The UK home office has just announced plans to protect paedophiles for well over a decade and they are paying millions to make it happen. Are you offended yet? You should be. The article (at https://www.theguardian.com/technology/2019/sep/17/home-office-artificial-intelligence-ai-dark-web-child-sexual-exploitation) is giving you that, yet you do not realise that they are doing that. The first part is ‘Money will go towards testing tools including voice analysis on child abuse image database‘, the second part is “Artificial intelligence could be used to help catch paedophiles operating on the dark web, the Home Office has announced” these two are the guiding part in this, and you did not even know it. To be able to understand this there are two parts. The first is an excellent article in the Verge (at https://www.theverge.com/2019/1/28/18197520/ai-artificial-intelligence-machine-learning-computational-science), the second part is: ‘AI does not exist!‘
Important fact is that AI will become a reality at some point, in perhaps a decade, yet the two elements making AI essential have not been completed. The first is quantum computing, IBM is working on it, and they admit: “For problems above a certain size and complexity, we don’t have enough computational power on Earth to tackle them.” This is true enough and fair enough. They also give us: “it was only a few decades ago that quantum computing was a purely theoretical subject“. Two years ago (yes only two years ago) IBM gives us a new state, a new stage in quantum computing where we see a “necessary brick in the foundation of quantum computing. The formula stands apart because unlike Shor’s algorithm, it proves that a quantum computer can always solve certain problems in a fixed number of steps, no matter the increased input. While on a classical computer, these same problems would require an increased number of steps as the input increases” This is the first true step towards creating AI, as what you think is AI grows, the data alone creates an increased number of steps down the line, coherency and comprehension become floating and flexible terms, whilst comprehension is not flexible, comprehension is a set stage, without ‘Quantum Advantage with Shallow Circuits‘ it basically cannot exist. In addition, this year we get the IBM Q System One, the world’s first integrated quantum computing system for commercial use, we could state this is the first true innovative computer acceleration in decades and it has arrived in a first version, yet there is something missing and we get to stage two later.
Now we get to the Verge.
‘The State of AI in 2019‘ published in January this year gives us the goods, and it is an amazing article to read. The first truth is “the phrase “artificial intelligence” is unquestionably, undoubtedly misused, the technology is doing more than ever — for both good and bad“, the media is all about hype and the added stupidity given to us by politicians connected the worst of both worlds, they are clueless and they are trying being dumb and clueless on the worst group of people, the paedophiles and they are paying millions to do what is cannot accomplish at present.
Consider a computer or a terminator super smart, like in the movies and consider “a sci-vision of a conscious computer many times smarter than a human. Experts refer to this specific instance of AI as artificial general intelligence, and if we do ever create something like this, it’ll likely to be a long way in the future” and that is the direct situation, yet there is more.
The quote “Talk about “machine learning” rather than AI. This is a subfield of artificial intelligence, and one that encompasses pretty much all the methods having the biggest impact on the world right now (including what’s called deep learning)” is very much at the core of it all, and it exists and it is valid and it is the point of set happening, yet without quantum computing we are confronted with the earlier stage ‘on a classical computer, these same problems would require an increased number of steps as the input increases‘, so now all that data delays and delays and stops progress, this is the stage that is a direct issue, then we also need to consider “you want to create a program that can recognize cats. You could try and do this the old-fashioned way by programming in explicit rules like “cats have pointy ears” and “cats are furry.” But what would the program do when you show it a picture of a tiger? Programming in every rule needed would be time-consuming, and you’d have to define all sorts of difficult concepts along the way, like “furriness” and “pointiness.” Better to let the machine teach itself. So you give it a huge collection of cat photos, and it looks through those to find its own patterns in what it sees” This learning stage takes time, yet down the track it becomes awfully decent in recognising what a cat is and what is not a cat. That takes time, yet the difference is that we are seeking paedophiles, so that same algorithm is used not to find a cat, but to find a very specific cat. Yet we cannot tell it the colour of its pelt (because we do not know), we cannot tell the size, shape or age of that specific cat. Now you see the direct impact of how delusional the idea form the Home Office is. Indirectly we also get the larger flaw. Learning for computers comes in a direct version and an indirect version and we can both put it in the same book: Programming for Dummies! You see, we feed the computer facts, but as it is unable to distinguish true facts from false facts we see a larger failing, the computer might start to look in the wrong direction, pointing out the wrong cat, making the police chase and grab the wrong cat and when that happens, the real paedophile had already hidden itself again. Deep Learning can raise flags all over the place and it will do a lot of good, but in the end, a system like that will be horribly expensive and paying 100 police officers for 20 years to hunt paedophiles might cost the same and will yield better results.
