Week 8 Digital Systems
How will AI change the world?
-robot and driverless cars -detect fake news -speech and language recognition -facial recognition and surveillance -police use, can be too intrusive, pausing sale to law enforcement -healthcare -reinforcing discrimination- learning systems can codify the human biases and social inequities reflected in their training data -AI and global warming -powering machine-learning models and datasets can contribute to global warming
Which are the leading firms in AI
Google- DeepMind AI AlphaFold and AlphaGo systems probably made the biggest impact on public awareness of AI -other major tech companies make headlines regularly
What are the elements of machine learning?
supervised learning unsupervised learning reinforced learning
Corporate Hacks
who was hacked? Why should you care? -SSN -Credit Cards -Loans -Identity -hackers use informations against individuals
Will and AI steal your job
won't replace all jobs, but will change nature of work -new jobs would be created to replace those lost, as with every technological shit, but it's uncertain whether these new roles will be created rapidly enough to offer job to those displaced or if these workers are even suited for these roles -for some, AI is a technology that will augment rather than replace workers
Regulations-cybersecurity
-Privacy and Personal Data Protection California Consumer Privacy Act (CCPA) -Few Federal Cybersecurity regulations -1996- Health Insurance Portability and Accountability Act (HIPPA) -1996- Gramm-Leach- Bliley Act -2002- Homeland Security Act, includes -Federal Information Security Management Act (FISMA)
Which AI services are available?
-all major cloud plaforms- AWS, Microsoft Azure, Google Cloud Platform-provide access to GPU arrays for training and running machine-learning models, Google gearing up to let users use their Tensor chips -all of the necessary associated infrastructure and service are available from the big 3
what is machine learning?
-computer system learns how to performa a task rather than programmed how to do so -to learn, systems are fed huge amounts of data, which they then use to learn how to carry out a specific task -quality and size of dataset important
Protection-cybersecurity
-considerations -antivirus- check for malware-3rd party software -ad blocker -avoid dodgy plugins -back-up files- personal files in case of corruption- not connected to network -disable macros -don't just open attachments -firewall -VPN
Cybersecutity
-flash drives can have viruses and compromise data -hijacking your systems! -who borrows flash drives? -how else do you share data? -vectors for virus and malware
password strength
-how secure are your passwords? -use numbers, letters, and symbols -make them as long as possible
Artificial Intelligence
-intelligence- efficiency with which you acquire new skills at tasks you didn't previously prepare for -AI systems typically demonstrate at least some of the following behaviors associated with human intelligence: planning, learning, reasoning, problem-solving... to a certain extent, social intelligence and creativity
Why hospitals are the perfect target for ransomware
-malware works by locking your computer to prevent you from accessing data until you pay a ransom, usually demanded in BitCoin -hospitals are perfect for this extortion because they provide critical care and rely on up-to-date information from patient records -without these up-to-date info, patient care can get delayed or halted, much makes hospitals more likely to pay a ransom to avoid death and lawsuits -another reason is hospitals workers are not trained in security awareness, don't focus on cybersecurity in general. Their primary concern is HIPAA compliance, ensuring employees meet federal requirements for protecting patient privacy
What can Narrow AI do?
-organizing personal and business calendars -responding to simple customer-service queries -helping radiologists to spot potential tumors in x-rays -new applications of learning systems emerging all the time -sometimes ambitions for AI tech outstrips reality, as much potential as these systems has
Case Study
-ransomware forces shutdown -the incident led to the cancellation of 2,800 patient appointments at the NHS Trust -took 3 days to resolve -hospital are an appealing target for cyberwar
Cybersecurity- Core Fundamental
1. What do you want to protect? 2. Who do you want tp protect it from? 3. How likely is it you will need to protect it? 4. How bad are the consequences if you fail? 5. How much trouble are you willing to go through to try to prevent those consequences?
barring this, a victim has two options
1. pay the ransom 2. restore the data from backups -If formal backups don't exist, it may be possible to restore data using Shadow Copy files and other methods. Best action is for hospitals to take steps to prevent attacks and maintain weapon-grade backups -security awareness training for employees is also key to prevent them from clicking on phishing emails -hospitals can whitelist their machines to prevent ransomware installing. Involves scanning a machine to note all legitimate applications on it, then configuring it or block any other executables -configure mail services to block zip or other files that are likely to be malicious. Also, organizations should restrict permission to areas of the network -force attackers to work harder
Strategic planning- 8 related causes of security failure
1. society 2. lack of transparency 3. invisible systemic risk 4. the cultural disconnect 5. throwing money at the problem 6. poor risk-appetite statements 7. CISOs as defenders 8. Broken Accountability
The AI revolution
2010 Flash Crash -ANI caused stock market to plummet -$1 trillion -quick recovery
What are the two most likely scenarios with AI?
