Artificial Intelligence; AI
Artificial Intelligence is human is having intellectual ability to artificially implemented in part or whole. In reality, it is just one step away and is a common material in SF water.
hen the term appeared in the beginning of 1956 in the United States Dartmouth Marvin Minsky , Claude Shannon artificial intelligence and information processing theory that people hold Engineers have contributed significantly to John McCarthy is from while using the term. But the concept of artificial intelligence has been around for a while. For example, it was in 1950 that Alan Turing proposed the feasibility and Turing test of a ‘thinking machine’ , and the first neural network model was proposed in 1943.
And, of course, not only the West was interested in this, but the Soviet Union, too, in the book “Red Books” by Anatoly Kittov in his book “ЕГин” Central control system ”, a theory aimed at the pursuit of a better planned economic system and society through computer networking. A further refinement by Soviet computer engineer Viktor Glushkov was the OGAS project.
The history of artificial intelligence has been around since the early 17th and 18th centuries as far back as the early 20th century, but at this time, it remained at the level of philosophical debate about the relationship between the brain and mind rather than artificial intelligence itself . At that time, there was no information processing machine except the human brain. But as time goes by and the stream of computer development innovation begins in earnest in the middle of the 20th century , can we do this well if we can make a brain with a computer and do what we do? This was the opinion that presenting many people considered plausibleAt a rapid pace, artificial intelligence began to enter the realm of learning.
Even in the mid-twentieth century, artificial intelligence research was a highly innovative study that allowed computers to solve problems that were really human-only, such as natural language processing and complex mathematical problems. It is a field made. Naturally, the AI industry was already large enough to create a $ 1 billion market in 1980, so it’s a myth that it wasn’t interested or practical for some reason in the past. However, the limitation of information processing ability at that time, the lack of information, and the trouble that the research funding is stopped for some reason, and the limitation of single layer neural network pointed out in 1969 by Marvin Minsky and Seymour Papert in the publication of “Perceptrons”. This problem was solved by the popularity of the first AI winter / AI winter in the 1970s. Needed.
Since the study of back propagation algorithms, expert systems, and neural network theory resumed in 1974, there has been a lot of research, but the growth is still very disappointing (2nd AI Winter / AI). winter) There have been visible areas such as text recognition and voice recognition, but many of them failed to achieve their goals, such as the failure to develop conversational AI. Conversational AI is also difficult to say that talking to humans. Because of this, since 1990, the goal of AI has become a more cautious and narrow field, focusing on problem solving and business, from the vaguely broad goal of realizing human intelligence.
Since the 21st century , deep learning papers were published by Geoffrey Hinton , 1947 ~ professor in 2006, which led to the unsupervised learning method that was considered impossible. . In addition to AlphaGo, 2017 artificial intelligence has a higher facial recognition rate or better recognition of objects compared to human ability, so that in the field of weak artificial intelligence, it is possible to quickly surpass human ability. This is spreading.
However, there are many points of view that this is just an excellent improvement and practical use rather than an innovative change from the existing paradigm. This is because research on human consciousness and brain realization, which was originally intended, is limited, and there is still no progress without any facts being revealed. Until now, the level of research on how the brain works is so slow that even neuronal circuits, which are the basis of brain analysis, are very slow. Except for being the motif of the neural network, there are few practical points of contact with artificial intelligence.
Although the field of artificial intelligence has become very close to the public, the robots shown in the media are often loaded with artificial intelligence, as is natural, but artificial intelligence receives some information, interprets it, and outputs the result. The problem is that hardware is a hardware- level problem , such as which part of the actuator is controlled and how the system is physically controlled . Alpha and the like, as a collaborative robot driven by artificial intelligence, and simple control algorithms and computer programs that run only in the presence of the complementary relationship which is not a sector closely tied to each other.
Look here and searched the Internet for information related to artificial intelligence inde often the word out about artificial intelligence Weak AI and steel AI Strong AI is John Searle in 1980, John R. Searle, 1932 ~ Professor of the famous Chinese Room and proposed a first argument This is the concept used.
Other documents conveniently describe the implementation of the human mind as complex information processing as strong artificial intelligence, or simply simulating some of the human ability, or weak artificial intelligence, for the purpose of such work. But to be more precise, Professor John Seol’s original intention was to develop “intelligent AI research,” which sees and conducts human minds from the same perspective as computer software. We defined the research on artificial intelligence as “weak artificial intelligence research” to criticize the study of the human mind as computers and software from a philosophical point of view.
