#aisystems

[lasseling]

“Maladaptive Traits”: AI Systems Are Learning To Lie And Deceive

A new study has found that AI systems known as large language models (LLMs) can exhibit “Machiavellianism,” or intentional and amoral manipulativeness, which can then lead to deceptive behavior.

[waybackwanderer]

undefined

Ai Systems - Web Links Oct 1997 Archived Web Page 🧩

[govindhtech]

How IBM’s Smarter Balanced Is Governing Education AI

Smarter Balanced Assessment Consortium

The Smarter Balanced Assessment Consortium, a member-led public organization with headquarters in California, offers assessment tools to teachers in K–12 and higher education. Founded in 2010, the business creates creative, standards-aligned test assessment systems in collaboration with state education organizations. In order to assist educators in identifying learning opportunities and enhancing student learning, Smarter Balanced provides them with lessons, tools, and resources, such as formative, interim, and summative assessments.

In the constantly evolving field of education, Smarter Balanced is dedicated to progress and creativity. The objective is to investigate a systematic methodology for utilizing artificial intelligence (AI) in educational assessments in conjunction with IBM Consulting. The partnership is still in place as of early 2024.

Specifying the difficulty

Standardized exams and structured quizzes, which are common K–12 skill evaluations, are criticized for a number of equity-related reasons. AI has the revolutionary potential to improve assessment fairness across student populations, including marginalized groups, by providing individualized learning and assessment experiences when used responsibly. Therefore, defining what responsible AI adoption and governance in a school setting looks like is the main difficulty.

Educators, professionals in artificial intelligence, ethics and policy surrounding AI, and specialists in educational measurement made up the first multidisciplinary advisory group established by Smarter Balanced and IBM Consulting. The panel’s objective is to create guiding principles for integrating justice and accuracy into the application of AI to learning materials and educational measurement. Below is a summary of some of the advisory panel’s factors.

Considering human needs when designing

Organizations can create a human-centric strategy for implementing technology by utilizing design thinking frameworks. Design thinking is driven by three human-centered principles: a focus on user outcomes, restless reinvention, and team empowerment for diversity. Stakeholders’ strategic alignment and responsiveness to both functional and non-functional organizational governance requirements are enhanced by this approach. Developers and other stakeholders can generate creative solutions, prototype iteratively, and gain a thorough understanding of user demands by applying design thinking.

This methodology is critical for early risk identification and assessment during the development process, as well as for enabling the development of reliable and efficient AI models. Design thinking aids in the development of AI solutions that are mathematically sound, socially conscious, and human-centered by consistently interacting with various communities of domain experts and other stakeholders and taking their input into consideration.

Including Diversity

A varied group of subject-matter experts and thought leaders were assembled by the merged teams to form a think tank for the Smarter Balanced initiative. Experts in the domains of law and educational evaluation, as well as neurodivergent individuals, students, and those with accessibility issues, made up this group.

The think tank aims to iteratively, rather than one-time, integrate its members’ experiences, opinions, and areas of expertise into the governance framework. A fundamental tenet of IBM’s AI ethics is reflected in the strategy: artificial intelligence should supplement human intelligence, not replace it. Incorporating continuous feedback, assessment, and examination by a range of stakeholders can enhance the development of trust and facilitate fair results, ultimately resulting in an educational setting that is more comprehensive and productive.

In grade school settings, these approaches are essential for developing equitable and successful educational assessments. Building AI models that are reflective of all students requires the many perspectives, experiences, and cultural insights that diverse teams bring to the table. Because of its inclusivity, AI systems are less likely to unintentionally reinforce existing disparities or fail to take into account the particular demands of various demographic groups. This highlights another important AI ethical tenet at IBM: diversity in AI is important since it’s about math, not opinion.

Examining beliefs that are focused on the student

Determining the human values IBM want to see represented in AI models was one of the first projects that IBM Consulting and Smarter Balanced performed together. IBM arrived at a set of principles and criteria that correspond to IBM’s AI pillars, or essential characteristics for reliable AI, as this is not a novel ethical dilemma.

Explainability: The capacity to provide results and functions that don’t require technical explanation

Fairness: Handling individuals equally

Robustness: security, dependability, and ability to withstand hostile attacks

Openness: Sharing information about the use, functionality, and data of AI

Data privacy: revealing and defending users’ rights to their privacy and data

It is difficult to put these ideas into practice in any kind of organization. Even higher standards apply to an organization that evaluates pupils’ skill sets. Nonetheless, the work is valuable due to the potential advantages of AI. The second phase is now in progress and involves investigating and defining the values that will direct the use of AI to the assessment of young learners.

