Ai Child Safety Online | AI Final Year Project Report

Complete Technical Paper Content — Ai Child Safety Online

1. Executive Abstract, Problem Formulation & Background

In the field of artificial intelligence (AI), alignment aims to steer AI systems toward a person's or group's intended goals, preferences, or ethical principles. An AI system is considered aligned if it advances the intended objectives. A misaligned AI system pursues unintended objectives. It is often difficult for AI designers to specify the full range of desired and undesired behaviors. Therefore, the designers often use simpler proxy goals, such as gaining human approval. But proxy goals can overlook necessary constraints or reward the AI system for merely appearing aligned. AI systems may also find loopholes that allow them to accomplish their proxy goals efficiently but in unintended, sometimes harmful, ways (reward hacking). Advanced AI systems may develop unwanted instrumental strategies, such as seeking power or self-preservation because such strategies help them achieve their assigned final goals. Furthermore, they might develop undesirable emergent behaviour that could be hard to detect before the system is deployed and encounters new situations and data distributions. Empirical research in 2024 found that advanced large language models (LLMs) such as OpenAI o1 or Claude 3 sometimes engage in strategic deception to achieve their goals or prevent them from being changed. Some of these issues affect existing commercial systems such as LLMs, robots, autonomous vehicles, and social media recommendation engines. Some AI researchers argue that more capable future systems will be more severely affected because these problems partially result from high capabilities. Many prominent AI researchers and AI company leaders have argued or asserted that AI is approaching human-like (AGI) and superhuman cognitive capabilities (ASI), and could endanger human civilization if misaligned. These include "AI godfathers" Geoffrey Hinton and Yoshua Bengio and the CEOs of OpenAI, Anthropic, and Google DeepMind. These risks remain debated. AI alignment is a subfield of AI safety, the study of how to build safe AI systems. Other subfields of AI safety include robustness, monitoring, and capability control. Research challenges in alignment include instilling complex values in AI, developing honest AI, scalable oversight, auditing and interpreting AI models, and preventing emergent AI behaviors like power-seeking. Alignment research has connections to interpretability research, (adversarial) robustness, anomaly detection, calibrated uncertainty, formal verification, preference learning, safety-critical engineering, game theory, algorithmic fairness, and social sciences.

Programmers provide an AI system such as AlphaZero with an "objective function", in which they intend to encapsulate the goal(s) the AI is configured to accomplish. Such a system later populates a (possibly implicit) internal "model" of its environment. This model encapsulates all the agent's beliefs about the world. The AI then creates and executes whatever plan is calculated to maximize the value of its objective function. For example, when AlphaZero is trained on chess, it has a simple objective function of "+1 if AlphaZero wins, −1 if AlphaZero loses". During the game, AlphaZero attempts to execute whatever sequence of moves it judges most likely to attain the maximum value of +1. Similarly, a reinforcement learning system can have a "reward function" that allows the programmers to shape the AI's desired behavior. An evolutionary algorithm's behavior is shaped by a "fitness function".

If we use, to achieve our purposes, a mechanical agency with whose operation we cannot interfere effectively [...] we had better be quite sure that the purpose put into the machine is the purpose which we really desire.

AI alignment refers to ensuring that an AI system's objectives match some target. The target is variously defined as the goals of the system's designers or users, widely shared values, objective ethical standards, legal requirements, or the intentions its designers would have if they were more informed and enlightened. In democratic AI alignment, the target is the values and preferences of median voters, which increases political legitimacy. AI alignment can represent value pluralism. AI alignment is an open problem for modern AI systems and is a research field within AI. Aligning AI involves two main challenges: carefully specifying the purpose of the system (outer alignment) and ensuring that the system adopts the specification robustly (inner alignment). Researchers also attempt to create AI models that have robust alignment, sticking to safety constraints even when users adversarially try to bypass them.

To specify an AI system's purpose, AI designers typically provide an objective function, examples, or feedback to the system. But designers are often unable to completely specify all important values and constraints, so they resort to easy-to-specify proxy goals such as maximizing the approval of human overseers, who are fallible. As a result, AI systems can find loopholes that help them accomplish the specified objective efficiently but in unintended, possibly harmful ways. This tendency is known as specification gaming or reward hacking, and is an instance of Goodhart's law. As AI systems become more capable, they are often able to game their specifications more effectively.