Anyway, the concept was strangely matched with the current state of artificial intelligence development, and it became known to the public by slightly twisting the contents. In other words,
strong artificial intelligence = artificial intelligence perfectly mimicking human ,
weak artificial intelligence = artificial intelligence designed as a useful tool You can see.
5.1. Pharmacological intelligence
Weak AI focuses on making a computer perform various problems that humans can easily solve but difficult to deal with, such as finding an object in a picture, listening to sound, and grasping a situation. Artificial intelligence, which is being developed with more realistic and practical goals, rather than aiming at vague human intelligence, is more likely to be used as a tool for solving specific problems rather than something with intelligence.
Based on the above definition, all artificial intelligence created by humans to date can be regarded as artificial intelligence. So far, AI has been developed so that objects developed based on predefined algorithms and vast amounts of data can be used to make actions or decisions that look relatively intelligent.Why can AI solve problems by finding rules? I don’t know if I solved it. The problem can only be solved in a limited range.
Nowadays, “learning” that delivers better output over time and input has been implemented, resulting in programs that outperform humans or perform similarly in limited areas. It is very difficult to see these things like humans. For example, much more than professional articles monarch put the well- alpha aad eventually ” Go Pass” and ” Monarch”I’m better off.” It mimics only a small part of the human ability, and the scope and utility of learning is so limited that it can’t be used for long-terms and any changes to Go’s rules can’t adapt. There is a limit to not being able to extend beyond one,
but as many people think wrongly, the feature of “tool” is not so small. That’s because AI doesn’t necessarily have to be caught up in human imitation, which is easy to understand when you think about the birth of airplanes. I tried to imitate birds, but nowadays the aircraft don’t imitate birds. Modern aircraft can’t sit in trees, eat fish from the river, fly in place in the wind, but no bird can fly at the speed of speed carrying hundreds of people and tens of tons of cargo at thousands of meters up in the air. Just as today’s aircraft is another possibility of flying because they have given up on bird imitation, weak artificial intelligence may not be something less than strong artificial intelligence, but a difference in direction.
The present artificial intelligence has already surpassed human abilities in their own functions, and there are many things that cannot be done by these artificial intelligences. Because they did not imitate human beings, they were not human, so they could transcend human beings. The ability to simply solve a given problem can outperform tough AI. This can be predicted to some extent, even though tough AI has not yet emerged. “Is it an outstanding advantage in solving any problem that mimicking human intelligence?” I can’t say “yes” to the question.
Robust artificial intelligence is a system that embodies human intelligence with information processing capability of computer. It doesn’t matter whether the program has the same intelligence as humans, or the whole brain is scanned and run on a computer. Naturally, it means that even a human can have an intelligent level of intelligence. Therefore, most of the controversial issues of artificial intelligence originate from strong AI.
The words are tremendous, but in fact, artificial intelligence, which can be called robust AI, has not been developed yet. It is said that it embodies the entire human mind in the beginning, but it is not clear what it is and how to define it. How much work does a machine have to be equal to a human being? In the past, if one could play chess or play chess, it would be similar to humans, but the definition was vaguely floating. . Because seeing pictures, listening to sounds, etc., most people can now do as much as they can with a computer.
Even if we can do something similar to humans, there are many barriers to the creation of strong AI. For example , how consciousness , mind , and thought occur is still in the unknown realm, and whether the person who is the party is physically present, or consciousness of artifacts other than the brain.May or may not be granted. In this regard, there is a mind and body problem, which is a question about whether the brain , which can be seen as hardware, and the mind , which can be viewed as software, can be separated from each other. The dualism, the view of the body and soul as separate beings, is called monism. Whether the river can be developed in artificial intelligence will soon Spirit of the brain inde whether or not whether to implement, not a computer, this should be reproduced artificially noeman spirit to appear possible. Therefore, seeing strong artificial intelligence as possible is in line with the mental and monotheistic view. Of course, the scientific community is taking the position of mind and body monism. He proved by various experiments, but “how?” Is still in need of further research.