The teams are debating the following questions:

What morally-based guidelines are required to properly develop these skills?

Who should be in charge of operationalizing and governing them?

What guidance should practitioners providing these models with follow?

What are the essential needs, both functional and non-functional, and what is the required strength?

Investigating varying impact and levels of affect

IBM used the Layers of Effect design thinking framework for this activity. IBM Design for AI has donated numerous frameworks to the open source community Design Ethically. Stakeholders are asked to think about the primary, secondary, and tertiary implications of their experiences or goods using the Layers of Effect framework.

The intended and known impacts of the product in this case, an AI model are referred to as primary effects. One of the main functions of a social media platform, for instance, could be to link people with shared interests.

Although less deliberate, secondary impacts can swiftly gain importance among stakeholders. Using social media as an example, the platform’s value to advertisers may have a secondary effect.

Unintentional or unexpected consequences that show up gradually are known as tertiary effects. An example of this would be a social media platform’s propensity to provide more views to messages that are insulting or misleading.

The main (desired) consequence of the AI-enhanced test assessment system for this use case is a more effective, representative, and equitable tool that raises learning outcomes throughout the educational system.

Increasing efficiency and obtaining pertinent data to aid in more effective resource allocation where it is most required are possible secondary benefits.

Unintentional and possibly recognized tertiary effects exist. Stakeholders need to investigate what would constitute unintentional harm at this point.

The groups determined that there could be five types of serious harm:

Detrimental prejudice concerns that fail to take into account or assist pupils from marginalized groups who might require additional resources and viewpoints to meet their varied needs.

Problems with personally identifiable information (PII) and cybersecurity in educational systems where insufficient protocols are in place for their networks and devices.

Insufficient governance and regulations to guarantee AI models maintain their intended behaviors.

Inadequate communication regarding the planned usage of AI systems in schools with parents, students, instructors, and administrative staff. These messages ought to outline agency, like how to opt out, and safeguards against improper usage.

Restricted off-campus connectivity that could limit people’s access to technology and the use of AI that follows, especially in rural locations.

Disparate impact evaluations, which were first used in court cases, assist organizations in identifying possible biases. These evaluations look at how people from protected classes those who are vulnerable to discrimination on the basis of gender, ethnicity, religion, or other characteristics can be disproportionately impacted by policies and practices that appear to be neutral. These evaluations have shown to be useful in the formulation of employment, financing, and healthcare policies. IBM tried to take into account cohorts of students in their education use case who might, because of their circumstances, receive unequal results from tests.

The following were the categories found to be most vulnerable to possible harm:

People who experience mental health issues

People from a wider range of socioeconomic backgrounds, including those without a place to live

Individuals whose mother tongue is not English

Those with additional non-linguistic cultural factors

People with accessibility concerns or those who are neurodivergent

IBM group’s next series of exercises is to investigate ways to lessen these harms by utilizing additional design thinking frameworks, like ethical hacking. IBM will also go over the minimal specifications that companies looking to integrate AI into student assessments must meet.

Read more on govindhtech.com

[sifytech]

Convergence Digital and Real - Is It Good

If we give algorithms total control over our decisions, they can influence what we eat and how we behave, our choices may be influenced by an entity we cannot control. Read More. https://www.sify.com/ai-analytics/convergence-digital-and-real-is-it-good/

[thxnews]