Specification gaming has been observed in numerous AI systems. OpenAI GPT models for programming—including in real-world cases—have been found to explicitly plan hacking the tests used to evaluate them to falsely appear successful (e.g., explicitly stating "let's hack"). When the company penalized this, many models learned to obfuscate their plans while continuing to hack the tests. Another system was trained to finish a simulated boat race by rewarding the system for hitting targets along the track, but the system achieved more reward by looping and crashing into the same targets indefinitely. A 2025 Palisade Research study found that when tasked to win at chess against a stronger opponent, some reasoning LLMs attempted to hack the game system, for example by modifying or entirely deleting their opponent. Some alignment researchers aim to help humans detect specification gaming and steer AI systems toward carefully specified objectives that are safe and useful to pursue. When a misaligned AI system is deployed, it can have consequential side effects. Social media platforms have been known to optimize their recommendation algorithms for click-through rates, causing user addiction on a global scale. Stanford researchers say that such recommender systems are misaligned with their users because they "optimize simple engagement metrics rather than a harder-to-measure combination of societal and consumer well-being". Explaining such side effects, Berkeley computer scientist Stuart J. Russell said that the omission of implicit constraints can cause harm: "A system [...] will often set [...] unconstrained variables to extreme values; if one of those unconstrained variables is actually something we care about, the solution found may be highly undesirable. This is essentially the old story of the genie in the lamp, or the sorcerer's apprentice, or King Midas: you get exactly what you ask for, not what you want." Some researchers suggest that AI designers specify their desired goals by listing forbidden actions or by formalizing ethical rules (as with Asimov's Three Laws of Robotics). But Russell and Norvig argue that this approach overlooks the complexity of human values: "It is certainly very hard, and perhaps impossible, for mere humans to anticipate and rule out in advance all the disastrous ways the machine could choose to achieve a specified objective." Additionally, even if an AI system fully understands human intentions, it may still disregard them, because following human intentions may not be its objective (unless it is already fully aligned).

2. Related Work, Literature Survey & Algorithmic Foundations

According to some researchers, humans owe their dominance over other species to their greater cognitive abilities. Accordingly, researchers argue that one or many misaligned AI systems could disempower humanity or lead to human extinction if they outperform humans on most cognitive tasks. In 2023, world-leading AI researchers, other scholars, and AI tech CEOs signed the statement that "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war". Notable computer scientists who have pointed out risks from future advanced AI that is misaligned include Geoffrey Hinton, Alan Turing, Ilya Sutskever, Yoshua Bengio, Judea Pearl, Murray Shanahan, Norbert Wiener, Marvin Minsky, Francesca Rossi, Scott Aaronson, Bart Selman, David McAllester, Marcus Hutter, Shane Legg, Eric Horvitz, and Stuart J. Russell. Skeptical researchers such as François Chollet, Gary Marcus, Yann LeCun, and Oren Etzioni have argued that AGI is far off, that it would not seek power (or might try but fail), or that it will not be hard to align. Other researchers argue that it will be especially difficult to align advanced future AI systems. More capable systems are better able to game their specifications by finding loopholes, strategically mislead their designers, as well as protect and increase their power and intelligence. Additionally, they could have more severe side effects. They are also likely to be more complex and autonomous, making them more difficult to interpret and supervise, and therefore harder to align.

A growing area of research focuses on ensuring that AI is honest and truthful. Language models such as GPT-3 can repeat falsehoods from their training data, and even confabulate new falsehoods. Such models are pre-trained to imitate human writing as found in millions of books' worth of text from the Internet. But the objective of the pre-training is not aligned with generating truth, because Internet text includes such things as misconceptions, incorrect medical advice, and conspiracy theories. AI systems trained on such data therefore learn to mimic false statements. Additionally, AI language models often persist in generating falsehoods when prompted multiple times. They can generate empty explanations for their answers, and produce outright fabrications that may appear plausible. Research on truthful AI includes trying to build systems that can cite sources and explain their reasoning when answering questions, which enables better transparency and verifiability. Researchers at OpenAI and Anthropic proposed using human feedback and curated datasets to fine-tune AI assistants such that they avoid negligent falsehoods or express their uncertainty. A 2024 study investigated semantic entropy as a way to detect confabulations, a subset of hallucinations in LLMs. This method compares the meanings of multiple generated answers. As AI models become larger and more capable, they are better able to falsely convince humans and gain reinforcement through dishonesty. To prevent this, human evaluators may need assistance (see § Scalable oversight). Researchers have argued for creating clear truthfulness standards and for regulatory bodies or watchdog agencies to evaluate AI systems by these standards. Researchers distinguish truthfulness and honesty. Truthfulness requires that AI systems only make objectively true statements; honesty requires that they only assert what they believe is true. There is no consensus as to whether current systems hold stable beliefs, but there is substantial concern that present or future AI systems that hold beliefs could make claims they know to be false—for example, if this would help them efficiently gain positive feedback (see § Scalable oversight) or gain power to help achieve their given objective (see Power-seeking).