In the beginning, the Chinese-language argument mentioned above is an argument for distributing Turing test that judges the intelligence of AI with this problem . In summary, the information processing machine should not be good at processing information. It is not necessary to understand itself. In other words, even if an object that claims to be artificial intelligence can seem and act like a person, it cannot be said to have a heart because it is only the result of thorough calculation and information processing . This has been highly countered, and in fact, AI has passed the Turing test in 2014, but in the end it has passed through the psychology of the lack of algorithms, not the improvement of AI. As in this case, Moro only needs to go to Seoul, so it is more likely to become meaningless to make any problems to determine the intelligence in the future . This is also a significant problem for strong AI, because when it is developed, it may be a human-like consciousness , but it can be seen as a program that acts incredibly well as if it were conscious . . So how do you judge this difference based on what?
Moreover, this is not the end, but there is a more fundamental problem. People generally think of human consciousness as free will as something different from a program that responds algorithmically, but the existence of free will has not yet been proven. In other words, as mentioned above, if you want to distinguish ‘real consciousness like humans’ from ‘only appearing conscious when viewed from the outside like a program,’ I have to prove it, but it’s not proven yet. So maybe the formula of human consciousness = just a more complex program is true. See also Chinese room , free will , and determinism .
In addition to the above, the problem of popping heads is scattered, so the current AI research is not based on the theory, but rather by directly studying the brain ‘s operation or research is aimed at simulation. However, it’s clear that any method is hard to see in the near future. What is human intelligence after all? The biggest obstacle is that the answer to the question is still not answered.
More recently, it embodied the concept of river AI Artificial General Intelligence Artificial General Intelligence and Artificial Consciousness Artificial Consciousness was divided into concept.
Often the best known methodologies for AI research are bottom-up and top-down.
The bottom-up approach is that artificial intelligence can be created if the brain’s electronic model can be interpreted by analyzing the neural network of the brain and analyzing chemical reactions. Therefore, it focuses on investigating how the brain works, including the basic interactions of brain cells, and mathematically modeling these behaviors to simulate them on a computer. If strong artificial intelligence is created in this way, it is highly likely to have a structure and operation method close to the human brain. Just tweaking the system to your liking will take more time. Simulating the desired object to see the results and making the simulation results as desired is another matter.
As a representative example of the bottom-up method, a pretty nematode with a detailed neuron map is revealed.There was a case that proved that the neuron’s connection information and the strength of the connection could be realized by the electronic similarity. There is no realization of living creatures, and since human intentions are not included at all, it is doubtful that this should be regarded as artificial intelligence. For example, to avoid light or to make it clutter, or not to make it properly, but before that humans can not manipulate the nervous system to produce the intended result. This is just a simulation. Even 1mm nematode studies are at this level, not to mention more complicated and sophisticated human studies.
On the contrary, the top-down approach focuses on solving algorithms that are very difficult for computers, but that humans can easily solve. Naturally, the purpose of development is more diverse than the bottom-up method, so the development of robust AI may be the goal, but only to solve the problem efficiently. Most of the artificial intelligence that has been researched and developed by humans until now has been born from top-down research and exists in various forms around us, from expert systems to machine learning . And that’s a pretty successful direction.
Modern AI research has been carried out with the proper combination of two parts, and artificial neural network structure, which is a primitive imitation of the structure of neurons, opens the way for the futur
It is not easy to clearly define what AI is and what to call intelligence. And this is not a philosophical problem, but the purpose and direction of the study is completely different depending on which answer you prefer.
One answer is that if a computer can handle a task that requires human ‘intelligence’, then it is artificial intelligence. Another answer is that artificial intelligence is only possible to understand in the same way as humans. These two answers are also detailed, “What is the need for intelligence?” Or “What is the same way as humans?” Depending on the answer to the question, there are many different kinds of answers. Of course, these two answers are not exclusive. It is a dream and hope for many computer engineers to create computers that can handle the things that require ‘intelligence’ with the same kind of intelligence as humans, but at least in the short term it is unlikely to reach that goal.
If we define AI as the one that handles ‘things that need intelligence’, then we don’t have to worry about how humans think, and we don’t have to consider things like emotion. Just go to Morro Garden Seoul. AI research in this direction was predominantly in the early stages of expert systems that deal with the needs of specialists through complex software. These expert systems have no distinction from ordinary software in the way they are implemented, and the way professionals solve problems as much as possible. We focused on providing a way to easily and accurately reflect the software.
