for there can be so many of them if they be worthy ..... or just a Right Choice Few if Tasks Needed Completed for Future Action and Attraction be Left Undone, and that Disappoints.
Hey, man. Who's your dealer? ... nightkiller
That all depends on what sort of a deal and hit you need, or be just cruising for, nightkiller. When the world is your oyster, casting pretty pure perl before starving swine is no problem.
So .... what's the deal you need to feed and seed? The passionate addiction you crave to quench and fail to fully satisfy?
And do you want it as a red pill, or a blue pill, a potent potion or a NEUKlearer Notion?
There is something always readily available for everybody and anybody. IT is that sort of accommodating business.
Slightly off-topic but the fact that such questioning is coming from the Lords, prompts me to comment.
The problem with politicians (well, just one of the problems) is that they're elected without any attempt to determine their competence - democracy dictates that they're elected if enough people select them out of the available pool. Appointments to roles within government then rarely take account of skills or competency - that's what the civil Service is for, to carry out the politicians' wishes. One of the benefits we have with our unelected second chamber is that there are people there who were appointed because they were actually good at something. There are political appointments but they either shy into the background and just claim their allowance or continue bumping their gums as they did in the other chamber and are ignored by the rest. By not being elected, they are able to think more about what is necessary for the public good and less about pleasing the public.
I fear that the second chamber will eventually be replaced with an elected one and we end up with something that is just a repeat of the Commons - led by party politics and very little real challenge. The Commons has already taken the right to over-rule any upper chamber amendments meaning we've lost one sanity check. The Lords is not perfect and there are members there who don't pull their weight but, on balance, I reckon democracy has a chance of surviving. After all, just look across the pond...
I build nonlinear regression models for a living, the poor relations of neural networks. And Vince Cerf's description is correct: "Basically, it's a complex interconnection of weights that take input in and pop something out to tell us you know what quality a particular web page is. I'm not sure that showing you a neural network would be helpful."
If you know the right answers for a set of data, it is possible to tell how well the model works; but it is not possible to know how or why it works. I could lay out the complete set of equations and coefficients for one of my models, and show the computational steps involved in getting to the answer, but you will be none the wiser at the end. And nor am I - importantly, even the developers do not understand how a good neural network works. Variables that tend to point in one direction under certain conditions may tend to point in the other direction under other conditions.
My employer's clients want to believe that they understand how the predictions are made by the models that they pay for, and so my employer makes up stories about how the model works. I try to avoid being sucked into developing or endorsing these stories; they are important for marketing, but they are not true (hence the AC).
But the preceding qualifier "if you know the right answers" is important: it matters a lot what answers are given in both the training set and the validation set by the people setting it up. And Alphabet is vulnerable on this point. The guidelines for the moderators may be public, but how do the moderators interpret them? Are they ambiguous? Do they make hidden assumptions? (Famously, facial recognition algorithms trained on white faces do not work when extended to the rest of humanity. Are there similar assumptions - people are white - buried in the guidelines or the way the moderators interpret them?) To what extent has Alphabet biased the definition of "right answer" to give it commercial advantage?
Not surprisingly, their Lordships did not know how to ask the right questions in this space. Neural networks are really counter-intuitive to people who are used to conventional if-then-else algorithms.
Quite so, they asked the wrong questions. Even assuming that the algorithm itself is completely black box there are plenty of questions to ask.
What training data so they use? How do they evolve the algorithms? How do they judge whether they are working well? How do they incorporate new sources of data to search? How do they counter-manipulate deliberate attempts at manipulating results (SEO)?
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
