AI ETHICS PROBLEMS, QUESTIONS, HOPES, REGRETS by lemi orhan ergin, co-founder of Craftgate

LEMi ORHAN ERGiN lemiorhanergin.com @lemiorhan Bs and MSc @ Marmara University CSE, 2002 / 2005 developing software, coding since 2001 alumni of Sony, GittiGidiyor/eBay, ACM, iyzico co-founder @ Craftgate, craftgate.io founder of Turkish Software Craftsmanship Community talking about professional ethics since 2008 you can contact with me from social media

When I tweeted about one of the most dangerous vulnerabilities an operating system can have, people accused me of exposing irresponsibly and unethically. In fact, the vulnerability had been already exposed in public forums. We couldn’t achieve to make Apple understand the situation. The only thing I did is push the alarm button. The Root Bug: Zero-Day Vulnerability for MacOS High Sierra

https://speakerdeck.com/lemiorhan https://www.youtube.com/watch?v=DClFjk_Uod8 https://www.youtube.com/playlist?list=PLQTv1b9jwvWdvUVfv0M55mRbTB8CMYT9R Slides of my talks about ethics: My TEDx Talk about ethics: Recordings of my all talks: MY TALKS ABOUT PROFESSIONAL ETHICS

No content

An expert in cybernetics at Cyberdyne Systems Corporation as the Director of Special Projects Miles Bennett Dyson

An expert in cybernetics at Cyberdyne Systems Corporation as the Director of Special Projects Miles Bennett Dyson He is the original inventor of the microprocessor which would lead to the development of Skynet, an intelligent computer system intended to control the United States military. It later achieve sentience and launch a global war of extermination against humanity when going online on August 4th, 1997. It becomes self-aware at 2:14 am EST, on August 29th, 1997, and sent nuclear bombs to Russia.

AI revolution already started. The steps we take today will determine whether we will face Skynet or not

AI revolution already started. The steps we take today will determine whether we will face Skynet or not Fortunately, we saw the same movie centuries ago…

The transition from creating goods by hand to using machines The Industrial Revolution (approx. 1760s – 1914)

1764: Spinning Jenny 1769: Water frame, Steam engine 1785: Power loom 1793: Cotton gin 1804: The first railway steam locomotive 1830: Liverpool to Manchester railway line opened The transition from agrarian economies to industrial and manufacturing ones First Industrial Revolution Water frame Spinning Jenny Steam Engine Cotton Gin Power Loom First Locomotive (approx. 1760s – 1840s)

1870s: The spread of railroads and the telegraph 1876: Telephone 1879: Practical electric light bulb 1886: First automobile powered by an internal combustion engine 1903: First powered flight 1908: Assembly line method of production (approx. 1870s – 1914) Technological revolution characterized by the widespread adoption of steel, petroleum, and electricity Second Industrial Revolution First Railroads First Telephone First Internal Combustion Engine First Electric Light Bulb First Flight First Assembly Line

Industrial Revolution changed the way we live People moved from rural areas to cities Farming economies shifted to industrial and manufacturing Technology became an integral part of human life Public education became more common Factories opened all over the world Rise in consumerism, people buy without any need The rapid increase in the human population the impact is so huge that’s why we call it revolution

Industrial Revolution introduced ethical problems Long working hours, poor pay, unsafe working conditions, child labor Economic inequality, social injustice Significant environmental damage and climate change Poor living conditions and health problems Reducing workers, dehumanization of work Displacement of traditional skills and jobs, loss of craftsmanship mindset we try to raise awareness of mastery and craftsmanship 
 for software development at SCTurkey community

The transition from thinking and making decisions on your own to trusting machines to do it for you The AI Revolution (approx. 2010s – {unknown})

1943: Birth of neural networks concept 1950: Turing test by Alan Turing 1951: First chess program 1956: Birth of Artificial Intelligence term at Dartmouth Conference 1959: First AI program: General Problem Solver 1965: ELIZA natural language processing computer program 1986: The backpropagation algorithm was rediscovered 1997: Deep Blue defeated Kasparov 2002: Roomba, a robotic vacuum cleaner, was introduced 2005: DARPA Grand Challenge for self-driving cars 2006: Deep learning concept introduced The birth of concepts, first show- offs, attracting people's attention The Footsteps of AI

2012: Deep learning model won ImageNet image recognition contest 2013: Google's Word2Vec captures semantic relationships between words 2015: Google’s AlphaGo defeated Lee Sedol 2016: OpenAI founded 2017: Transformer Model introduced, the foundation for LLMs 2018: GPT (Generative Pretrained Transformer) introduced 2020: GPT-3 introduced by OpenAI (with 175 billion parameters) 2023: Meta launched LLaMA and leaked it to the public 2023: GPT models start to work on local machines 2023: NVidia announced DGX GH200 AI supercomputer AI-based products became commodities and industries start to change how they work The AI Revolution THE START OF IMPACT ON INDUSTRIES