Is also in this field and is actually too simple a class of AI, but chess machines are very early on in AI research. It’s very difficult to get chess into operations. Looking at one more number requires 26 times more operations on average, so no matter how modern computers are developing at high speeds, it’s best to look at five to six numbers, and you can’t match tens of chess knights. Do not. For this reason, real chess machines enter a large amount of chess notation so far and then process them by contrasting each other to see if the same shape has ever been found. IBM’s chess machine, which beat the world champion, used more than 70,000 notations. But this is no longer the field of artificial intelligence, it’s just a demonstration of the supercomputer’s ability to perform data parallelism at high speed. In other words, he did not implement complex intelligence, but created a structure that could unfold one piece out of a myriad of things that could be done in a higher accident. According to the definition of the word intelligence, there are professors / scientists who call this an artificial intelligence, and some subtractors, but that doesn’t change anymore.
As detailed in the AI study, when a computer is able to perform a certain task such as chess, it tends to be taken out of the condition of AI directly, rather than the result of AI development. Yes, because Because of the lack of artificial intelligence research.
The approach that has been or has been tried in the meantime is the brain Driving simulation and the like (Brain Simulation), Search, Bottom-Up approach.
In modern times, researches dealing with probability and random algorithms are the most popular. Generally, “A is B!” The problem that can be assumed is relatively easy to access with a computer. However, as ‘art’ may be ‘art’ or ‘technology’, there are many possible answers, and in this case, you need to consider the surrounding situation such as ‘context’. And or ‘technology’ “is difficult to cut and answer. Solving this kind of problem using complex mathematics to deal with statistics and probabilities. In fact, modern artificial intelligence research has given a category corresponding to each word, so that the category is interpreted in many senses when viewed in the whole sentence. A simple example of extreme, when called ‘Music is an art’, by guessing the ‘art’ of two words that mean the category that contains the music and art of the sentence shall be interpreted according to the context, the alpha and also this way Belongs to.
Everledger founder and CEO Leanne Kemp is now using blockchain to track 900,000 karats of diamonds.
Everledger / Hannah Photography
With the price of bitcoin more than doubling from $3,400 to $10,000 since last year’s Fintech 50 list, it’s perhaps no surprise to see some of the largest, and most innovative cryptocurrency companies still holding onto their place on the annual list of startups blazing new trails in financial technology. But what is notably different this year are some mind-numbingly imaginative applications of cryptocurrency that make bitcoin look like child’s play and some new applications of blockchain that don’t involve cryptocurrency at all.
Returning to this year’s Fintech 50 list after a hiatus is cryptocurrency investigation and compliance company Chainalysis, which generated $8 million in revenue in the last complete fiscal year, helping government agencies track down criminals using cryptocurrency and businesses comply with complex regulatory requirements, and in the process becoming the first cryptocurrency company to earn a spot on the Forbes Next Billion-Dollar Startups list. Also new is blockchain startup Everledger, which is now tracking 900,000 carats worth of diamonds on its blockchain, and MakerDAO, which generated $10 million in interest from loans using cryptocurrency as collateral.
Gone from the list are three notable crypto veterans. Bitcoin mining firm Bitfury fell from the list after generating $500 million of revenue in 2018, with an expected drop this year, though to help diversify, it has also expanded its non mining services for enterprises adopting blockchain. Early cryptocurrency firm Circle last year spun off the Poloniex cryptocurrency exchange it acquired for a reported $400 million, refocusing its work on so-called stablecoins that serve no purpose as speculative instruments, but can be used to make purchases (as can more traditional cryptocurrencies), without the lengthy delays associated with cross-border transactions. Cryptocurrency exchange Gemini was also dropped from the list after declining to share information about how its core trading business is doing. Interestingly, Gemini is also getting into the stablecoin space, which may end up having long-term value, but which is currently difficult to monetize.
Veterans on the list, last year, still on in 2020 are Axoni, which just launched a new equity swaps platform with Goldman Sachs and Citigroup as early users, Coinbase, which is focusing on institutional investors, and Ripple, the payments company whose founders created the XRP cryptocurrency, now being used in 10% of Moneygram transactions from the U.S. to Mexico.
Headquarters: New York
Axoni co-founders Jeff and Greg Schvey
Uses blockchain technology to overhaul financial markets infrastructure, most notably the DTCC’s Trade Information Warehouse, which tracks credit derivatives around the globe. Goldman Sachs, Citigroup and other banks just helped launch a new Axoni-built infrastructure for conducting equity swaps.