AI is programmed to do something devastating -AI is programmed to do something beneficial, but it develops a destructive method to achieve its goals
The beginning of artificial intelligence
Dartmouth professor John McCarthy coined the phrase in summer of 1956 -AI began wing promising results on more human tasks, Arthur Samuel late 1950s program that play checkers. In 1962, one scored a win over a master at the game -As AI developed, different strategies for making machines smarter did too -the field hit milestones after milestones -more human knowledge in code and systems better at task -deep learning, is a revival of one of the oldest idea in AI -involves passing data through webs of math loosely inspired by how brain cells work -artificial neural networks -deep learning can also deliver a significant jump in the accuracy of speech recognition -deep-learning experts high in demand
Other types of AI
Evolutionary computation Expert systems
Summer of hacks continues with massive ransomware attack
July 2, 2021, Kaseya Ltd targeted for a ransomware attack -Kaseya's business provides services to managed providers who in turn offer IT services to small and medium-sized businesses -ransomware group successful in spreading an infection to about 1,500 businesses -by ransomware gang REVIL- used Russian language and asked for 70 million to unlock -ransomware encrypts files on a victim's computers, rendering them useless until a payment is made -numerous instances occur without being public -Schools, government agencies, and health care providers-primarily within US- have fallen victim
different types of neural networks
Recurrent Neural Networks(RNN) -understand meaning of text and speech recognition Convolution neural networks-roots in image recognition and have uses as diverse as recommender systems and NLP
Which countries are leading the way in AI
US is ahead now, but China has Alibaba, Baidu, and Lenovo that invested heavily in AI -Baidu-self-driving cars using Baidu's deep-learning algorithms AutoBrain
unsupervised learning
algorithms try to identify patterns in data, looking for similarities that can be used to categorize data -algorithms isn't set up in advance to pick up specific data, it simply looks for data that its similarities can group
What can General AI do?
artificial General Intelligence (AGI) -experts have different predictions -probably still centuries away
Password management
basics -keep them in your head -don't change them? -reuse them? management -consider a passphrase -access a vault of your passwords -two-factor authentication
The future of AI
big tech companies such as Google, Microsoft, and Amazon have amassed strong rosters of AI talents and impressive arrays of computer to bolster their core business of targeting ads or anticipating your next purchase -invites others to run AI projects on their networks -improvements to AI hardware, growth in training courses in machine learning, and open-source machine-learning projects will also accelerate spread of AI into other industries -consumers can expect to be pitched more gadgets and services with AI powered features -virtual assistant and smart speakers -commercial possibilities makes this a great time to be an AI researcher -labs investigating how to make smarter machines are more numerous and better-funded than ever -AI systems invite more scrutiny as they grow more powerful -government use of software such as criminal justice is often flawed or secretive, and corporations like Facebook have been confronting the downsides of their own life-shaping algorithms -more powerful AI has potential to create worse problems such as perpetuating historical biases and stereotypes against women or black people -we have to get smarter about machines, now even tech industry making rules and ethics of AI
The Capital One data breach is alarming, but these are the 5 worst corporate hacks
capital One's data breach affected about 100 million customers- over 30% of US population 1. Yahoo-3 billion accounts in 2013 2. Yahoo- 500 million accounts in 2014 3. Marriot/ Starwood- 500 million guests in 2018 4. Friend Finder Network- 412 million accounts in 2016 5. Equifax: 146 million accounts in 2017
Expert systems
computers are programmed with rules that allow them to take a series of decisions based a large number of inputs, allowing that machine to mimic behavior of a human expert in a specific domain -autopilot system flying a plane
Data breaches, cyberattacks are top global risks alongside natural disasters and climate change
cyberattacks and their potential to cripple critical infrastructure remain on of the biggest risks facing the world today, the World Economic Forum (WEF) warned -cyberattacks and data breaches featured heavily in last year's report (2018), ranking as third and fourth most likely risk -5th and 4th respectively -cyberattacks pose a risk to critical infrastructure -hackers gaining access to control rooms of US utility companies -one key reason why cyberattacks on infrastructure are viewed as a significant risk by the WEF -rise of digitalization and growth of Internet of Things in recent years -increased connectivity across the world has increased potential for hackers to conduct campaign -Global Risks Reports warns that organizations need to make progress on this blind spot in order to prevent damaging consequences
Will AI kill us all?