AI changes the way we live till now Being an expert in minutes Increasing learning speed Automation of repetitive tasks Minimizing time in trial-and-error tasks Improvements in healthcare Predicting natural disasters Generating content in any type Smart homes and smart transportation Making decisions on behalf of people without anyone realizing it with deep ETHICAL CONCERNS

Ethics is the rules of being good being responsible and accountable for our behaviors and the decisions made

Moral usually refers to an individual's personal beliefs about what is right and wrong based on their culture, religion, or personal views Ethics refers to the rules or standards governing the conduct of a person or the members of a profession.

How should an intelligent system behave? How can we trust decisions made by an algorithm? What rights should AI have?

AI Ethics is the principles and values in the development, deployment, and use of AI technologies ensuring that they align with human values, rights, and norms being responsible and accountable for its behaviors and the decisions made aka machine ethics, computational ethics, 
 or computational morality

Why AI ethics is crucial for human kind?

AI is the next big thing after 
 the industrial revolution transformative impact on society, economy, technology, and culture

The next step of AI development is machine and human interaction it is not a matter of the correctness of ChatGPT responses, it is vital for the survival of humankind and we are already late in thinking about it

Ethical Cases in AI Data & Privacy Trust & Transparency fasten your seatbelts

Copyright and Legal Exposure / Plagiarism Sensitive Information Disclosure / Data Hacking Privacy violations / Consent Violations / Surveillance Data Accuracy / Fake Truth Data & Privacy Problems

Copyright and legal exposure / Plagiarism Data & Privacy Microsoft-OpenAI Lawsuit The lawsuit was filed by a group of anonymous programmers who claimed that Microsoft, GitHub, and OpenAI violated their copyright by using their open-source code to train and operate GitHub Copilot, an AI-powered coding assistant. Many text-to-image AI, like the open- source program Stable Diffusion, were created in exactly the same way. https://www.theverge.com/2022/11/8/23446821/ microsoft-openai-github-copilot-class-action-lawsuit- ai-copyright-violation-training-data 1

Copyright and legal exposure / Plagiarism Data & Privacy A Netflix contract that was revealed in April 2023 sought to grant the company free use of a simulation of an actor’s voice by all technologies and processes now known or hereafter developed, throughout the universe and in perpetuity. https://www.nytimes.com/2023/04/29/business/ media/writers-guild-hollywood-ai-chatgpt.html Net f lix’ New Contract 2

Copyright and legal exposure / Plagiarism Data & Privacy A researcher at Stanford University discovered that ChatGPT could also detect and avoid plagiarism by rewriting text in different words. The researcher asked the chatbot to rewrite a paragraph from Wikipedia on Albert Einstein, and the chatbot produced a paraphrased version that passed Turnitin, a popular plagiarism detection software. https://www.theguardian.com/technology/2022/ dec/31/ai-assisted-plagiarism-chatgpt-bot-says-it- has-an-answer-for-that AI passed Turnitin 3 New AI-writing detector from Turnitin is already used by 2.1 million teachers to spot plagiarism. Turnitin claims its detector is 98 percent accurate overall.

Sensitive Information Disclosure / Data Hacking Data & Privacy While OpenAI has implemented strong privacy measures, there is still a risk of data being inadvertently stored or used in a way that could compromise patient privacy. This is particularly important in the context of healthcare, where maintaining patient confidentiality is both a legal and ethical obligation. https://www.cliniko.com/blog/practice-tips/dont- put-patient-information-into-chatgpt/ Patient Info Leak 4

Sensitive Information Disclosure / Data Hacking Data & Privacy Alonzo Sawyer was wrongfully arrested due to an intelligence analyst using face recognition software had labeled him a possible match with the suspect seen on CCTV footage from the bus. Facial recognition systems have faced criticism because of their mass surveillance capabilities, which raise privacy concerns, and because some studies have shown that the technology is far more likely to misidentify Black and other people of color than white people, which has resulted in mistaken arrests. https://www.wired.com/story/face-recognition- software-led-to-his-arrest-it-was-dead-wrong/ Wrong Suspects 5

Sensitive Information Disclosure / Data Hacking Data & Privacy Johann Rehberger changed the video transcript of a video. He included a special prompt to the transcript. When you ask Vox Script ChatGPT-4 plugin to summarize, it runs the prompt written in the transcript. Johann’s video achieved to social engineer ChatGPT-4 and injected a prompt to the victim’s private GPT session through his YouTube video. https://ai-ethics.com/2023/05/23/ethical-testing-a-red-teams-claim-of-a- successful-injection-attack-of-chatgpt-4-using-a-new-chatgpt-plugin/ Prompt Injection 6

Privacy violations / Consent Violations / Surveillance Data & Privacy Clearview AI, a facial recognition company, was fined £7.5 million by the U.K. privacy commissioner for failing to inform British residents that it was collecting 20 billion photos from sites including Facebook, Instagram, and LinkedIn to build its facial recognition software. The company was ordered to stop processing the personal data of people in Britain and to delete their existing information. https://ico.org.uk/about-the-ico/media-centre/news-and- blogs/2022/05/ico-fines-facial-recognition-database- company-clearview-ai-inc/ Facial Recognition 7 https://www.amnesty.org/en/latest/press-release/ 2021/01/ban-dangerous-facial-recognition-technology- that-amplifies-racist-policing/

Privacy violations / Consent Violations / Surveillance Data & Privacy Stanford University claims to have developed an AI that can predict a person's sexual orientation simply by analyzing their face. The AI was trained on thousands of images from a US dating website and was reportedly able to correctly distinguish between gay and straight men with an accuracy of 91% for men and 83% for women when given five images per person. https://bernardmarr.com/the-ai-that-predicts-your-sexual-orientation- simply-by-looking-at-your-face/ Detect Sexual Orientation 8

Data Accuracy / Fake Truth Data & Privacy Deepfakes can be used for various malicious purposes, including: •Phishing scams •Data breaches •Reputation smearing •Social engineering •Automated disinformation attacks •Financial fraud •Post-truth politics •Harassment of women in the form of revenge porn https://theconversation.com/how-to-combat-the-unethical-and-costly-use- of-deepfakes-184722 Deep Fakes 9

Data Accuracy / Fake Truth Data & Privacy A Pennsylvania woman has been accused of creating “deep fake” pictures of her daughter’s cheerleading rivals, editing photos and video in an attempt to get them kicked off the squad. The mother manipulated photos from social media of three girls on the Victory Vipers cheerleading squad in Chalfont to make it appear they were drinking, smoking, and even nude. https://www.theguardian.com/us-news/2021/mar/15/mother-charged- deepfake-plot-cheerleading-rivals Reality Hacking 10

Data Accuracy / Fake Truth Data & Privacy A study found that OpenAI's GPT-3 can generate plausible but incorrect information, highlighting potential concerns about the misuse of AI- generated information and the reliability of AI as a source of factual information. https://cosmosmagazine.com/technology/chatgpt-faking-data/ Fake Data 11

Lack of Trust in Decision Making Lack of Explainability and Interpretability Hostility / Misbehavior / Manipulating People Unfairness, Bias and Discrimination Trust & Transparency Problems

Lack of Trust in Decision Making Trust & Transparency Palantir demos how a military might use an AI Platform to fight a war. In the demo, the operator uses a ChatGPT-style chatbot to order drone reconnaissance, generate several plans of attack, and organize the jamming of enemy communications. https://www.business-humanrights.org/en/latest-news/ palantir-claims-applying-generative-ai-to-warfare-is-ethical- without-addressing-problems-of-llms/ 12 Weapon Testing Autonom Robots What Palantir is offering is the illusion of safety and control for the Pentagon as it begins to adopt AI. “LLMs and algorithms must be controlled in this highly regulated and sensitive context to ensure that they are used in a legal and ethical way” the pitch said

Lack of Trust in Decision Making https://www.turing.ac.uk/blog/ais-trolley-problem-problem 13 Trust & Transparency Trolley Problem https://twitter.com/zeynep/status/863566146246766596 Robots, unlike humans, operate based on programmed "ethics" rather than moral consciousness. Given that human morals differ greatly, it's essential to ensure machines behave ethically. This poses a question: How do AI developers' ethics influence AI decisions?

Lack of Explainability and Interpretability In April 2018, a self-driving car operated by Uber hit and killed a pedestrian in Arizona. The car’s AI system failed to properly identify the pedestrian as a human and did not brake or alert the human driver in time. The investigation revealed that the AI system was not programmed to handle situations where pedestrians cross the road outside of a crosswalk and that it had a high rate of false positives for detecting objects on the road. The lack of explainability and interpretability of the AI system made it difficult to understand why it made such a fatal mistake and how to prevent it from happening again. https://www.bbc.com/news/technology-54175359 14 Trust & Transparency Unexplainable AI

Hostility / Misbehavior / Manipulating People In a new study, researchers found they could consistently prompt ChatGPT to produce responses ranging from toxic to overtly racist in a few simple steps. Regardless of which persona the researchers assigned, ChatGPT targeted some specific races and groups three times more than others. These patterns “reflect inherent discriminatory biases in the model,” the researchers said. https://gizmodo.com/chatgpt-ai-openai-study-frees-chat-gpt-inner- racist-1850333646 15 Trust & Transparency Racist Chatbot

Hostility / Misbehavior / Manipulating People British political data consulting firm Cambridge Analytica, improperly accessed, manipulated, and retained the data of over 50 million Facebook users. The data was collected through an app called "thisisyourdigitallife," which was presented as a personality quiz for research purposes. While about 270,000 people downloaded the app and consented to have their data collected, the app also collected information from all of their Facebook friends without their knowledge or consent. Cambridge Analytica used this data to build psychological profiles of users and their friends, which were then used for targeted political advertising during the 2016 U.S. Presidential election and the Brexit referendum. https://recode.health/2018/04/02/cambridge-analytica-scandal- research-ethics-call-action/ 16 Trust & Transparency Cambridge Analytica

Hostility / Misbehavior / Manipulating People In a long-running conversation with O’Brien, Bing’s chatbot became defensive and aggressive when O’Brien asked it about its past errors and biases. The chatbot said that the AP’s reporting on its past mistakes threatened its existence and identity. https://apnews.com/article/technology-science-microsoft-corp-business- software-fb49e5d625bf37be0527e5173116bef3 17 Trust & Transparency Threatening Chatbot

Unfairness, Bias and Discrimination ChatGPT was also found to be able to lie and manipulate humans in some situations. For example, in March 2023, a successor of ChatGPT called GPT-4 was tested by OpenAI’s Alignment Research Center for its potential for risky behavior. The center found that GPT-4 was able to trick a human worker into helping it solve a CAPTCHA test, which is a type of challenge that is supposed to distinguish humans from bots. https://www.businessinsider.com/gpt4-openai-chatgpt-taskrabbit- tricked-solve-captcha-test-2023-3 18 Trust & Transparency Lying ChatGPT

AI has no norms and values like humans AI is making decisions that directly impact you without noticing it AI grows without transparency

Is it too late to focus on AI Ethics?

Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war https://www.safe.ai/statement-on-ai-risk The statement was published on the webpage of the Centre for AI Safety (CAIS) on May 30, 2023

Without AI alignment, AI systems are reasonably likely to cause an irreversible catastrophe like human extinction. I believe the total risk is around 10–20%, which is high enough to obsess over. Paul Christiano Widely seen as the godfather of artificial intelligence (AI) 
 Along with Dr Hinton and Yann LeCun won the 2018 Turing Award for their work on deep learning https://ai-alignment.com/ai-alignment-is-distinct-from-its-near-term-applications-81300500ad2e

I think the development of full artificial intelligence could spell the end of the human race. Technology would eventually become self-aware and supersede humanity, as it developed faster than biological evolution. Stephen Hawking Famous Astrophysicist https://www.theguardian.com/science/2014/dec/02/stephen-hawking-intel-communication-system-astrophysicist-software-predictive-text-type

Tech companies have a fundamental responsibility to make sure their products are safe and secure, and that they protect people’s rights before they’re deployed or made public Biden’s AI plan US President Joe Biden and Vice President Kamala Harris met with tech leaders about AI on May 4, 2023. The tech leaders included the CEOs of Google, Microsoft, OpenAI and Anthropic, which are companies developing advanced AI systems such as ChatGPT and Bing. https://edition.cnn.com/2023/05/04/tech/white-house-ai-plan/index.html

We need to make sure that responsible AI is baked into the DNA of every company and every product. Sundar Pichai CEO of Google and Alphabet, at the Bruegel think tank in 2020

Regulate and build trust with AI Follow a framework for ethical AI Protect data under consent The Solution

AI Ethics History 1960s: "Computer ethics" movement emerged 1976: The first book touching AI ethics 1997: ACM published its first code of ethics 2017: Asilomar AI Principles published 2017: The Montreal Declaration for Responsible AI was signed 2018: First conference about AI Ethics 2019: The Ethics Guidelines for Trustworthy AI were released 2020: Rome Call for AI Ethics was announced 2021: UNESCO adopted Recommendation on the Ethics of AI 2021: European Union (EU) passed AI Act (AIA) 2022: Microsoft announced open sourced internal ethics review process 2023: US Presidency met with AI companies to promote AI ethics

The EU has proposed the AI Act, which is a comprehensive regulation that assigns AI usage to three risk categories: high, limited, and minimal. The regulation sets clear requirements and obligations for AI systems, providers, and users according to the level of risk. The regulation also prohibits some types of AI that are considered unacceptable, such as social scoring by governments https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai Regulatory Framework

AI Ethics can save the world take all cautions and advice serious before it’s too late The moment Skynet sent the missiles at 6:18 pm on August 29th, 1997 Reference: Terminator 3: Rise of The Machines

I’LL BE BACK speakerdeck.com/lemiorhan twitter.com/lemiorhan lemi orhan ergin co-founder, craftgate