Funding: $59 million from Citi, Goldman Sachs, JP Morgan, Nyca, Andreessen Horowitz; latest valuation of $171 million*
Bonafides: Trade Information Warehouse handles data on $10 trillion worth of credit derivatives
Co-founders: CEO Greg Schvey, 33, and CTO Jeff Schvey, 34, brothers who previously founded Tradeblock, a service provider for institutional bitcoin traders
Headquarters: New York
After first building tools to help enforcers track down crime and regulatory violations on blockchains, in 2018 it rolled out a product that helps financial institutions comply with know your customer and anti-money laundering rules when it comes to crypto. Newest product, Kryptos, is a clearinghouse of info on players in the cryptocurrency space. Revenues in 2018 hit $8 million, and more than doubled last year, earning company a place on the 2019 Forbes Next Billion-Dollar Startups list.
Funding: $45 million from Accel, Benchmark, and others; latest valuation of $266 million*
Bonafides: 250 customers, including the U.S. government, Barclays and Bittrex
Cofounder & CEO: Michael Gronager, 49; CSO Jonathan Levin, 29, a veteran of the Forbes 30 Under 30 list in Europe; former CTO Jan Moller, 48
Headquarters: San Franscisco
Coinbase co-founder Brian Armstrong
After making its mark as a safe and regulatory-compliant crypto exchange, Coinbase has branched out to offer crypto custodial service to institutions, plus a personal wallet and new currencies designed to appeal to those seeking more privacy.
Funding: $525 million from Andreessen Horowitz, Tiger Global Management, Union Square Ventures and others; latest valuation of $8.1 billion
Bonafides: Coinbase Custody now holds $8 billion in assets for 200 institutional clients
Cofounders: CEO Brian Armstrong, 37, a billionaire based on his Coinbase holdings; board member Fred Ehrsam, 31
Developed a blockchain to track the movement of goods from raw materials source to sales, with its first application tracking diamonds to make sure they don’t come from conflict zones. Now expanding to track cobalt, having signed a contract with the U.S. Department of Energy and a battery trade group in New Zealand.
Funding: $20 million Tencent, Graphene, Rakuten and others; latest valuation of $100 million
Bonafides: 2 million diamonds totaling 900,000 carats are currently being tracked on Everledger’s blockchain, which is being used by 100 stores at the Fred Meyer Jewelers chain in the U.S.
Founder & CEO: Australian Leanne Kemp, 47
Headquarters: New York
MakerDAO co-founder and CEO Rune Christensen
MakerDAO / Carsten Andersen
This decentralized finance platform lets borrowers use volatile cryptocurrency as collateral for loans of stablecoins (called dai) pegged to the U.S. dollar. The borrower pays interest on the loans, but if the crypto collateral falls too far, it’s sold to pay off the loan.
Funding: $63 million from Andreessen Horowitz, Dragonfly Capital, Polychain; latest valuation of $500 million
Bonafides: Maker generated $10 million in interest last year; its platform has already attracted more than 400 outside developers
Founder and CEO: Rune Christensen, 29, founded Try China while still in college, to bring English teachers to China
Headquarters: San Francisco
Facilitates international payments, including with a cryptocurrency created by its founders, XRP, for 300 institutional clients, including Standard Chartered and Santander. In 2019, sold $500 million of XRP, using proceeds to expand and invest $50 million in Moneygram, which now uses XRP in 10% of its cross-border transactions to Mexico.
Funding: $293 million from Accenture, Andreessen Horowitz, Google Ventures and others; latest valuation of $10 billion
Bonafides: Still owns $12 billion worth of XRP tied up in escrow accounts
Cofounders: Executive chairman Chris Larsen, 59; Jed McCaleb, 49; Arthur Britto
CEO: Brad Garlinghouse, 49, a former AOL president
Pedestrians pass a JPMorgan Chase & Co. bank branch near the New York Stock Exchange in 2018.
Bloomberg | Bloomberg | Getty Images
JPMorgan Chase is in talks to merge its marquee blockchain unit Quorum with Brooklyn-based start-up ConsenSys, according to people familiar with the discussions.
The deal is likely to be formally announced within six months, but financial terms are unclear, the people said.
Around 25 people work on Quorum, and it’s unclear whether they would join ConsenSys after the merger, the people said.