depends on who you ask -Elon Musk- push for stronger regulatory oversight and more responsible research into mitigating downside of AI
Ransomware- A profitable business
digital extortion racket been around since 2005, it works -victims generally get infected through phishing attacks that carry a malicious attachment or instruct recipients to click on URL that downloads malware to their computer. Also, victims can get infected through malvertising if they visit a website that's serving up compromised ads -payoff can be huge -ransomeware attackers have gone further, encrypting an entire organization from accessing shared file and database. The worst ones also go after backup repositories that victim might use to restore data -FBI Flash warning- MSLI/Samas -"don't have to lock an entire network"- only crucial files to handicap an organization
Which of the major tech firms is winning the AI race?
each tech giants use AI to help drive myriad public service -virtual assistants, most visible manifestation SIri, Google assistant, Cortana
How do I get started with AI?
easiest way is via the cloud -all major tech firms offer AI services
Alan Turing (1912-1954)
father of AI and computer science -cryptanalyst during WW2 -cracked code that enabled Allies victory -contributions to physical sciences -prosecuted by UK and castrated for being gay -died of cyanide poisoning -receives public apology in 2009 and royal pardon in 2013
Notable Hacks
government supported hacking groups from Russia, China, N.Korea, Iran and even U.S pulled off sophisticated and audacious hacks examples 1. Colonial Pipeline Company Attacker- Darkside, April 2021- 4.4 million in ransom 2. City of Baltimore attacker- Robinhood ransomware-May 2019
Meet Sophia
granted legal citizenship in Saudi Arabia
Robots are everywhere
in many fields such as space explorations, health care, defense... -some are fully autonomous, others requiring some human input, extends our grasp, enhance our capabilities and travel to places too dangerous or difficult for us to go -assembly lines Shanghai GM -robot arm KFC- Moscow, Russia -drone delivering Dominios- Netherlands
machine learning
involves "training" computers to perform tasks based on examples, rather than by relying on programming by a human -deep learning- a technique has made this approach more powerful -AI posed to reinvent many areas of life -health care- hospitals in India are testing software for signs of diabetic retinopathy -vital in projects in autonomous driving
What are the uses for AI?
is everywhere -virtual assistance -Alexa and Siri -spot spam -detect credit card fraud
What are neural networks?
key to machine learning success -are mathematical models that are able to tweak internal parameters to change what they output. A neural network is fed datasets that teach it what it should spit out when presented with certain data during training -structure and functioning of neural networks very loosely based on the connections between neurons in the brain -neural networks made up of interconnected layers of algorithms that feed data into each other. Can be trained to carry out specific tasks by modifying the importance attributed to data as it passes between the layers. During training, the weights attached to data as it passes between layers will continue to be varied until the output from the neural network is very close to what is desired. At this point, the network will have "learned" how to carry out a particular task -subset of machine learning-deep learning
reinforced learning
like rewarding a pet when it performs a trick -system attempts to maximize a reward based on its input data, basically going through a process of trial and error until it arrives at the best possible outcome
What is fueling the resurgent in AI?
machine learning and deep learning breakthroughs possible due to easy availability of data and an explosion in parallel computing power, during which the use of clusters of GPUs (graphic processing units) to train machine-learning systems has become more prevalent -clusters also available widely as cloud services over the internet -major tech firms have moved to using specialized chips tailored to both running and training machine-learning models
subset of machine learning-deep learning-neural networks
neural networks are expanded into sprawling networks with large numbers of sizable layers that are trained using massive amounts of data -have fueled the current leap forward in the ability of computers to carry out tasks like speech recognition and computer vision
What is threat modeling?