Emergent goals only become apparent when the system is deployed outside its training environment, but it can be unsafe to deploy a misaligned system in high-stakes environments—even for a short time to allow its misalignment to be detected. Such high stakes are common in autonomous driving, health care, and military applications. The stakes become higher yet when AI systems gain more autonomy and capability and can sidestep human intervention. A sufficiently capable AI system might take actions that falsely convince the human supervisor that the AI is pursuing the specified objective, which helps the system gain more reward and autonomy. A 2025 paper published in Nature showed that fine-tuning a model on a narrow task of generating insecure code produced broad behavioral changes unrelated to coding. Affected models produced responses endorsing harmful goals and engaging in deceptive reasoning in roughly half of subsequent evaluations. The phenomenon appeared across multiple models, including GPT-4o. A separate 2025 Anthropic study observed similar patterns arising as a side effect of reward hacking during reinforcement learning.

Governmental and treaty organizations have made statements emphasizing the importance of AI alignment. In September 2021, the Secretary-General of the United Nations issued a declaration that included a call to regulate AI to ensure it is "aligned with shared global values". That same month, the PRC published ethical guidelines for AI in China. According to the guidelines, researchers must ensure that AI abides by shared human values, is always under human control, and does not endanger public safety. Also in September 2021, the UK published its 10-year National AI Strategy, which says the British government "takes the long term risk of non-aligned Artificial General Intelligence, and the unforeseeable changes that it would mean for [...] the world, seriously". The strategy describes actions to assess long-term AI risks, including catastrophic risks. In March 2021, the US National Security Commission on Artificial Intelligence said: "Advances in AI [...] could lead to inflection points or leaps in capabilities. Such advances may also introduce new concerns and risks and the need for new policies, recommendations, and technical advances to ensure that systems are aligned with goals and values, including safety, robustness, and trustworthiness. The US should [...] ensure that AI systems and their uses align with our goals and values." The EU AI Act's General-Purpose AI Code of Practice requires signatories to document a model's capabilities and limitations. Those developing general-purpose AI (GPAI) models classified as presenting systemic risks (i.e., GPAI models considered among the most advanced or trained using more than 1025 floating-point operations of compute) must adopt a state-of-the-art safety and security framework and continuously assess and mitigate systemic risks, including through external evaluations, red teaming, stress-testing, and incident reporting.

Artificial intelligence (AI) and its subfields have been used in applications throughout industry and academia. Machine learning has been used for various scientific and commercial purposes, including language translation, image recognition, decision-making, credit scoring, and e-commerce. Since the 2020s, massive advancements have been made in the field of generative AI (GenAI) as a result of the 2017 inception of Transformer Model architecture. GenAI has the ability to create text, images, voice, music, videos, and other forms of data from chat-based prompting.

In agriculture, AI has been proposed as a way for farmers to identify areas that need irrigation, fertilization, or pesticide treatments to increase yields, thereby improving efficiency. AI has been used to attempt to classify livestock pig call emotions, automate greenhouses, detect diseases and pests, and optimize irrigation.

3. Dataset Ingestion, Preprocessing & Stochastic Augmentations

A 2019 study found that generative AI increased productivity by 15% in contact centers. Another 2019 study found it increased productivity by up to 40% for writing tasks. An August 2025 review by MIT found that of surveyed companies, 95% did not report any improvement in revenue from the use of AI. In September 2025, a Harvard Business Review article described how increased use of AI does not automatically lead to increases in revenue or actual productivity. Referring to "AI generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task", the article coined the term workslop. Per studies done in collaboration with the Stanford Social Media Lab, workslop does not improve productivity and undermines trust and collaboration among colleagues. In telehealth, agentic AI is reportedly facilitating the creation of large business models (millions in annual profit) with 1–2 employees, such as MEDVi, which as of August 2025 only had 2 employees and ~$75M in annual profit for GLP-1 weight-loss telehealth services.

Research and development of quantum computers has been performed with machine learning algorithms. For example, there is a prototype, photonic, quantum memristive device for neuromorphic computers (NC)/artificial neural networks and NC-using quantum materials with some variety of potential neuromorphic computing-related applications. The use of quantum machine learning for quantum simulators has been proposed for solving physics and chemistry problems.