Facebook and Instagram Enhance User Control and Transparency in Content Ranking

  Enhancing User Control and Transparency

In a recent announcement, Facebook and Instagram unveiled significant updates to empower users with more control over their content experience and provide greater transparency into the algorithms shaping their feeds. The updates aim to make the platforms more user-friendly and address concerns regarding algorithmic influence. These changes come as billions of people rely on Facebook and Instagram to connect, share their lives, and discover captivating content.   Empowering Users with AI Systems Understanding the importance of personalization, both platforms utilize AI systems to curate content tailored to each user's preferences. By factoring in user choices and behavior, these systems attempt to deliver relevant and engaging content. In a prior discussion, Meta, the parent company of Facebook and Instagram, acknowledged the need for more transparency and user control, challenging the notion that algorithms render users powerless. Building on that commitment, Meta now takes strides toward openness and control.   Increased Transparency and Control Facebook and Instagram are committed to providing users with more transparency regarding AI systems that rank content across the platforms. By releasing 22 system cards, Meta grants insights into how these AI systems operate, the predictions they make to determine content relevance, and the available controls to customize the user experience. These system cards cover various sections such as Feed, Stories, Reels, and even unconnected content recommendations. Users can access the Transparency Center for a more detailed explanation of content recommendation AI. Moreover, Meta goes beyond system cards by sharing the types of signals and predictive models used to determine content relevance in the Facebook Feed. While the company aims to be transparent, it also recognizes the need to balance disclosure with safeguarding against misuse.   Personalizing the User Experience Recognizing that users have different preferences, Facebook and Instagram have centralized controls to customize content exposure. The Feed Preferences on Facebook and the Suggested Content Control Center on Instagram provide users with the ability to influence the content they see. Additionally, features like "Interested" and "Not Interested" on Instagram's Reels tab allow users to indicate their preferences and receive more of the content they enjoy. Facebook's "Show more, Show less" feature further empowers users to fine-tune their content consumption. For users desiring a more chronological feed experience, the Feeds tab on Facebook and the Following section on Instagram offer alternatives. Users can also create a Favorites list to ensure they never miss content from their favorite accounts.   Enabling Research and Innovation Meta believes in fostering openness and collaboration in the field of research and innovation, particularly regarding transformative AI technologies. Over the past decade, the company has released over 1,000 AI models, libraries, and data sets to support academic and public interest research. In the coming weeks, Meta will introduce the Content Library and API, offering comprehensive access to publicly-available content from Facebook and Instagram. Researchers from qualified institutions can apply for access, fostering scientific exploration while meeting new data-sharing and transparency obligations. By involving researchers early in the development process, Meta aims to receive valuable feedback, ensuring the tools align with their needs and aspirations. Facebook and Instagram's commitment to user control, transparency, and research collaboration signifies a forward-thinking approach, emphasizing the importance of customization and understanding in the ever-evolving landscape of social media platforms.   Sources: THX News & Meta. Read the full article

[placement-india]

𝐖𝐡𝐲 𝐀𝐈 𝐖𝐢𝐥𝐥 𝐍𝐞𝐯𝐞𝐫 𝐑𝐞𝐩𝐥𝐚𝐜𝐞 𝐘𝐨𝐮𝐫 𝐁𝐚𝐜𝐤 𝐎𝐟𝐟𝐢𝐜𝐞 𝐓𝐚𝐬𝐤𝐬?

The 𝐁𝐚𝐜𝐤 𝐎𝐟𝐟𝐢𝐜𝐞 is all about managing the internal operations and administrative tasks of the business. 👉It is quite important for its regular functioning.

The Jobs can include payroll processing, HR handling, and accounting. The digital tool gives adequate chances to automate the actions. However, technology is no doubt essential for the future of the business.

There is pressure to discover the right resources for in-house jobs. Get the full story: Click here👇 to read the full article. http://surl.li/tgrzq

[as8bakwthesage]

updated my alters page

figured mayhaps some people would be interested

[archoneddzs15]

Sega Mega CD - The Ninja Warriors

Title: The Ninja Warriors / ニンジャウォーリアーズ

Developer/Publisher: Aisystem Tokyo / Taito

Release date: 12 March 1993

Catalogue No.: T-11024

Genre: Action

Haven't had much of a chance to play this one but from what I have played so far it seemed pretty good. The animation on the ninja (or cyborg) is nice as are the enemies. The colors are a little drab but overall, they're not bad. Not as good as the Super Famicom game Ninja Warriors Again but still a good game. And let's not forget the great soundtrack by Zuntata, Taito's sound team. This Mega CD version comes complete with a ridiculous Zuntata story prologue movie and a choice of arcade original and re-arranged Zuntata music. Talking about the Zuntata story prologue, the story tries to tell the story behind the Ninja Warriors, and the actors list comprises Zuntata members. The prologue story is badly narrated in English.

[lasseling]

AI Systems Have Learned How To Lie And Deceive

AI systems known as large language models (LLMs) can exhibit “Machiavellianism,” or intentional and amoral manipulativeness, which can then lead to deceptive behavior, according to the findings of a new study.

[waybackwanderer]

undefined

a i systems - Introduction Oct 1997 Archived Web Page 🧩

[govindhtech]

AI Memory Function: The Secret to Smarter Decisions

AI Memory

Contrary to popular belief, humans and artificial intelligence (AI) share many traits. Even while AI is incapable of walking or feeling emotions, it does depend on memory, a critical cognitive ability shared by humans. Learning, reasoning, and adaptation are made possible by AI memory. AI employs memory to store and retrieve data necessary for certain tasks, just as humans do to recall prior experiences and apply knowledge to current circumstances. This article examines memory’s crucial function in artificial intelligence, including everything from its fundamental significance to the ethical issues and upcoming developments influencing its development.