All that is contained in the quote: “Machine learning systems can’t explain their thinking, and that means your algorithm could be performing well for the wrong reasons” more importantly it will be performing for the wrong reasons on wrong data making the learning process faulty and flawed to a larger degree.
The article ends with “In the here and now, artificial intelligence — machine learning — is still something new that often goes unexplained or under-examined” which is true and more important, it is not AI, the fact that we were not really informed about, there is not AI at present, not for some time to come and it makes us wonder on the Guardian headline ‘Home Office to fund use of AI to help catch dark web paedophiles‘, how much funds and the term ‘use of AI‘ requires it to exist, which it does not.
The second missing item.
You think that I was kidding, but I was not, even as the Quantum phase is seemingly here, its upgrade does not exist yet and that is where true AI becomes an optional futuristic reality. This stage is called the Majorana particle, it is a particle that is both matter and antimatter (the ability to be both positive and negative), and one of the leading scientists in this field is Dutch Physicist Leo Kouwenhoven. Once his particle becomes a reality in quantum computing, we get a new stage of shallow circuits, we get a stage where fake news, real news, positives and false positives are treated in the same breath and the AI can distinguish between them. That stage is decades away. At that point the paedophile can create whatever paper trail he likes; the AI will be worse than the most ferocious bloodhound imaginable and will see the fake trails faster than a paedophile can create it. It will merely get the little pervert caught faster.
The problem is that this is decades away, so someone should really get some clarification from the Home Office on how AI will help, because there is no way that it will actually do so before the government budget of 2030. What will we do in the meantime and what funds were spend to get nothing done? When we see: “pledged to spend more money on the child abuse image database, which since 2014 has allowed police and other law enforcement agencies to search seized computers and other devices for indecent images of children quickly, against a record of 14m images, to help identify victims“, in this we also get “used to trial aspects of AI including voice analysis and age estimation to see whether they would help track down child abusers“, so when we see ‘whether they would help‘, we see a shallow case, so shallow that the article in the Verge well over half a year ago should indicate that this is all water down the drain. And the amount (according to Sajid Javid) is set to “£30m would be set aside to tackle online child sexual exploitation“, I am all for the goal and the funds. Yet when we realise that AI is not getting us anywhere and Deep Learning only gets us so far, and we also now consider “trial aspects of AI including voice analysis and age estimation” we see a much larger failing. How can voice analyses help and how is this automated? and as for the term ‘trial aspects of AI‘, something that does not exist, I wonder who did the critical read on a paper allowing for £30 million to be spend on a stage that is not relevant. How about getting 150 detectives for 5 years to hunt down these bastards might be cheaper and in the end a lot more results driven.
In the end of the article we see the larger danger that is not part of AI, when we see: “A paper by the security think-tank Rusi, which focused on predictive crime mapping and individual risk assessment, found algorithms that are trained on police data may replicate – and in some cases amplify – the existing biases inherent in the dataset“, in this Rusi is right, it is about data and the data cannot be staged or set against anything, which makes for a flaw in deep learning as well. We can teach what a cat is by showing it 1,000 images, yet how are the false images recognised (panther, leopard, or possum)? That stage seems simple in cats, in criminals it is another matter, comprehension and looking past data (showing insight and wisdom) is a far stretch for AI (when it is there) and machine learning and deeper learning are not ready to this degree at present. We are nowhere near ready and the first commercial quantum computer was only released this year. I reckon that whenever a politician uses AI as a term, he is either stupid, uninformed or he wants you to look somewhere else (avoiding actual real issues).
For now the hypes we see are more often than not the lie of AI, something that will come, but unlikely to be seen before the PS7 is setting new sales records, which is still many years away.