Blockchain emerged over a decade ago as the software tracking cryptocurrency transactions. Since then, banks and other large corporations have been investing millions of dollars to develop and test a range of business applications using the nascent technology. Efforts have had mixed results, with few projects achieving significant impact.
JPMorgan built the Quorum blockchain internally using the ethereum network, the software that underpins ether, one of the best-known cryptocurrencies.
It is being used by JPMorgan to run the Interbank Information Network, a payments network that involves more than 300 banks. JPMorgan, the largest U.S. bank by assets, also said it would use Quorum to issue a digital currency called JPMorgan Coin that it designed to make instantaneous payments using blockchain.
A merger with ConsenSys would have no impact on the Interbank Information Network and other JPMorgan projects running on Quorum, one of the people said.
JPMorgan has been considering spinning off Quorum for around two years, evaluating options including setting up an open-source foundation, creating a new start-up or merging it with another company, the person said.
A merger with ConsenSys was chosen as the best path forward because both organizations work with ethereum and have been involved in joint initiatives in the past.
ConsenSys, a prominent blockchain start-up that grew rapidly during the 2017 crypto bubble, was founded by Joe Lubin, one of the co-founders of ethereum. The company announced last week that it had laid off around 14% of its staff in a restructuring to separate its software development business from its venture activities.
A merger with Quorum would align with its shift toward growing its software division.
Quorum is open source, meaning its code is free and may be modified and redistributed. The plan after the merger is to maintain the Quorum brand and keep the technology open source, one of the people said.
Finding what cloud services employees are using is half the battle–integrating Microsoft Cloud App Security and Defender Advanced Threat Protection lets you track, block, or audit cloud app usage.
Shadow IT used to mean someone writing Excel macros or setting up an unofficial file server for their team. Now, putting AWS and Salesforce subscriptions on the company credit card or using Dropbox to share documents with suppliers and partners is just the tip of the iceberg. The average enterprise has over 1,500 cloud apps in use, most of them not provisioned by the IT team, and is uploading more than 80GB a month to cloud services from within the business.
There are various tools that promise to detect cloud services in use inside your network, usually by scanning network traffic and checking firewall logs (although services like Zylo plug into financial systems to find SaaS subscriptions that departments are buying directly without going through IT). Not only is that cumbersome and likely to impact network speed, but it doesn’t help with remote users: you can only detect and block web applications by sending everyone through your corporate network — something that will be both unpopular and hard to enforce.
You get more control by analysing and controlling cloud app usage on the device directly. Microsoft Cloud App Security (MCAS, a Cloud Application Service Broker, which is itself a cloud service) now includes a shadow IT discovery tool that integrates with Defender ATP to discover cloud app and service usage on any managed device. Defender already monitors what processes are running and what files are being opened as it checks for malware, and that same information lets it report back on what cloud services are being used. It’s an automated process that creates a catalog of cloud apps that are in use, by which users and on which devices, with security and compliance risk scores for each app.
But because MCAS is integrated with Defender, you get the option to block and whitelist apps directly on the device. That works for all devices, not just the ones on the corporate network — and it even lets you enforce read-only access to your corporate resources for external users like suppliers and partners.
Defender ATP can look at the labels you apply with Azure Information Protection, so you can tell when data tagged as sensitive or confidential is being sent to a cloud service (or if files have been downloaded and stored on a device) — in real time, or later on, even if that device is no longer connected for you to scan or available for you to look at physically. Niv Goldenberg, principal group PM manager for cloud security, compares it to a flight recorder: the data is stored for six months in your cloud tenant, so you can go back and audit which files were accessed and what happened to the data if you need to do an investigation, which is an additional layer of protection even for approved cloud applications.
Cloud app reputation
The idea is to give organisations back the visibility and control over users and data that they lost when SaaS services arrived, Goldenberg told TechRepublic. “If I want to start using a new application, all I need to do is open my browser and type in a URL and I’m using an application that the IT department probably isn’t familiar with and doesn’t have any visibility or control over. Knowing what applications employees are using is one thing, but before you can decide if this is an application I want to encourage employees to use, or an application that I want to block, you need to know whether the application is secure enough. Based on the metrics we see of our customers, typically organisations use more than 1,500 cloud applications and there’s no way an IT admin knows all these applications and how secure they are.”