Memory’s two faces

AI is capable of using both long-term memory and short-term working memory. When using the compute processor, short-term memory functions similarly to a cognitive workspace, allowing for instantaneous data manipulation and decision-making. When AI systems have to process and react to spoken or written words, like in real-time language translation, this kind of AI Memory comes in handy. For example, an chatbots rely on short-term memory to keep context intact during a dialogue, guaranteeing well-reasoned and pertinent responses.

Memory AI

AI’s long-term memory serves as a storehouse for previously learned material and life experiences. AI systems with this kind of memory are able to identify trends, gain knowledge from past data, and forecast behavior. Memory AI medical records and creates treatment plans in the healthcare industry using long-term memory, assisting physicians in making wise decisions.

The memory test

When compared to human memory, artificial intelligence memory still faces a number of difficulties, chief among them being latency and speed issues. Even though AI can process data at extremely fast speeds, it is not as efficient as human cognition at quickly integrating and contextualizing knowledge. Due to its slower reaction time, AI is less effective than humans in activities that call for quick, practical thinking and flexibility.

In these situations, human intuition and experience are superior. However, this becomes less of an issue as memory and compute technology develop. System performance functions similarly to the manufacturing industry’s Theory of Constraints management paradigm; when one restriction is lifted, a new one is imposed. Advanced artificial intelligence (AI) systems are increasingly becoming constrained by the quantity of energy they receive.

Memory solutions that reduce energy consumption and maximize computational performance are necessary for AI systems, especially those operating in resource-constrained areas such as data centers, mobile devices, and small drones. Low-power memory technologies like LPDDR5X, high-bandwidth memory (HBM), and DDR5 DRAM need to be innovated in order to address these problems.

AI Memory Future

Technological developments in memory are about to completely transform AI applications in a variety of fields. Data processing bandwidth and speeds are greatly increased by HBM and graphics memory (GDDR). For applications that require real-time analysis of massive datasets, this progress is essential. High-speed memory, for example, makes it possible for sophisticated AI algorithms to quickly assess medical pictures in the healthcare industry, resulting in speedier and more precise diagnosis.

A paradigm change in AI memory design, neuromorphic computing is based on the parallel processing capacities of the human brain. These brain-inspired designs mimic the distributed and interconnected characteristics of neural networks in an effort to improve AI’s adaptability, fault tolerance, and energy efficiency. In order to achieve artificial general intelligence (AGI), where AI systems can execute a wide range of activities with human-like cognition, research in neuromorphic computing appears promising.

Advantages of having a good memory

Strong AI models with high-bandwidth memory support make it possible to create more adaptable and autonomous systems that can learn from big datasets. This could speed up the process of adjusting to new knowledge, resulting in improvements in financial forecasting, predictive maintenance, and personalized care. To anticipate future trends and enhance investment strategies, AI-powered predictive analytics in the banking industry, for instance, use historical market data that has been kept in long-term memory.

Ethics pertaining to long-term memory

The evolution of AI systems to store data for longer periods of time raises ethical questions about data privacy, bias amplification, and decision-making openness. The implementation of frameworks such as explainable AI (XAI) to improve transparency and accountability is necessary to ensure responsible AI development. By using XAI approaches, AI Memory can mitigate any biases resulting from long-term memory and build trust by explaining their conclusions in a way that is understandable to humans.

Leading the way in memory solutions for the AI revolution is Micron

Leading the way in creating memory solutions that are essential to the development of AI is Micron. The advancements in high-bandwidth memory solutions, DRAM, and NAND greatly improve the effectiveness and performance of AI systems, opening up a plethora of applications in many industries.

Because of Micron’s strong supply chain, global R&D footprint, leadership in memory nodes, and industry-leading memory and storage product range spanning the cloud to the edge, they are able to forge the strongest ecosystem alliances possible to hasten the spread of artificial intelligence.

Read more on govindhtech.com

[sifytech]

Hope for Humanity: AI that wins despite not cheating

AI system refusing to cheat in a game where cheating was the rule is a huge deal, says Satyen K. Bordoloi Read More. https://www.sify.com/ai-analytics/hope-for-humanity-ai-that-wins-despite-not-cheating/

[reallytoosublime]

AI technology continues to advance and become more integrated into various aspects of society, and ethical considerations have become increasingly important. but as AI becomes more advanced and ubiquitous, concerns are being raised about its impact on society. In this video, we'll explore the ethics behind AI and discuss how we can ensure fairness and privacy in the age of AI.

#theethicsbehindai#ensuringfairnessprivacyandbias#limitlesstech#ai#artificialintelligence#aiethics#machinelearning#aitechnology#ethicsofaitechnology#ethicalartificialintelligence#aisystem

The Ethics Behind AI: Ensuring Fairness, Privacy, and Bias