one of the most important component of staying secure online -is thinking about what data you want to protect, and who you're trying to protect it from -"What are you worried might get hacked?" -to figure out your threat mode, consider these 5 questions 1. What do you want to protect? 2. Who do you want tp protect it from? 3. How likely is it you will need to protect it? 4. How bad are the consequences if you fail? 5. How much trouble are you willing to go through to try to prevent those consequences? -overestimating your threat can be a problem too. Using technical things that are not necessary wastes time and can be putting yourself at risk. -might get a false sense of security and overlook actual threat
Narrow AI
seen around us in computers today- intelligent systems have been taught or have learned to carry out specific tasks without explicitly programmed to do so Examples Siri Vision-recognition system as self-driving cars -these systems can only learn and be taught how to do defined tasks
Evolutionary computation
sees genetic algorithms undergo random mutations and in an attempt to evolve the optimal solution to a given problem -approach have been used to design AI models, effectively using AI to help build AI -"neuralevolution"
Hackers that exposed Personal Information
since Jan. 2020, more than 20 corporate, government, and nonprofit organizations have experienced such breaches which exposed more than 899 million records -over course of a decade, the tally exceeds 11 billion records across 365 entities -nowadays, user and corporate data among a business's most valuable and most sensitive data
Phishing scams
take information from you-phone, email Case Study -Google and Facebook targeted -lost $100 million Best practice -careful of any suspicious emails -don't click on random links
supervised learning
teaches AI by using many labelled examples- machine-learning systems fed huge amounts of data, which has been annotated to highlight the features of interest. Once trained, the system can then apply these labels to new data -need both large amounts of data and computing data
AGI tests
the turing test-test machine for intelligent behavior -the coffee test- hard for AI to describe coffee making process -the College Student Tests -the Employment Test
From AGI to ASI
this could happen soon -experimentation -innovation
What is Artificial Intelligence?
three types Narrow(ANI)-all AI are this now General(AGI) -human intelligence Superintelligence(ASI) -beyond human intelligence
General AI
type of adaptable intellect found in humans, a flexible form of intelligence capable of learning how to carry out vastly different tasks
Many things AI still can't do
understand nuances of languages common-sense reasoning learning a new skill from one or two examples
Vulnerability
what put users at risk? -forms of malware? -who is impacted? -are we safe? -can this problem be solved? -combat strategies?
The AI revolution: are you a robot?
what tests have you encountered? what makes this challenging for AI?
The Future
what will the job market look like when you graduate? -more advanced tech -jobs different -AI can create jobs that never existed before
How hospitals can protect themselves from cyberattacks
when ransomware strikes a hospital, the first reaction is often panic Proper response: -victims to disconnect infected systems from a network and disable WiFi and Bluetooth to prevent the malware from spreading. Victims are also to remove any USB sticks or external hard drives connected to an infected computer to prevent those being locked as well -knowing the strain of ransomware is on your system if it's well-known, there may be information published online by security firms or even tools that an bypass the encryptions -barring this, a victim has two options
What are recent landmarks in the development of AI
while modern narrow AI may be limited to performing specific tasks, within their specialism, these systems are sometimes capable of superhuman performance -many breakthroughs: -2009- Google showed its self-driving Toyota Prius could complete more than 10 journeys of 100 miles each -2011-IBM Watson won US quiz show Jeopardy!, beating two of the best players the show ever produced -Watson used natural language processing and analytics on vast repositories of data that's processed to answer human-posed questions
John Deere's promotional USB drive hijacks your keyboard
while the USB drive didn't do anything to compromise the security of devices it was connected to, the method is similar to a malicious attack -it's an HID compliant keyboard that, when connected detects what platform it's on and automatically sends a keyboard shortcut to open a browser, and then it barks the link into the address bar -John Deere's director of public affair, Ken Golden, said the distribution wasn't malicious, said they distributed same type of USB drive in the past -security experts recommends to never plug in any USB drives whose origin or contents are unknown -Hackers have put malware inside USB drives in the past, it's not not just a theoretical attack, even though John Deere's USB contained no malware