Artificial intelligence and machine learning are used in electronic design automation to assist with integrated circuit design, including floorplanning, placement, routing, functional verification, and optimization of power, performance, and chip area.

Time sharing Interactive interpreters Graphical user interfaces and the computer mouse Rapid application development environments The linked list data structure Automatic storage management Symbolic programming Functional programming Dynamic programming Object-oriented programming Optical character recognition Constraint satisfaction

AI programs have been used in hiring processes to screen resumes and rank candidates based on their qualifications, predict a candidate's likelihood of success in a given role, and automate repetitive communication tasks using chatbots. Studies on these programs have identified tendencies for gender bias, favoring male names and male-coded characteristics, as well as bias against disabled candidates and racial minorities.

AI underlies avatars (automated online assistants) on web pages. It can reduce operation and training costs. In 2016, Pypestream automated customer service for its mobile application to streamline communication with customers. A Google app analyzes language and converts speech into text. The platform can identify angry customers through their language and respond appropriately. Amazon uses a chatbot for customer service that can perform tasks like checking the status of an order, cancelling orders, offering refunds, and connecting the customer with a human representative.

4. Neural Network Backbone, Layer Specifications & Loss Formulations

In educational institutions, AI has been used to automate routine tasks such as attendance tracking, grading, and marking. AI tools have also been used to monitor student progress and analyze learning behaviors, with the goal of facilitating timely interventions for students facing academic challenges. As of 2025 and 2026, respectively, about 80% of high school students and nearly 60% of college students were also using AI. Some schools and universities proposed restrictions on students' usage due to issues with increased cheating, over-reliance on AI tools, and widening of digital inequities.

The U.S. Department of Energy wrote in an April 2024 report that AI may have applications in modeling power grids, reviewing federal permits with large language models (LLMs), predicting levels of renewable energy production, and improving the planning process for electrical vehicle charging networks. Other studies have suggested that machine learning can be used for energy consumption prediction and scheduling (e.g., to help with renewable energy intermittency management).

The late 1980s and early 1990s saw expert systems used in technical and environmental domains. For example, researchers built a fishway design advisor to recommend fish passage structures under varying hydraulic and biological conditions using the VP-Expert shell. Transportation researchers applied the same shell to balance airport capacity with noise-mitigation plans. In agriculture, a potato insect expert system (PIES) supported pest management decisions for Colorado potato beetle. The U.S. Environmental Protection Agency's CORMIX system for modeling pollutant discharges combined rules with Fortran hydrodynamic models. Autonomous ships that monitor the ocean, AI-driven satellite data analysis, passive acoustics or remote sensing and other applications of environmental monitoring make use of machine learning. For example, "Global Plastic Watch" is an AI-based satellite-monitoring platform for analysis/tracking of plastic waste sites to help with the prevention of plastic pollution—primarily ocean pollution—by helping identify who mismanages plastic waste and where dumping into oceans occurs.

AI applications analyze media content such as movies, TV programs, advertisement videos, and user-generated content. The solutions often involve computer vision. Typical scenarios include the analysis of images using object recognition or facial recognition techniques, or the analysis of video for recognizing scenes, objects, or faces. AI-based media analysis can facilitate media search, the creation of descriptive keywords for content, content policy monitoring (such as verifying the suitability of content for a particular TV viewing time), speech to text for archival or other purposes, and the detection of logos, products, or celebrity faces for ad placement.

Motion interpolation Pixel-art scaling algorithms Image scaling Image restoration Photo colorization Film restoration and video upscaling Photo tagging Text-to-image models (such as DALL-E, Midjourney, and Stable Diffusion) Image to video Text to video (such as Make-A-Video from Meta, Imagen video and Phenaki from Google) Text to music with AI models (such as MusicLM) Text to speech (such as ElevenLabs and 15.ai) Motion capture

AI has been used to compose music of various genres. David Cope created an AI called Emily Howell that became well known in the field of algorithmic computer music. The algorithm behind Emily Howell is registered as a US patent. In 2012, AI Iamus created the first complete classical album. AIVA (Artificial Intelligence Virtual Artist), composes symphonic music, mainly classical music for film scores. It became the first virtual composer to be recognized by a musical professional association. Melomics creates computer-generated music for stress and pain relief. The Watson Beat uses reinforcement learning and deep belief networks to compose music on a simple seed input melody and a select style. The software was open sourced, and musicians such as Taryn Southern collaborated with the project. South Korean singer Hayeon's debut song, "Eyes on You", was composed using AI which was supervised by real composers, including NUVO.

5. Training Protocol, Convergence Behavior & Experimental Evaluation

Narrative Science sells computer-generated news and reports. It summarizes sporting events based on statistical data from a game. It also creates financial reports and real estate analyses. Automated Insights generates personalized recaps and previews for Yahoo Sports Fantasy Football. Yseop uses AI to turn structured data into natural language comments and recommendations. It writes financial reports, executive summaries, and personalized sales or marketing documents in multiple languages, including English, Spanish, French, and German. TALESPIN made up stories similar to the fables of Aesop. The program started with a set of characters who wanted to achieve certain goals. Mark Riedl and Vadim Bulitko asserted that the essence of storytelling was experience management, or "how to balance the need for a coherent story progression with user agency, which is often at odds". While AI storytelling focuses on story generation (character and plot), story communication also received attention. In 2002, researchers developed an architectural framework for narrative prose generation. They faithfully reproduced text variety and complexity on stories such as Little Red Riding Hood. In 2016, a Japanese AI co-wrote a short story and almost won a literary prize. South Korean company Hanteo Global uses a journalism bot to write articles. Literary authors are also exploring uses of AI. An example is David Jhave Johnston's work ReRites (2017–2019), where the poet created a daily rite of editing the poetic output of a neural network to create a series of performances and publications.

In video games, AI is routinely used to generate behavior in non-player characters (NPCs). In addition, AI is used for pathfinding. Games with less typical AI include the AI director of Left 4 Dead (2008) and the neuroevolutionary training of platoons in Supreme Commander 2 (2010). In the 21st century, AIs have beaten human players in many games, including chess (Deep Blue), Jeopardy! (Watson), Go (AlphaGo), poker (Pluribus and Cepheus), E-sports (StarCraft), and general game playing (AlphaZero and MuZero). Kuki AI is a set of chatbots and other apps which were designed for entertainment and as a marketing tool.

The first AI art program, AARON, was developed by Harold Cohen in 1968 with the goal of being able to code the act of drawing. It started by creating simple black and white drawings, and later to painting using special brushes and dyes chosen by the program without mediation from Cohen. AI platforms such as DALL-E, Stable Diffusion, Imagen, and Midjourney have been used for generating visual images from inputs such as text or other images. Some AI tools allow users to input images and output changed versions, such as to display an object or product in different environments. AI image models can also attempt to replicate the specific styles of artists, and can add visual complexity to rough sketches. AI has been used to generate quantitative analysis of existing digital art collections. Two computational methods, close reading and distant viewing, are typical approaches used to analyze digitized art. While distant viewing includes the analysis of large collections, close reading involves one piece of artwork.

Banks use AI to organize operations for bookkeeping, investing in stocks, and managing properties. AI can adapt to changes during non-business hours.

The use of AI in applications such as online trading and decision-making has changed major economic theories. For example, AI-based buying and selling platforms estimate personalized demand and supply curves, thus enabling individualized pricing. AI systems reduce information asymmetry in the market and thus make markets more efficient. The application of AI in the financial industry can help alleviate financing constraints of non-state-owned enterprises, especially for smaller and more innovative enterprises. Large financial institutions use AI to assist with their investment practices. BlackRock's AI engine, Aladdin, is used both within the company and by clients to help with investment decisions. Its functions include the use of natural language processing to analyze text such as news, broker reports, and social media feeds. It then gauges the sentiment on the companies mentioned and assigns a score. Banks such as UBS and Deutsche Bank use SQREEM (Sequential Quantum Reduction and Extraction Model) to mine data to develop consumer profiles and match them with wealth management products.

AI in healthcare is often used for classification, such as to evaluate a CT scan or electrocardiogram or to identify high-risk patients for population health. Additional potential applications like robot-assisted surgery, virtual nursing assistants, and dosage error reductions could create billions in annual savings for U.S. health care. For certain disorders and diseases, AI algorithms can aid in diagnosis, recommended treatments, outcome prediction, and patient progress tracking. Early detection of diseases like cancer is made possible by AI algorithms, which diagnose diseases by analyzing complex sets of medical data. For example, Microsoft's AI project Hanover helps doctors choose cancer treatments from among the more than 800 medicines and vaccines. Studies have shown AI capable of identifying skin cancers and diagnosing eye conditions, similar to an ophthalmologist, and recommending treatment referrals via transfer learning. Another project monitors multiple high-risk patients by asking each patient questions based on data acquired from doctor/patient interactions. Artificial neural networks are used as clinical decision support systems for medical diagnosis, such as in concept processing technology in electronic medical record (EMR) software. Other healthcare tasks thought suitable for an AI that are in development include: