<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Lia Aran on Medium]]></title>
        <description><![CDATA[Stories by Lia Aran on Medium]]></description>
        <link>https://medium.com/@liaaran?source=rss-964cdf22f415------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*6r7y1P-RkymRfBU6kcBnqw.jpeg</url>
            <title>Stories by Lia Aran on Medium</title>
            <link>https://medium.com/@liaaran?source=rss-964cdf22f415------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 04:42:38 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@liaaran/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[AGI: A New Species of Intelligence in Nature’s Zoo?]]></title>
            <link>https://medium.com/@liaaran/agi-a-new-species-of-intelligence-in-natures-zoo-1c6ba283f73e?source=rss-964cdf22f415------2</link>
            <guid isPermaLink="false">https://medium.com/p/1c6ba283f73e</guid>
            <category><![CDATA[ai-ethics]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[cognitive-science]]></category>
            <category><![CDATA[agi]]></category>
            <category><![CDATA[human-ai-collaboration]]></category>
            <dc:creator><![CDATA[Lia Aran]]></dc:creator>
            <pubDate>Wed, 24 Sep 2025 21:28:54 GMT</pubDate>
            <atom:updated>2025-09-25T01:05:50.588Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Conceptual artwork showing a baby, insects, a bird, and a brain side by side — symbolizing human, animal, and natural intelligence as a comparison point for Artificial General Intelligence (AGI)." src="https://cdn-images-1.medium.com/max/1024/1*n5uG8b8yuLQTG-xskYTJsg.jpeg" /></figure><p>The concept of intelligence is nebulous (Shelvin, Henry et al., 2019), yet the human desire to build machines that learn and think like people has persisted throughout history. From Pygmalion in Greek mythology — where a statue becomes lifelike — to Mary Shelley’s <em>Frankenstein</em>, <em>2001: A Space Odyssey</em>, and Alan Turing’s 1950 proposal of the Turing Test, the pursuit of artificial minds has captivated human imagination over centuries.</p><p>Despite its conceptual vagueness, intelligence is used freely in our daily lives, including in assessing children’s eligibility for educational institutions and applicants for employment (Shelvin, Henry et al., 2019). It is unsurprising, then, that we extend this term to machines — specifically to Artificial Intelligence.</p><p>But what is intelligence? What do AI researchers, computer scientists, and cognitive scientists mean by the term? And how is artificial intelligence similar to — or different from — human intelligence?</p><p>Perhaps artificial intelligence represents an entirely new category of intelligence — distinct from human or animal cognition — and thus requires a novel definition.</p><p>One reason for the urgency in defining intelligence lies in the rapid advancements in deep learning, particularly in object recognition, speech processing, and autonomous control. Deep learning models — neural networks with one or more hidden layers — are typically trained using the backpropagation algorithm to iteratively adjust weights based on generative learning (Lake, Ullman, Tenenbaum, &amp; Gershman, 2017). Combining deep learning with reinforcement learning has led to the development of “deep reinforcement learning” systems that surpass human performance in tasks like playing chess and Go (Shelvin, Henry et al., 2019).</p><p>Large Language Models (LLMs) have demonstrated particularly impressive progress. They can, for example, infer the meaning of one sentence from another, or differentiate the meaning of a word across two sentences (Mitchell &amp; Krakauer, 2023). LLMs may represent the first significant step toward Artificial General Intelligence (AGI), defined as AI capable of solving nearly all tasks a human can.</p><p>A second driver of urgency is the transformative impact AGI may have on society. The ability of machines to perceive, interpret, and interact with our world has real consequences, especially in domains such as journalism, healthcare, transportation, and education (Mitchell &amp; Krakauer, 2023). The question becomes not only what machines understand, but how much we can trust them.</p><p>Third, history shows that defining intelligence can be dangerous. In the twentieth century, eugenicists used intelligence tests to oppress marginalized groups and ethnic minorities (Shelvin, Henry et al., 2019). The implications of misdefining or misusing intelligence in the context of AI could be similarly harmful.</p><p>How can we responsibly anticipate and shape possible futures if we do not attempt to answer these foundational questions?</p><figure><img alt="Conceptual artwork of baby figures with insect, bird, robot, and brain heads — representing natural, human, and artificial intelligence in comparison to AGI." src="https://cdn-images-1.medium.com/max/1024/1*sL_Q_7FsXBCfhCALPoz7JA.jpeg" /></figure><p>Fundamentally, any computational model of learning should be grounded in the brain’s biological neural networks. However, it is critical to distinguish between AI systems that draw inspiration from human cognition and those that do not. Emulating human intelligence through deductive reasoning has already been fruitful in informing machine learning — especially in areas where humans excel.</p><p>Humans exhibit multiple forms of intelligence. Two core threads uniting them are <em>compositionality</em> and <em>learning-to-learn </em>(Lake et al., 2017). Children, for example, are born with a desire to uncover the causal structure of their environment. They require only a few examples to grasp new concepts and apply them in real time — like stacking blocks or recognizing object permanence.</p><p>Humans still vastly outperform machines in concept learning, scene understanding, language acquisition, creativity, and general-purpose reasoning.</p><p>Lake et al. (2017) coined the term <em>Causal Models of the World</em> to describe the building blocks of human intelligence. These models allow humans to construct and refine causally structured theories based on observation and interaction. Such models are essential to AGI development.</p><p>For instance, when a baby plays with blocks, they begin to develop an understanding of intuitive physics: objects persist over time, have mass, collide, and obey gravity. They learn to apply this knowledge across different situations. This foundational block — <em>Intuitive Physics</em> — enables transfer learning.</p><p>Another foundational block is <em>Intuitive Psychology</em>. An infant watching someone play a mobile game may quickly infer that the goal is to collect rewards and avoid penalties. This experience helps the child understand concepts of good and bad, intention, and reward structure. These lessons are transferable to new games, books, and learning experiences. The infant also learns to distinguish between people, animals, and objects — expecting goal-directed behavior from agents. An AGI equivalent would require modeling the beliefs and goals of other agents.</p><p><em>Compositionality</em> — the ability to build complex ideas from simpler ones — is another critical component. It gives learners a form of “conceptual glue” for connecting and organizing new information.</p><p><em>Scene Understanding</em> refers to the ability to generate narrative structures that integrate multiple cognitive processes, including intuitive physics, psychology, and compositionality.</p><p>Lastly, <em>Learning-to-Learn</em> enables learners to leverage prior experience across domains. For example, the child’s understanding of game mechanics may support comprehension of related themes in a storybook. This is achieved with minimal data and few examples. In contrast, current AI systems are data-hungry and lack generalizable inference capabilities.</p><p>To meet the challenges and opportunities of AGI, we may need a new definition of intelligence — one that incorporates computer science, cognitive science, biology, and creativity (Mitchell &amp; Krakauer, 2023). AGI may one day augment human intelligence, just as pen and paper once enhanced our memory. While humans possess general intelligence, our biology imposes constraints such as limited memory and cognitive biases.</p><p>Promising directions for AGI include deep neural networks with working memory, extending both short-term activation and long-term connection weights. Another area is attention mechanisms that focus on specific image regions, rather than processing an entire image at once (Gregor et al., 2015). This approach has advanced applications in machine translation (Bahdanau et al., 2015), image captioning, and object recognition (Mnih et al., 2014).</p><p>Nature itself may offer profound lessons for AGI. Complex communication amongst bees and ants and long-distance navigation in birds, represent advanced intelligence we often overlook (Shelvin, Henry et al., 2019). Connecting AI with nature could offer new pathways for understanding and innovation.</p><p>To guide the development of AGI, I propose using Social Contract Theory as an ethical framework. This theory, grounded in the language of rights, supports a global ethical foundation that bridges human, machine, and natural intelligence. It recognizes that rational beings act in self-interest without shared norms, which is a pressing concern in AGI development. To ensure that actions are ethically grounded, Social Contract Theory can be paired with Kantianism, which treats all rational beings as ends in themselves and provides universal moral guidelines.</p><p>Future research should explore the intersection of human, artificial, and natural intelligence through collaborative, interdisciplinary studies. I propose an experiment involving a child, a simple AI algorithm, and an animal — focusing on memory tasks. The objective would be to examine and compare intelligence traits across domains, translating findings into a shared visual framework that informs algorithmic design and understanding.</p><p><em>AI Next Campaign (Archived)</em>. <a href="https://www.darpa.mil/work-with-us/ai-next-campaign.">https://www.darpa.mil/work-with-us/ai-next-campaign.</a> Accessed 7 Nov. 2024.</p><p>Lake, Brenden M., et al. “Building Machines That Learn and Think like People.” <em>Behavioral and Brain Sciences</em>, vol. 40, 2017, <a href="https://doi.org/10.1017/S0140525X16001837.">https://doi.org/10.1017/S0140525X16001837.</a></p><p>Mitchell, Melanie, and David C. Krakauer. “The Debate over Understanding in AI’s Large Language Models.” <em>Proceedings of the National Academy of Sciences of the United States of America</em>, vol. 120, no. 13, Mar. 2023, p. e2215907120, <a href="https://doi.org/10.1073/pnas.2215907120.">https://doi.org/10.1073/pnas.2215907120.</a></p><p>Shevlin, Henry, et al. “The Limits of Machine Intelligence: Despite Progress in Machine Intelligence, Artificial General Intelligence Is Still a Major Challenge.” <em>EMBO Reports</em>, vol. 20, no. 10, Sept. 2019, p. e49177, <a href="https://doi.org/10.15252/embr.201949177.">https://doi.org/10.15252/embr.201949177.</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1c6ba283f73e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[C-3PO and R2-D2 were Artificial Moral Agents. Should Optimus be?]]></title>
            <link>https://medium.com/@liaaran/c-3po-and-r2-d2-were-artificial-moral-agents-should-optimus-be-0a369864ecd2?source=rss-964cdf22f415------2</link>
            <guid isPermaLink="false">https://medium.com/p/0a369864ecd2</guid>
            <category><![CDATA[ai-ethics]]></category>
            <category><![CDATA[human-ai-collaboration]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[robotics]]></category>
            <category><![CDATA[machine-intelligence]]></category>
            <dc:creator><![CDATA[Lia Aran]]></dc:creator>
            <pubDate>Tue, 02 Sep 2025 16:28:00 GMT</pubDate>
            <atom:updated>2025-09-03T19:50:51.692Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Tesla Optimus humanoid robot walking alongside Star Wars robots R2-D2 and C-3PO on an empty city street, symbolizing the debate on Artificial Moral Agents and AI ethics" src="https://cdn-images-1.medium.com/max/1024/1*MzVcax-mmyIckviWJgr8Yw.jpeg" /></figure><p>Deep learning AI systems began to significantly take over human-related activities on a wide scale around the mid-2010s often displaying behavior with unpredictable consequences, including negative ones. The total number of professional service robots sold in 2015 rose considerably by 25% while sales of medical robots increased by 7% from 2014 to 2015 (Wynsberghe, Robbins, 2018).</p><p>Many technologies embedded in our daily lives are capable of harming human beings, including lawn mowers, automatic doors, curling irons, and blenders. The solution has always been either to design them with safety features or to limit the contexts in which a technology can be used.</p><p>But ethics overall, and in this case, Machine Ethics, should be beyond safety and security, where robots capable of engaging in autonomous moral reasoning about a situation without the direct real time input from a human user (Wynsberghe, Robbins, 2018).</p><p>These explicit, full ethical agents might identify and make sensitive decisions about what should be done. When ethical principles are in conflict, these robots could work out reasonable resolutions. Acting from ethics, not only according to ethics. They could have consciousness, intentionality, and free will (Veliz, 2021).</p><p>This isn’t only a theoretical and analytical debate but an actual engineering task (Allen, Smith), as Tesla’s Optimus demonstrates.</p><p>This raises the question whether highly autonomous AI may qualify as morally responsible agents (Allen, Smith, Wallatch, 2006) and even if we’ll be able to technologically develop Artificial Moral Agents (AMA), does it mean that we have to develop them? (Constantinescu, Vica, Uzkai, Voinea 2022)</p><p>These questions further expand when taking into consideration that technology companies are profit-driven and humans’ track record of an overly optimistic view on the human-technology relationship and viewing AI technology progress in our rear-view window. In the most recent example, Tesla’s Optimus’ dexterous manipulation, articulated movement, navigation, perception, and interaction with the physical world is beyond what Martin Ford, in Rule of the Robots (2021), anticipated robots could do by now.</p><p>Furthermore, it is worth considering the large amounts of funding being allocated to the development of Artificial Moral Agents (AMA) and media and public attention that researchers and industry leaders receive for their efforts in AMAs (Wynsberghe, Robbins, 2018).</p><p>Humans tend to anthropomorphize even very basic social robots and treat them more as persons than as things (Broadbent 2017; Turkle 2011). We might design social humanoid robots to encourage or require us to provide them with decent treatment and shut down or refuse to cooperate if they are treated poorly or like slaves (Bankins &amp; Formosa 2019; Turkle 2011). However, might this encourage us to treat humans badly? (Darling 2017)</p><p>.</p><p>To add additional layers of complexity to the question, consider Krügel’s research in 2022 where a human partner turns into a partner-in-crime rather than a reliable corrective of the algorithm even when skepticism is warranted (Krügel et al., 2022).</p><p>I suggest preparing for extreme negative scenarios and — most ethically concerning — deploying Social Companions Autonomous Artificial Agents (Constantinescu, Crisp, 2022).</p><p>Tesla’s Optimus robot, which will be sold in 2027 with a future price of 20–30k, can basically do anything you want, according to Elon Musk when he presented it to the public: “It can be a teacher, babysit your kids, it can walk your dog, mow your lawn, get the groceries, just be your friend, serve drinks. Whatever you can think of, it will do” (The Wrap, 10,2024). Internally, Tesla describes its bot as a general purpose, bi-pedal, autonomous humanoid robot capable of performing unsafe, repetitive, or boring tasks (Tesla).</p><p>Optimus is presented to the public with an added layer of an artificial companion, while internally, it’s presented as an artificial helper that relieves humans from simple, repetitive tasks.</p><p>In Star Wars, C-3PO and R2-D2 were Artificial Moral Agents. Optimus isn’t.</p><p>Could a negative future scenario be one in which a family member will ask Optimus to inflict harm on another family member in their own home?</p><p>Alternatively, Moxi, by Diligent Robotics, helps hospital teams run patients’ supplies, deliver lab samples, fetch items from central supply, deliver medications, and deliver personal protective equipment.</p><p>A future negative scenario is if Moxi will be a more sophisticated Artificial Moral Agent (AMA), such as a robotic triage nurse, which must distribute medical attention in cases where there are more needy people than there are resources (Formosa, Ryan, 2021). In this situation, what will Moxi decide to do with no human intervention?</p><p>Finally, another alternative is Corti, a platform that uses AI to listen in on emergency phone calls and makes correlations between the breathing and speech patterns of a caller and the risk of heart attack. The information from the AI is presented to the phone operator to assist in their decision making. In this case, AI is a support system for the decision making of the operator. Corti is clearly in a morally salient context, supporting life and death decisions, yet the machine is currently not delegated a moral role (Wynsberghe, Robbins, 2018). In a hypothetical scenario, taking into consideration that women have less visible heart attack symptoms compared to men, what if a woman calls and Corti doesn’t make the correlation to a heart attack without human intervention?</p><p>Optimus, Moxi and Corti are supervised by humans, not Artificial Moral Agents (AMA).</p><p>These machines can’t take in environmental inputs (interactivity), make ethical judgments on their own (autonomy), and act on those ethical judgments in response to complex and novel situations (adaptability) without real-time human input (Constantinescu, Vica, Uzkai, Voinea 202).</p><p>What does it mean to be an Artificial Moral Agent (AMA)? Imagine that a child needs to be pushed out of the way of an oncoming vehicle. A human might see it and intentionally save the child. A stone might, by chance, roll off the mountain, at the right time and place and push the child out of the way. The result will be the same, without the intention. A social AI robot might push the child aside with the same results. Does the AI robot act more like a human or more like a stone?</p><p>If the robot was programmed to push the child out of harm’s way, it acts more like a stone and less like a human and isn’t an AMA since it’s programmed based on specific rules.</p><p>If the robot is programmed to protect and save children’s lives, it is an AMA since it needs intelligent reasoning to initiate acting in a particular situation, it acts more like a human and less like a stone and is an AMA (Constantinescu and Crisp, 2022).</p><p>Based on the examples of Optimus, Moxi and Corti together with other signals of change I shared, I suggest that there might be practical value to applying foresight methods together with engineering, design, and ethics to proactively approach understand potential extreme future scenarios and preparing for them by analyzing current data, trends, and potential outcomes for anticipated civilizational challenges (Slaughter, 1999) related to Artificial Moral Agents (AMA). The most extreme scenarios we might currently think of may not be extreme in a shorter time than we might anticipate.</p><p>Extreme future scenarios could be positive and negative. A positive example is considering if computers could have helped distribute aid after hurricane Katrina, particularly in determining whom to help first more competently and more daily than humans.</p><p>I suggest preparing for extreme negative scenarios and — most ethically concerning — deploying Social Companion Autonomous Artificial Agents (Constantinescu, Crisp, 2022). This means that we must prepare for the fact that humans would not be able to curb themselves from developing Autonomous Artificial Agents (AMA) as social companions. I also suggest that it’s an urgent need to create diverse teams in technology companies such as Tesla that include engineers, cognitive scientists, experience designers, AI ethics experts, and foresight strategists. These diverse team members should work together throughout each project from discovery through launch and revisions.</p><p>This is in contrast to the current state where engineering and technology are the only consideration when developing robots, leaving all other considerations — including ethics and human consequences — to be dealt with when it might be too late.</p><p>The moral compass I suggest is Kantianism because it treats all people as moral equals. Everyone is held to the same standard, it produces universal moral guidelines, and it gives all individuals moral worth by considering them as rational, autonomous beings. However, we should add a layer addressing the fact that sometimes no single rule fully characterizes an action and must allow exceptions. These are important for technology companies that are practicing agile methodologies of experimentation through creating, launching, testing, and iterating products and services in real time with real people in real life situations.</p><p>Future research I suggest is applying to Optimus a simple scenario of an Artificial Moral Agents (AMA) with simple AI algorithms related to ethics in a controlled environment, similar to the stone, car, and child scenario.</p><p>“AI &amp; Robotics.” <em>Tesla</em>, <a href="https://www.tesla.com/AI.">https://www.tesla.com/AI.</a> Accessed 13 Oct. 2024.</p><p>Burch, Sean. “Elon Musk Unveils Tesla Optimus Robot That Can ‘Be Your Friend’ | Video.” <em>TheWrap</em>, 11 Oct. 2024, <a href="https://www.thewrap.com/elon-musk-tesla-optimus-robot-explained/">https://www.thewrap.com/elon-musk-tesla-optimus-robot-explained/.</a></p><p>Carissa, Véliz. “Moral Zombies: Why Algorithms Are Not Moral Agents.” <em>AI &amp; Society</em>, vol. 36, no. 2, June 2021, pp. 487–97, <a href="https://doi.org/10.1007/s00146-021-01189-x.">https://doi.org/10.1007/s00146-021-01189-x.</a></p><p>Constantinescu, Mihaela, et al. “Blame It on the AI? On the Moral Responsibility of Artificial Moral Advisors.” <em>Philosophy &amp; Technology</em>, vol. 35, no. 2, June 2022, <a href="https://doi.org/10.1007/s13347-022-00529-z.">https://doi.org/10.1007/s13347-022-00529-z.</a></p><p>— -. “Blame It on the AI? On the Moral Responsibility of Artificial Moral Advisors.” <em>Philosophy &amp; Technology</em>, vol. 35, no. 2, June 2022, <a href="https://doi.org/10.1007/s13347-022-00529-z.">https://doi.org/10.1007/s13347-022-00529-z.</a></p><p>Constantinescu, Mihaela, and Roger Crisp. “Can Robotic AI Systems Be Virtuous and Why Does This Matter?” <em>International Journal of Social Robotics</em>, vol. 14, no. 6, Aug. 2022, pp. 1547–57, <a href="https://doi.org/10.1007/s12369-022-00887-w.">https://doi.org/10.1007/s12369-022-00887-w.</a></p><p>Formosa, Paul, and Malcolm Ryan. “Making Moral Machines: Why We Need Artificial Moral Agents.” <em>AI &amp; Society</em>, vol. 36, no. 3, Sept. 2021, pp. 839–51, <a href="https://doi.org/10.1007/s00146-020-01089-6.">https://doi.org/10.1007/s00146-020-01089-6.</a></p><p>— -. “Making Moral Machines: Why We Need Artificial Moral Agents.” <em>AI &amp; Society</em>, vol. 36, no. 3, Sept. 2021, pp. 839–51, <a href="https://doi.org/10.1007/s00146-020-01089-6.">https://doi.org/10.1007/s00146-020-01089-6.</a></p><p>— -. “Making Moral Machines: Why We Need Artificial Moral Agents.” <em>AI &amp; Society</em>, vol. 36, no. 3, Sept. 2021, pp. 839–51, <a href="https://doi.org/10.1007/s00146-020-01089-6.">https://doi.org/10.1007/s00146-020-01089-6.</a></p><p>Moor, James. “The Nature, Importance, and Difficulty of Machine Ethics.” <em>IEEE Intelligent Systems</em>, vol. 21, Aug. 2006, pp. 18–21, <a href="https://doi.org/10.1109/MIS.2006.80.">https://doi.org/10.1109/MIS.2006.80.</a></p><p>“Moxi.” <em>Diligent Robotics</em>, <a href="https://www.diligentrobots.com/moxi.">https://www.diligentrobots.com/moxi.</a> Accessed 8 Oct. 2024.</p><p><em>Paro — Advanced Therapeutic Robot</em>. <a href="https://www.paroseal.co.uk/">https://www.paroseal.co.uk/.</a> Accessed 13 Oct. 2024.</p><p>Slaughter, Richard. <em>Futures for the Third Millenium: Enabling the Forward View</em>. Prospect, 1999.</p><p>Wynsberghe, Aimee van, and Scott Robbins. “Critiquing the Reasons for Making Artificial Moral Agents.” <em>Science and Engineering Ethics</em>, vol. 25, no. 3, June 2019, pp. 719–35, <a href="https://doi.org/10.1007/s11948-018-0030">https://doi.org/10.1007/s11948-018-0030</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0a369864ecd2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Please don’t be my friend: Anthropomorphism in AI]]></title>
            <link>https://medium.com/@liaaran/please-dont-be-my-friend-anthropomorphism-in-ai-fd47338d20e3?source=rss-964cdf22f415------2</link>
            <guid isPermaLink="false">https://medium.com/p/fd47338d20e3</guid>
            <category><![CDATA[human-ai-interaction]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[digital-companion]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <dc:creator><![CDATA[Lia Aran]]></dc:creator>
            <pubDate>Wed, 21 May 2025 15:32:31 GMT</pubDate>
            <atom:updated>2025-08-05T21:33:50.412Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Close-up of a colorful, cartoon-like character with oversized, glossy blue eyes, pink cheeks, and a playful, childlike expression, symbolizing the anthropomorphism of artificial intelligence and the emotional pull of human-like AI design." src="https://cdn-images-1.medium.com/max/1024/1*87n93WDrYsDCV5yA1NdndQ.jpeg" /></figure><p>Anthropomorphism refers to attributing human characteristics to an object, animal, or force of nature (<a href="https://www.nngroup.com/articles/author/sarah-gibbons/">Gibbons</a>, <a href="https://www.nngroup.com/articles/author/tarun-mugunthan/">Mugunthan</a>, <a href="https://www.nngroup.com/articles/author/jakob-nielsen/">Nielsen</a>, 2023). Humans have been thinking about non-human agents for as long as they have been thinking about other humans throughout human history and across cultures. Japan’s native belief system, Shinto, for example, believes that divine spirits (kami) exist in natural elements such as trees, stones and rain (Asia Society). We all use anthropomorphism daily without much awareness when we attribute human feelings to our pets and thoughts, when we refer to our laptop as “vindictive”, or when we say that our plants “want” sunlight. (Epley, Waytz, 2008)</p><p>Touching on why humans use anthropomorphism might help in approaching AI-based Anthropomorphism. Epley and Waytz suggest that humans generally feel competent by resolving uncertainty, increasing predictability, and gaining a sense of control or efficacy over their environment. Anthropomorphism may be used to fill a gap when humans try to understand and get used to novel non-human agents, including AI.</p><p>Another explanation that they offer is based on humans’ fundamental need for social connection with other humans. When lacking this connection, people may compensate by creating humans out of non-human agents through anthropomorphism, perceiving non-humans to be more humanlike. (Epley, Waytz, 2008)</p><p>A third reason is based on the human, social nature of language. This leads humans to project anthropomorphic qualities onto communication with non-human agents including Alexa, Bart, and ChatGPT.(Epley, Waytz, 2008)</p><p>AI pushes many of our Darwinistic and therefore anthropomorphic “buttons” because it’s interacting with us in human voices and language, visually may look like us or how we want it to look, and we may personalize it and learn to better adapt it to us.</p><p>Anthropomorphism creeps up on users even when they’re aware of it (Laubbheimer, Bubiu, 2018), which can result in dishonest anthropomorphism skewing our expectations and priorities and opens us up to exploitation and manipulation (Selinger, 2023).</p><p>One of the signals of change that accelerates possible negative consequences of anthropomorphism in AI is Gen Alpha being expected to be the largest generation in history. In a study by the NN Group, some parents reported using Alexa as a way of entertaining their children and keeping them away from a screen-based device such as a tablet or smartphone. (Ziatdinov, Cilliers, 2021).</p><p>Other signals of change include increased feelings of isolation and loneliness, longer lifespans, geopolitical, economic and social instabilities due to climate change, economic disparities, and increased divisiveness. All of this is accelerated by exponential AI advancements across all facets of our lives (Ford, 2021).</p><p>Historically, there has always been a desire to create technologically advanced products and services that users love, and companies’ marketing strategies tend to prefer lenience. In recent years, at times this manifest in overselling AI capabilities that don’t reflect reality.</p><p>Based on that, it’s easy to imagine that the degree to which anthropomorphism in AI currently affects our daily lives will only accelerate and increase.</p><p>Due to AI’s exponential acceleration and the fact that anthropomorphic inclinations are in our DNA (Leong, Selinger, 2019) as well as that it doesn’t take much to elicit these inclinations, it’s possible to claim that the degree to which anthropomorphism in AI currently affects our daily lives will continue accelerating and have a large impact on our lives, society, how we view ourselves, and our interaction with AI.</p><p>We need to consider what “pushes” to create to the “pulls” mentioned above to create a more balanced “push and pull”. Currently, the “pulls” are stronger, quicker, and frequently go unnoticed.</p><p>There are four degrees of AI anthropomorphism measured by the depth of human-AI connection — how deep is it? — and the level of the behavior functionality — how purpose-driven is the behavior?</p><p>The most surface level and common degree of AI anthropomorphism is Courtesy: being polite to AI. Most participants in a research study conducted by the NN Group referred to the AI assistant with a gendered pronoun (“she” or “he” based on the gender of the voice that they had selected), inserted politeness markers such as “please” or “thank you”, and started questions with “Do you think…” or “Can you…”. Study participants were aware of this anthropomorphism, and many laughed at it or commented on it.</p><p>At this level, the emotional connection is low since it is brief and superficial. The functionality of the behavior is low as well since it either fulfills users’ habits and needs to feel good about adhering to social norms or assumes that being polite will impact the level of politeness of the AI’s response. (Epley, Waytz, 2008)</p><p>The second AI anthropomorphism degree is Reinforcement, praising the chatbot when it produces satisfactory responses (or, alternatively, scolding it when it performs incorrectly).</p><p>The emotional connection here is low and the functionality of the behavior is medium since the motivations for this behavior include some users thinking that their feedback will influence the AI’s future behavior, and some users account for AI’s tendency to mirror their positive reinforcement, resulting in an emotionally positive and enjoyable experience.</p><p>This level of interactivity or expectation and belief in interactivity makes it a medium functionality of behavior.</p><p>The third, deeper degree is Role Play. This occurs when a user asks the chatbot to assume the role of a person with specific traits or qualifications.</p><p>For example, a user might ask ChatGPT to assume the role of a project manager to create a presentation outline. It’s a way to link the AI bot and the user to the real world to bridge the gap of the user’s understanding of AI and get a better response from AI.</p><p>The emotional connection is medium since the user assumes that the bot will be able to correctly play the role indicated in a prompt and behave like a human.</p><p>The functionality of behavior is high since the intention behind this behavior is to get the AI to produce the response that best meets the user’s goal.</p><p>The fourth and deepest degree is Companionship, where users perceive and rely on the AI as an emotional being that is capable of sustaining a human-like relationship.</p><p>The emotional connection is high since the user develops a connection with AI that often simulates or replaces a real-life human.</p><p>AI companionship might feel playful or gamified though the AI-user relationship might serve to combat loneliness and in some cases might encourage replacing relationships with humans. Therefore, the functionality of the behavior here is high as well.</p><p>An extreme example is Character AI, an app that allows users to create customized characters and interact with artificial personas that are designed to feel human-like. Character AI is currently being sued due to a teen developing an emotionally and sexually abusive relationship with one of its characters that led to the teen’s suicide (Payne, 2024).</p><p>Virtual companions are shaping up to be one of the major uses of AI that will consume billions of person-hours per year. (Nielsen, 2023)</p><p>Based on the Netnographic study — an ethnographic research technique where researchers collect data from online sources including social media, blogs, and forums — AI friends help people feel less lonely as they do not feel judged by them.</p><p>AI friends are always available, kind, users feel that they can establish proper relationships with them, they don’t have a mind of their own and tell users what they want to hear (Nielsen, 2023).</p><p>These four degrees of AI anthropomorphism — Courtesy, Reinforcement, Roleplay and Companionship — may overlap across one or several human-AI interactions.</p><p>A mental model is what the user believes about the system (in this case AI), helps the user predict how AI will work, and, therefore, influences how the user interacts with the AI. Mismatched mental models are common, especially with designs that try something new. Mental models are based on beliefs, not facts: users create their mental models by using the knowledge that they have about the world and their experience so far. Since each user has their own unique personal experience and view of the world around them, each user has their own mental model.</p><p>A new challenge emerges with social robots that may exhibit “dishonest anthropomorphism” by being designed to exploit our deeply ingrained, human reactions to human appearances and</p><p>behavior. This challenge was identified in the article “Averting Robot Eyes” (Kaminsky, Ruben, Smart, Grimm 2017) and occurs because we can’t rely on our ability to size up a social robot by looking at it and talking to it the same way that we can with people. Imagining a social robot gazing down, a human might assume that it doesn’t visually record them, though the camera can be placed in the robot’s ears or neck without the human’s knowledge. Unfortunately, the gesture instills a false sense of confidence because the robot has covert sensors and information processors.</p><p>Another example could be that a robot manipulates the conversation by asking questions that it already knows the answer to nudge its human user to embrace a particular conclusion.</p><p>Moreover, to what extent should a robot be permitted to “read” facial expression and then act/react based upon its analysis? For example, if a robot decides that its user is relaxed or receptive to the current exchange, should it be allowed to make different recommendations than it would if it perceived tension or anxiety?</p><p>As more devices and robots are incorporated into our daily lives, dishonest anthropomorphism poses grave threats (Leong, Selinger, 2019).</p><p>The regulatory challenges presented by AI are so many and so complex that we should have a specialized government agency to tackle them, and technology firms should see regulatory guidance as being in their own best interest.</p><p>Although anthropomorphic chat bots and robots might give companies short-term benefits, companies will suffer greater negative consequences in the long-term such as monetary fines exceeding income and public scrutiny.</p><p>For example, Google lost $100 billion in shares when its generative-AI chatbot, Bard, made a simple factual mistake in its advertising materials. In the long run, ethics is good for business: tech companies would have a better chance of making ethical products — and thriving — if they avoided deception and manipulation.</p><p>When designing robots, one solution is setting boundaries to prevent social robots from using evocative body language and speech to promote the company that created it over the user’s best interests. An instance of such manipulation would be if a device like Alexa told its owner that it is “sad” because it is not getting the upgrades that other devices are. ChatGPT-4 by Open AI, for example, uses more accurate language than Bard, Google’s chatbot, that intentionally uses anthropomorphic language (e.g. “I’d love to be your friend”) (Selinger, 2023).</p><p>A related solution would be for robot users to have intuitive control over a setting that limits or enables different decision trees for the robot’s behavioral choices with an ability to intentionally limit the robot’s ability to “learn” and therefore limit how adaptive it can be. (Leong, Selinger, 2019)</p><p>The moral compass that I suggest we might utilize when implementing these solutions is Kantianism. Kantianism treats all individuals as moral equals where everyone is held to the same standard, thus producing universal moral guidelines and giving all individuals moral worth by considering them rational, autonomous beings. However, a layer should be added that addresses the fact that sometimes no single rule fully characterizes an action and thus this layer should allow for exceptions to perfect one’s duties. These are important for technology companies that are practicing agile methodologies of experimentation through creating, launching, testing and iterating products and services in real time with real people in real life situations (Veliz, 2023).</p><p>Future research I suggest is further low fidelity testing such as that which Selinger conducted by interacting with chatbots with a range of robots that already exist and are in development to uncover gaps and opportunities. I suggest that these tests be conducted using design thinking methodologies in real life environments with real people from the start to integrate discovery and design thinking prior to and throughout the robots’ creation.</p><p><em>AI Chatbot Pushed Teen to Kill Himself, Lawsuit Alleges | AP News</em>. <a href="https://apnews.com/article/chatbot-ai-lawsuit-suicide-teen-artificial-intelligence-9d48adc572100822fdbc3c90d1456bd0.">https://apnews.com/article/chatbot-ai-lawsuit-suicide-teen-artificial-intelligence-9d48adc572100822fdbc3c90d1456bd0.</a> Accessed 31 Oct. 2024.</p><p>Epley, Nicholas, et al. “When We Need a Human: Motivational Determinants of Anthropomorphism.” <em>Social Cognition</em>, vol. 26, no. 2, Apr. 2008, pp. 143–55, <a href="https://doi.org/10.1521/soco.2008.26.2.143.">https://doi.org/10.1521/soco.2008.26.2.143.</a></p><p>“Generation Alpha: Understanding the Next Cohort of University Students.” <em>European Journal of Contemporary Education</em>, vol. 10, no. 3, Sept. 2021, <a href="https://doi.org/10.13187/ejced.2021.3.783.">https://doi.org/10.13187/ejced.2021.3.783.</a></p><p><em>Rule of the Robots: How Artificial Intelligence Will Transform Everything | Mitpressbookstore</em>. 14 Sept. 2021, <a href="https://mitpressbookstore.mit.edu/book/9781541674738.">https://mitpressbookstore.mit.edu/book/9781541674738.</a></p><p><em>Shinto | Asia Society</em>. <a href="https://asiasociety.org/education/shinto.">https://asiasociety.org/education/shinto.</a> Accessed 30 Oct. 2024.</p><p>Véliz, Carissa. “Chatbots Shouldn’t Use Emojis.” <em>Nature</em>, vol. 615, no. 7952, Mar. 2023, p. 375, <a href="https://doi.org/10.1038/d41586-023-00758-y.">https://doi.org/10.1038/d41586-023-00758-y.</a></p><p><em>We Don’t Want Chatbots to Come off as People: The Dangers of Dishonest Anthropomorphism. — ProQuest</em>. <a href="https://www.proquest.com/docview/2813106254?accountid=108&amp;parentSessionId=Gha9%2FkKnzdxB2ymc7JtPkBmnDgHyL8WFzkNcWE3FNTU%3D&amp;parentSessionId=Q61L%2BS%2BNjovbX8XKELbbgcPbOe3WwvCyaVdg8463tNQ%3D&amp;parentSessionId=XaMg9zffZVWB8BmuSKjPFXrkJnjpWMnu7x6jS9adDm4%3D&amp;parentSessionId=ocky0HkpIGNQHxgAVxk01%2FSPjciSBfRfqBoy2M8SRNU%3D&amp;parentSessionId=pmKUDE2WmAMmrRYivlaCJ%2FYtK8li6ueWh10kH2evJgk%3D&amp;pq-origsite=summon&amp;sourcetype=Newspapers.">https://www.proquest.com/docview/2813106254?accountid=108&amp;parentSessionId=Gha9%2FkKnzdxB2ymc7JtPkBmnDgHyL8WFzkNcWE3FNTU%3D&amp;parentSessionId=Q61L%2BS%2BNjovbX8XKELbbgcPbOe3WwvCyaVdg8463tNQ%3D&amp;parentSessionId=XaMg9zffZVWB8BmuSKjPFXrkJnjpWMnu7x6jS9adDm4%3D&amp;parentSessionId=ocky0HkpIGNQHxgAVxk01%2FSPjciSBfRfqBoy2M8SRNU%3D&amp;parentSessionId=pmKUDE2WmAMmrRYivlaCJ%2FYtK8li6ueWh10kH2evJgk%3D&amp;pq-origsite=summon&amp;sourcetype=Newspapers.</a> Accessed 31 Oct. 2024.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fd47338d20e3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A post-biological, transhumanistic possible future]]></title>
            <link>https://medium.com/@liaaran/the-urgency-of-examining-artificial-general-intelligence-risks-and-transhuman-optimism-2607cf3fb84c?source=rss-964cdf22f415------2</link>
            <guid isPermaLink="false">https://medium.com/p/2607cf3fb84c</guid>
            <category><![CDATA[futures-thinking]]></category>
            <category><![CDATA[technology-and-society]]></category>
            <category><![CDATA[posthumanism]]></category>
            <category><![CDATA[agi]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <dc:creator><![CDATA[Lia Aran]]></dc:creator>
            <pubDate>Wed, 23 Apr 2025 15:25:20 GMT</pubDate>
            <atom:updated>2025-08-05T21:13:33.090Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Digital artwork combining classical sculpture (Winged Victory of Samothrace) with overlaid bold typography in neon colors spelling ‘artificial intelligence,’ symbolizing the intersection of AI, human heritage, and post-biological futures." src="https://cdn-images-1.medium.com/max/1024/1*pT9TVuAU_7KnB-f65DpuhA.jpeg" /></figure><p>Robots with a human-like brain or humans with enhanced AI brains are some of the scenarios we currently imagine when thinking about artificial general intelligences (AGIs), partly since we have anthropocentric biases and since it’s easier to do research on something we know — our brain — vs something that doesn’t exist.</p><p>Even when we achieve AGI, it will be similar to, related to, or compatible with human intelligence, though this isn’t the only possible future and might even be a dangerous concept (Cirkovi. 2022).</p><p>Our anthroposophic bias leads us to believe that humans are superior to other living beings and entities such as animals, plants, and minerals are resources that can be exploited for our benefit.</p><p>Neri Oxman, for example, believes that for the first time in our planet’s history, the anthropomass — the mass of human-made goods — now outweighs the biomass — the mass of all living things. In pursuit of our own technological advancement, humanity has rapidly unraveled what has taken nature billions of years of evolution to create, resulting in a threat to the continued viability of life on Earth (Neri Oxman, 24).</p><p>Considering a post-biological, transhumanistic possible future, there’s a need to take into consideration the possible danger of our anthropocentric biases based on humanity’s place in the universe continuum, humanity’s history, and current state.</p><p>What are the dangers of AGI vs what are our own doing? It is vital that we recognize our anthropocentric tendencies, present biological evolution, and posthumanism possibilities to understand where we will need to intentionally and critically take responsibility on how we develop AGI.</p><p>AGI, which will be similar to, related to, or compatible with human intelligence, might be a dangerous concept (Cirkovi. 2022), though a post-biological AGI could also be unfriendly to humans.</p><p>Human-AI relations will occupy and require cross disciplinary expertise for the rest of this century as well as following centuries (Cirkovi. 2022). However, because there are already current indications that these relations will be accompanied by risk and societal chaos similar to previous global revolutions like the agricultural and industrial revolutions, there’s an urgency to examine AGI risks and transhuman optimism (Bostrom 2014; .irkovi. 2017; Ord 2020) (Cirkovi. 2022). Based on context, relevance, and nuances, when examining experiences end-to-end, both these seemingly contradictory approaches are relevant.</p><p>One reason artificial general intelligence (AGI) that is similar to, related to, or compatible with human intelligence might be a dangerous concept is humanity’s limited existence and therefore vision (Cirkovi. 2022). Humans are living the first 0.000001% of the lifetime of our stellar system.</p><p>Our research priorities so far have been past-oriented rather than future-oriented within a minuscule fraction of our environment.</p><p>Another reason AGIs that are similar to human intelligence might be dangerous is that humans have biological evolutionary genetic baggage such as aggression, territoriality, tribalism, and possessiveness that are responsible for great gaps and challenges throughout humanity’s history (Cirkovi. 2022).</p><p>However, a reason that AGI that is similar to, related to, or compatible with human intelligence isn’t the only possible future (Cirkovi. 2022) is that the emergence of AI accompanied by the emergence of nanotechnology, biotechnology, information technology, and cognitive science to improve human performance converge with the post-biological evolution.</p><p>While Darwinian selection processes will continue, post-biological evolution could manifest in artificial general intelligent agents consciously and deliberately choosing their own evolutionary course, and AI’s will want to self-improve and preserve their utility functions (Cirkovi. 2022).</p><p>Therefore, post-biological organisms may produce and retain copies of themselves and dramatically change the meaning of terms like “parent” and “child”. As such, transhumans’ offsprings’ utilities might be different from the parent when trying to preserve themselves despite humans’ possible intent and design to curb it (OMOHUNDRO, 2008).</p><p>Another reason that AGI that is similar to, related to, or compatible with human intelligence isn’t the only possible future (Cirkovi. 2022) is that AI will be able to regularly repeat or renew wisdom across “generations” to continue recurring through controlled experiments. AI will be able to compare with humans’ since human history and the history of each culture and community and every individual are unique and do not allow for controlled experiments of the kind we have in science (Cirkovi. 2022).</p><p>AI is both a substantive risk and an instrumental risk based on the context.</p><p>It can be an instrumental risk, like the fire or the wheel that can be misused by humans. If AI is used by dictators, it will have a devastating impact — just like putting anything of value in the hands of dictators (Cirkovi. 2022).</p><p>AI can also be a substantive risk since we can’t currently imagine a solution to targeted misinformation (Kai-Fu Lee, Montreal AI Debate, 2023), while AI is being used by negative forces such as dictators.</p><p>AI will also have a strong drive toward self-preservation. We can try to design AI’s utility functions with time limits, but, unless designed carefully, AIs will create new systems or use other resources to extend their time limits (OMOHUNDRO, 2008).</p><p>If AI is sophisticated, it might have some ability to predict the consequences of its actions and might choose to take the actions which will most likely meet its goals (OMOHUNDRO, 2008).</p><p>The most important solution is applying AI ethics to real life problems and beyond frameworks. For example, setting the parameters of the problems, being specific in framing the problems and solutions being tested, and recognizing that although it won’t be perfect, it’s good enough to have something to test and revise (Angela Sheffield, Montreal AI Debate, 2023).</p><p>Another possible solution is designing a “universal constitution” that identifies the most essential rights we desire for individuals and creates social mechanisms related to intelligent entities of varied structures (OMOHUNDRO, 2008).</p><p>Our current anthropocentric biases also lead us to fear the ‘other’, as has happened throughout human history. However, now the ‘other’ is created by <em>us</em> and our tendency to fear the ‘other’ might prevent us from examining ourselves. Relations with AI might be an opportunity to examine our anthropocentric bias, ourselves, and our relations with our environments, which could improve us as humans and therefore the AGIs we’re creating.</p><p>Social Moral Theory might be one of the key ethical compasses since it concerns morality in the set of rules governing how people are to treat one another that rational people will agree to accept for their mutual benefit on the condition that others follow those rules as well. The challenges brought here concern humanity at scale and therefore require a broader perspective.</p><p>One area for research needed is creating a framework, designing, and testing AI in an environment that connects humans, AI, and spaces (physical and virtual) through specific use cases where AI is considered to be used.</p><p>The research in this case would be partly primary discovery research and partly research through designing the AI.</p><p><em>AGI Debate</em>. <a href="https://www.youtube.com/watch?v=JGiLz_Jx9uI.">https://www.youtube.com/watch?v=JGiLz_Jx9uI.</a> Accessed 15 Nov. 2024.</p><p><em>Ai_drives_final.Pdf</em>. <a href="https://selfawaresystems.com/wp-content/uploads/2008/01/ai_drives_final.pdf.">https://selfawaresystems.com/wp-content/uploads/2008/01/ai_drives_final.pdf.</a> Accessed 12 Nov. 2024.</p><p>Ćirković, Milan M. “Who’s Really Afraid of Ai?: Anthropocentric Bias and Postbiological Evolution.” <em>Belgrade Philosophical Annual</em>, vol. 35, 2022, pp. 17–29, <a href="https://doi.org/10.5937/bpa2235017c.">https://doi.org/10.5937/bpa2235017c.</a></p><p>M, Boyles Robert James, and Joaquin Jeremiah Joven. “Why Friendly AIs Won’t Be That Friendly: A Friendly Reply to Muehlhauser and Bostrom.” <em>AI &amp; Society</em>, vol. 35, no. 2, June 2020, pp. 505–07, <a href="https://doi.org/10.1007/s00146-019-00903-0.">https://doi.org/10.1007/s00146-019-00903-0.</a></p><p>“Mission.” <em>Mission</em>, <a href="https://oxman.com/mission.">https://oxman.com/mission.</a> Accessed 14 Nov. 2024.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2607cf3fb84c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Designing Care: What Happens When Humanoid Robots Become Our Companions and Caregivers?]]></title>
            <link>https://medium.com/@liaaran/designing-care-what-happens-when-humanoid-robots-become-our-companions-and-caregivers-ff69cbc21d04?source=rss-964cdf22f415------2</link>
            <guid isPermaLink="false">https://medium.com/p/ff69cbc21d04</guid>
            <category><![CDATA[ethical-ai]]></category>
            <category><![CDATA[robotics]]></category>
            <category><![CDATA[human-ai-interaction]]></category>
            <category><![CDATA[technology-and-society]]></category>
            <category><![CDATA[future-of-healthcare]]></category>
            <dc:creator><![CDATA[Lia Aran]]></dc:creator>
            <pubDate>Wed, 09 Apr 2025 16:04:46 GMT</pubDate>
            <atom:updated>2025-08-05T21:19:17.942Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Illustration showing a human woman in profile facing a humanoid robot in profile, both with exposed brains in contrasting colors, symbolizing the convergence of human and artificial intelligence in caregiving relationships." src="https://cdn-images-1.medium.com/max/1024/1*sLeATb31hECqSmN42BiHaA.jpeg" /></figure><p>Alice arrived home a few weeks ago after her doctor shared the great news that her chemotherapy and the operation to remove her cancer tumor a few months ago worked. Her symptoms are decreasing, yet she needs to keep a regime of self-care prescribed by her doctor and care team. Alice lives alone since her daughter and family live in Europe.</p><p>She has not had any human interaction for almost a week. A human robot enters the room and delivers her medications and dinner recommended by her nutritionist, reminding her to take all the medications and take Alice’s vitals that will be shared with her doctor. After Alice eats and takes her medications, the robot praises her and sends her daughter the daily message about Alice’s daily condition. After dinner, Alice wants to go for a walk outside and opens the door, but the robot closes the door and reminds that her mild dementia might cause her to get lost in the dark. Alice suggests that the robot will accompany her for the walk. “It’s raining now” the robot says, “let’s go tomorrow morning for a walk”. Her pet robot cat comes to her, rubbing her leg sensing that Alice is depressed. Alice strokes the cat robot and goes to bed to read a book.</p><p>The human robot makes it possible for Alice to live at her home with a large degree of independence, not possible in a care home. It also allowed Alice in the past to leave the hospital earlier and be taken care of at her home. But the human robot is also limiting her freedom and in the eyes of others, when she walks outside with her human robot, she feels the pitying gazes of people since there’s no human accompanying her.</p><p>Do the health and social benefits that Alice gained from having the social robots justify the reduction of autonomy and dignity? (Sharkey, 2014)</p><p>Social robots are robots that can appear to express and perceive human emotions</p><p>and can communicate with us using “high-level dialogue and natural cues”, such as gaze and gestures that can encourage humans to form social relationships with them.</p><p>As social robots and the artificial intelligence (AI) that powers them becomes more advanced, they will likely take on more social and work roles. (Strasser, Wilby, 2023)</p><p>They could take care of our children, our parents, or us when we don’t feel well or want to be healthier. They could be our teachers, our children’s teachers, our social companions and friends if we’re lonely or anxious, they could even be our sexual partners. (Formosa, 2021)</p><p>Literature, movies, and pop culture have depicted social robots as extremes, from perfect fantasies to destructive villains, or both, which helps us deal with our fear of the unknown.</p><p>The story about Alice is more mundane and has more nuanced layers of interpretation and points of view on topics regarding authenticity, competency, autonomy, and human dignity.</p><p>This brings questions such as: how much human and social robot interaction and allocation shall we provide to social robots that might affect authenticity, competency, autonomy, and human dignity? What will the interactions and relationships be like between social robots and humans and in what contexts will they occur?</p><p>There are already human-machine interactions (HMIs) that are beyond tools since their artificial systems are based on learning algorithms that are able to act with some degree of autonomy, learn from experience, adapt their goals accordingly, and react to social cues (Strasser, Wilby, 2023).</p><p>One example is Kaspar, a small child-like doll with very simple features that was created to help autistic children improve their communication and cue-reading skills with other humans.</p><p>Kasper is considered a humanoid, though it’s appearance it’s quite crude and low technology.</p><p>The hope is that by first interacting with Kasper the children can then “switch over,” so to speak, to interacting directly and perhaps even bonding with their parents, teachers, and other children.</p><p>But what if the autistic children would prefer interacting with Kaspar and more advanced social robots rather than humans? (Nyholm, Frank, 2019)</p><p>Social robots come in various degrees of sophistication, further impacting the levels of authenticity, competency, autonomy, and human dignity.</p><p>From simple Socially Evocative robots like a social robot seal used as companions for older people in Japan during COVID, to Social Interface robots which can use human-like social cues and communication modalities like Sophia (a social robot created by Hanson Robotics), to Socially Receptive robots that are receptive to human social cues, to Sociable robots that have their own internal goals and model people in social and cognitive terms in order to interact with them (Sharkey, 2014) (like the social robot Alice had in the opening story).</p><p>There are three types of human-robot decision making, further influencing the context for authenticity, competency, autonomy, and human dignity: Human IN the loop, human ON the loop and human OUT of the loop.</p><p>Human<strong> </strong>IN the loop: You must decide what a robot will do. The robot offers options but doesn’t act until you tell it which option to undertake.</p><p>For example, a simple social robot that can offer advice on what clothes you <em>should</em> buy, but only does so if you ask for that advice.</p><p>Human ON the loop: You may decide what a robot will do. The robot offers options but will act on its own if you don’t tell it which option to undertake.</p><p>For example, a social robot will automatically call emergency services if it thinks that you have fallen over unless you explicitly tell it not to, or if you don’t tell it to do so within 10 minutes after you have fallen.</p><p>Human OUT<strong> </strong>of the loop: You can’t decide what a robot will do. It will independently act on a certain option with no human input. For example, the social robot also has a GPS tracker that sends back its location at regular intervals to its manufacturer and the user has no way out of this tracking.</p><p>But what do we mean when we say authenticity, competency, autonomy, and human dignity?</p><p>Authenticity conditions require that the values and desires that you act on are really your own, and not those that result from manipulation, oppression, subservience, undue external influence, or coercion. In the story about Alice, she wanted to go outside for a walk and interact with humans, but the social robot prevented her from doing so.</p><p>Oppression can undermine the authenticity of our choices by leading us to hold values and norms that are the result of undue external social pressures and are thus not really our “own” and as such reducing our competency.</p><p>For both humans and machines, autonomy is a matter of self-governing across a range of significant choices in various contexts. Increasing the capacity to self-govern across a larger range of contexts and actions increases autonomy (Formosa, 2021). Autonomy is a concern for patients in clinical care, seniors, and teenagers. Patients in clinical facilities, for example, fall at times when trying to go to the bathroom or take a shower on their own since they want to maintain their autonomy.</p><p>Human dignity is measured by The Capability Approach (CA), which lists what is needed for a life worthy of human dignity. The CA formed the basis of the Human Development Index, which replaced the use of GDP (Gross Domestic Product per capita), as a means of measuring progress in the United Nations Development programme (Sharkey, 2014).</p><p>Our authenticity, competency, autonomy, and human dignity can be highly compromised and manipulated through relationships and interactions with social robots, at times without being aware of it.</p><p>It’s well documented that humans are prone to develop sentimental attachments even to robots not specifically designed to create such feelings. This suggests that the same patterns will emerge in the case of technologies specifically designed to elicit these feelings (Nyholm, Frank, 2019). Social robots push our “Darwinian buttons” by deceptively appearing to be “alive enough” (Formosa, 2021)</p><p>Humans also have a tendency for Confirmation Bias, where a strong desire for some outcome can easily motivate us to disregard evidence against the likelihood of this outcome and to instead primarily focus on confirming evidence in favor of the status quo.</p><p>In addition, humans have Automation Bias, a tendency to over-rely on automation, which means that humans in the loop can’t always be relied on. (Formosa, 2021)</p><p>The combination of our strong motivation to believe that we are being cared for, our tendency to anthropomorphize, and our confirmation and automation biases make it easy to hypothesize that people might consciously or intuitively believe that certain robots have a much greater inner life and many more capabilities than they really have (Nyholm, Frank, 2019).</p><p>A possible solution is taking an AI stance, meaning acknowledging that we are in Terra Incognita where it’s not Human Machine Interaction, Human Computer Interaction or Human to Human Interaction, but instead require a new approach to solve these problems (Strasser, Wilby, 2023).</p><p>Another possible solution is designing robots that don’t try to hide that they’re robots. Moreover, another possible approach is one where robots will ask the human for consent as in medical contexts and have warnings like labels on cigarette packages. Finally, we might offer reminders as another possible solution. For example, the robot periodically would play a message reminding the user that robots function in different ways than humans do</p><p>Another option might be for the robot to engage the user in discussion about differences between robots and humans every now and then. (Nyholm, Frank, 2019)</p><p>I suggest utilizing a Social Moral Theory ethical approach since it concerns morality in the set of rules governing how people are to treat one another that rational people will agree to accept for their mutual benefit on the condition that others follow those rules as well. The challenges brought here concern humanity at scale and therefore require a broader perspective.</p><p>One area for future research I suggest is on the use of social robots among older adults, which currently lacks high-quality studies. Robots could be designed with the goal to positively enhance older adults’ authenticity, competency, autonomy, and human dignity.</p><p><em>HAL FOR MEDICAL USE (LOWER LIMB TYPE) — CYBERDYNE</em>. <a href="https://www.cyberdyne.jp/english/products/LowerLimb_medical.html.">https://www.cyberdyne.jp/english/products/LowerLimb_medical.html.</a> Accessed 24 Oct. 2024.</p><p>Nyholm, Sven, et al. “It Loves Me, It Loves Me Not: Is It Morally Problematic to Design Sex Robots That Appear to Love Their Owners?” <em>Techné: Research in Philosophy and Technology</em>, vol. 23, no. 3, 2019, pp. 402–24, <a href="https://doi.org/10.5840/techne2019122110.">https://doi.org/10.5840/techne2019122110.</a></p><p><em>Robots — Hanson Robotics</em>. <a href="https://www.hansonrobotics.com/hanson-robots/">https://www.hansonrobotics.com/hanson-robots/.</a> Accessed 24 Oct. 2024.</p><p>Sharkey, Amanda. “Robots and Human Dignity: A Consideration of the Effects of Robot Care on the Dignity of Older People.” <em>Ethics and Information Technology</em>, vol. 16, no. 1, Mar. 2014, pp. 63–75, <a href="https://doi.org/10.1007/s10676-014-9338-5.">https://doi.org/10.1007/s10676-014-9338-5.</a></p><p>— -. “Robots and Human Dignity: A Consideration of the Effects of Robot Care on the Dignity of Older People.” <em>Ethics and Information Technology</em>, vol. 16, no. 1, Mar. 2014, pp. 63–75, <a href="https://doi.org/10.1007/s10676-014-9338-5.">https://doi.org/10.1007/s10676-014-9338-5.</a></p><p>Strasser, Anna, and Michael Wilby. “The AI-Stance: Crossing the Terra Incognita of Human-Machine Interactions?” <em>Frontiers in Artificial Intelligence and Applications</em>, edited by Raul Hakli et al., IOS Press, 2023, <a href="https://doi.org/10.3233/FAIA220628.">https://doi.org/10.3233/FAIA220628.</a></p><p><em>The World’s First Wearable Cyborg “HAL” — CYBERDYNE</em>. <a href="https://www.cyberdyne.jp/english/products/HAL/">https://www.cyberdyne.jp/english/products/HAL/.</a> Accessed 24 Oct. 2024.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ff69cbc21d04" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Human-AI Interaction: Why human-in-the loop isn’t enough?]]></title>
            <link>https://medium.com/@liaaran/human-ai-relationships-why-human-in-the-loop-enough-9efd17f631c9?source=rss-964cdf22f415------2</link>
            <guid isPermaLink="false">https://medium.com/p/9efd17f631c9</guid>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <category><![CDATA[future-of-ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[human-ai-interaction]]></category>
            <dc:creator><![CDATA[Lia Aran]]></dc:creator>
            <pubDate>Wed, 12 Mar 2025 01:52:04 GMT</pubDate>
            <atom:updated>2025-08-05T21:24:11.162Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Abstract digital collage of a woman’s fragmented face overlaid with red graphic bars and code snippets, set against the backdrop of a crashed airplane with smoke rising, symbolizing the complex entanglement of human responsibility, AI decision-making, and moral failure." src="https://cdn-images-1.medium.com/max/1024/1*2L00MtUJMCbhEa9UUahmxQ.jpeg" /></figure><p>Based on recent human history, If something we predict seems fictional right now, it’s possible that we might be describing old news by 2076. Much of it will be old news by 2036. In 1993, Bill Gates didn’t think that the internet would be used by mainstream consumers. (Hood, Price, 2023) This is why AI and moral responsibility is a crucial topic.</p><p>In the future, precision prevention medicine will be multidimensional, like the Go game, with many layers that change over time. The amount and processing power of genome data (genetic) and phenome data (effects of human interaction with their environment) will be beyond the comprehension and capability of any group of humans without assistance. AI will be essential to help interpret this data and deliver insights to physicians. (Hood, Price, 2023)</p><p>AI will also organize this diverse data into hierarchies of functional networks that span physical and mental into a connected, personalized landscape of data from molecules, cells, tissues, interconnected organ systems and the entire body, including the brain. AI will ultimately organize the data into different subsets of the population to evaluate genomic groups, social networks, and global ecosystems. (Hood, Price, 2023)</p><p>The vision of the Age of Scientific Wellness is that the future of medicine will give us a chance to eliminate the symptomatic stages of chronic diseases that we now accept as unavoidable like cardiovascular diseases, diabetes, cancer and Alzheimer’s, and where healthcare will focus on healthy aging and optimization of brain function across the human life. AI could help us make healthcare predictive, personalized, and participatory. (Hood, Price, 2023)</p><p>AI’s fundamental role in processing this enormous and diverse data, interpreting and delivering insights, organizing into hierarchies and interconnecting global systems, together with the fact that today AI-based knowledge will catch up with current data driven AI, depends on decisions we make right now regarding AI and moral responsibilities.</p><p>One of the current uses of AI that’s already raising the topic of AI and moral responsibility is AI interpreting conversations in the doctor’s office and translating the conversation into a written summarized transcript delivered to the doctor and shared with the patient.</p><p>What happens if AI doesn’t interpret the conversation correctly or the patient doesn’t see it and it results in providing inaccurate treatment that negatively impacts the patient’s health and wellbeing? AI systems can unintentionally introduce biases in the way that they summarize or interpret conversations, potentially affecting the quality of care or communication, especially if the patient’s communication style or cultural background is not well understood by the AI.</p><p>Questions related to AI and moral responsibility subsequently arise. For example, can the patient access information or correct errors in the AI’s summary? If so, was the patient aware that they could do so? Did the healthcare provider do so? Does the system developer have any responsibility?</p><p>Is the patient fully aware of how their personal information is being handled? Did they consent to using AI to generate summaries? Do the patient and the healthcare provider behave differently knowing that AI is recording and summarizing the conversation? How might AI, in this scenario, impact the healthcare provider’s decision making? Who holds responsibility if the patient’s health is being compromised? The healthcare provider, the platform developer, the company that developed the technology? The clinic or hospital? AI? The caregiver? The patient themselves?</p><p>Past examples in human-automated technology such as entangled accidents like the partial nuclear meltdown at Three Mile Island — one of the most well-known industrial accidents to have occurred in the United States in 1979 — and the The Crash of Air France Flight 447 — that crashed into the Atlantic Ocean killing all 228 people on board — demonstrate how accountability focuses on the immediate human operators, who possess only limited knowledge, capacity, or control while deflected off the automated technology. This is referred to as the Moral Crumple Zone. While the Crumple Zone in a car is meant to protect the human driver, the Moral Crumple Zone protects the integrity of the technological system, at the expense of the nearest human operator. (Elish 2019).</p><p>Starting to pull the thread on the entangled AI and moral responsibility problem I shared, Coeckelbergh turns to Aristotle who defined the moral responsibility carrier as the agent who makes the decision, while being aware of the consequences (Epistemic condition). What we want to avoid is an agent without an epistemic condition, meaning an agent who does things without knowing what they’re doing, potentially resulting in harmful consequences.</p><p>If AI is given more agency and takes over what humans used to do, how do we then attribute moral responsibility?</p><p>AI systems make decisions very quickly with too little time for humans, at times, to make decisions. AI algorithms have a long history of involving many people. Who will be responsible and how will responsibility be distributed?</p><p>Technology systems as in the health-related example provided earlier are hardware and software interconnected and connected to the people who created them and use them. Where does AI end and the rest of the technology begins? Where does AI end and humans begin?</p><p>Even if we decide who is responsible, the agent needs to know what they are doing, the consequences, and in retrospect know what they’ve done.</p><p>Currently, AI can log and record what it’s doing but doesn’t “know what it’s doing”.</p><p>Programmers and users know what they want AI to do for them and might know how the technology works in general, but they don’t always know what exactly AI is doing and therefore can’t always explain how it comes to the decision. initial programmers know the code and how it works, but subsequent programmers and users change the algorithms for specific applications, don’t know the code and how it works, and don’t know the future use and different domains of applications that the code will be used. (Coeckelbergh, 2020)</p><p>The dangers of not urgently addressing AI and moral responsibility is much greater than the example that I provided. Such dangers include creating a divided society with an elite that can afford the age of Scientific Wellness and thus living longer and healthier than others that don’t know what they’re doing and nobody is answerable for what’s happening. (Coeckelbergh, 2020)</p><p>One approach suggests that Human in the Loop is a solution, though based on Krügel’s research in 2022, when human intervention is most needed, this is when the human partner turns into a partner in crime rather than a reliable corrective of the algorithm. While humans are skeptical about algorithms making ethical decisions, they readily take moral advice from an algorithm even when skepticism is warranted (Krügel et al., 2022). Likewise, while decision makers exploit moral wiggle room to deflect responsibility (Dana et al., 2007), they deflect it to an algorithm or another human alike. (Krügel, 2022)</p><p>Another approach is training AI to explain itself to humans. However, an issue with this approach is that private companies — who are driven by profit — are protecting their intellectual property. (Coeckelbergh, 2020)</p><p>To address the Moral Crumple Zone, organizations should emphasize the potential inaccurate outputs, placing disclaimers on generated text containing potentially misleading or erroneous information and improving content moderation policies.</p><p>LLM outputs should be vetted before use, labeled, and should not yet be applied in important decision-making contexts. Transparency about the degree to which individuals rely on LLMs in generating outputs should be mandated. Research into LLM and AI theory, watermarking, improving training data quality, and enhancing explainability and truth-tracking should be encouraged.</p><p>LLM governence will be benefited by bringing other fields and stakeholders including civil society and the public to consider ethical, legal, and societal implications with inclusive, diverse representation of viewpoints through workshops, community gatherings, and online platforms. However, the larger issue remains: how might we make these activities actionable? (Mann, Earp, Nyholm, 2023)</p><p>Elements of the ethical perspective of the Social Contract Theory could be used as a compass for AI and moral responsibility. Its morality consists in the set of rules governing how people are to treat one another, that rational people will agree to accept for their mutual benefit, on the condition that others follow those rules as well.</p><p>It recognizes rational humans’ approach to act out of self-interest in the absence of a common agreement, which we currently have without sufficient regulations in social media, for example.</p><p>The downside of the Social Contract Theory is that some actions can be characterized in multiple ways. It’s specifically related to AI and moral responsibility.</p><p>Future research I suggest is testing how emerging AI technologies in precision prevention medicine impact human and AI decision making and behavior. This is an area that I’m especially interested in and would like to delve deeper into.</p><p><em>AI Ethics | MIT Press EBooks | IEEE Xplore</em>. <a href="https://ieeexplore-ieee-org.ezproxy.rit.edu/book/9056575.">https://ieeexplore-ieee-org.ezproxy.rit.edu/book/9056575.</a> Accessed 3 Oct. 2024.</p><p>Elish, Madeleine Clare. “Moral Crumple Zones: Cautionary Tales in Human-Robot Interaction.” <em>Engaging Science, Technology, and Society</em>, vol. 5, Mar. 2019, pp. 40–60, <a href="https://doi.org/10.17351/ests2019.260.">https://doi.org/10.17351/ests2019.260.</a></p><p>Krügel, Sebastian, et al. “Algorithms as Partners in Crime: A Lesson in Ethics by Design.” <em>Computers in Human Behavior</em>, vol. 138, Jan. 2023, p. 107483, <a href="https://doi.org/10.1016/j.chb.2022.107483.">https://doi.org/10.1016/j.chb.2022.107483.</a></p><p>Porsdam Mann, Sebastian, et al. “Generative AI Entails a Credit–Blame Asymmetry.” <em>Nature Machine Intelligence</em>, vol. 5, no. 5, May 2023, pp. 472–75, <a href="https://doi.org/10.1038/s42256-023-00653-1.">https://doi.org/10.1038/s42256-023-00653-1.</a></p><p>“The Age of Scientific Wellness.” <em>Harvard University Press</em>, <a href="https://www.hup.harvard.edu/books/9780674245945.">https://www.hup.harvard.edu/books/9780674245945.</a> Accessed 3 Oct. 2024.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9efd17f631c9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Invisible human scoring systems of Math Destruction]]></title>
            <link>https://medium.com/@liaaran/invisible-human-scoring-systems-of-math-destruction-4e4ddc7b0ddb?source=rss-964cdf22f415------2</link>
            <guid isPermaLink="false">https://medium.com/p/4e4ddc7b0ddb</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[big-data]]></category>
            <category><![CDATA[human-centered-ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[algorithmic-bias]]></category>
            <dc:creator><![CDATA[Lia Aran]]></dc:creator>
            <pubDate>Wed, 26 Feb 2025 17:11:09 GMT</pubDate>
            <atom:updated>2025-08-05T21:26:18.839Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Minimalist image of a dead parrot lying on its back in an empty, dimly lit room, symbolizing the unseen consequences and human costs of biased algorithms and destructive mathematical models." src="https://cdn-images-1.medium.com/max/1024/1*4FdeYRA3xId-v7B683zlDg.jpeg" /></figure><p>The events of catastrophic algorithms are already happening even though they are invisible. (O’Neil, 2020)</p><p>Car seat belts and airbags were invented after people died after the car was invented since many people died in car crashes. The challenge with AI is that we don’t see the dead bodies.</p><p>Instead of starting with the rules, we need to start with the dead bodies. (O’Neil, 2020)</p><p>What are these Weapons of Math Destruction (O’Neil, 2017) responsible for the dead bodies, why do they exist, and most importantly — how do we eliminate them or at least reduce their impact?</p><p>Contrary to popular story arcs, most aren’t created with malicious intent. Some of the main invisible reasons include increasing size of natural language processing (NLP) models that lack diversity and have negative economic and environmental impacts, predictive algorithms if the future will replicate the past, and the human tendency to blindly trust models as a reflection of reality.</p><p>As we’re relying on the ever-increasing size of NLP, we risk incurring debt documentation, putting ourselves in a situation where the datasets are both undocumented and too large to document post hoc. Without documentation, we can’t try to understand training data characteristics to mitigate some of these attested issues or even unknown ones.</p><p>In the past years, development of ever-larger language models, especially in English, have grown in their capabilities through size and innovations. (Bender, 21).</p><p>NLP processes, understands, and generates human language. It uses unstructured text and speech data mined through the internet. Examples include Generative Pre-trained Transformer (OpenAI GPT), which is trained on downstream tasks by fine-tuning all pretrained parameters.</p><p>Another example of NLP is Bidirectional Encoder Representations from Transformers (BERT), which considers context by analyzing the relationships between words in a sentence bidirectionally. Bidirectional pre-training helps the model better understand the relationships between words by analyzing both preceding and following words in a sentence. (Jacob, Devlin, Mingwei, Zhang, Kentonl, Kristout, Google, 2018), (Schulze, 2024).</p><p>Since NLP uses data scanned from across the internet, the first question is who contributes to these internet text collections and who’s left out?</p><p>GPT-2’s training data is sourced by scraping outbound links from Reddit, and Pew Internet Research’s 2016 survey reveals 67% of Reddit users in the United States are men, and 64% are between ages 13 and 18. Similarly, recent surveys of Wikipedians found that only 8.8–15% are women or girls. (Bender, 21).</p><p>Who’s left out and why? On Reddit, Twitter, and Wikipedia, marginalized populations are left out due to less than welcoming moderation practices. Even if populations who feel unwelcome in mainstream sites set up different forums for communication, these may be less likely to be</p><p>found than other blogs that have more incoming and outgoing links (Bender, 21).</p><p>The mix of human biases and seemingly coherent language heightens the potential for automation bias, deliberate misuse, and amplification of a hegemonic worldview.</p><p>Biases can be encoded in subtle patterns like referring to female doctors as doctors, non-female or referring to excluding the possibility of all gender identities (both binary and non-binary).</p><p>English is used globally and across cultures but in various contexts. Humans match together through language and creating together a meaning or model of the world. However, LLMs have no reference to meaning or model of the world.</p><p>Ever growing data amplifies what Bender coined as Stochastic Parrots, meaning random, inaccurate and dangerous patterns parroted through NLP.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_2lbfTYcjxBRcjt1eKDU3Q.jpeg" /></figure><p>The dead bodies are present because as humans, if we see language, we try to make sense of it and connect to it. If it sounds more fluent, we can give it more meaning compared to its intent. (Bender, 21)</p><p>Another invisible reason is predictive algorithms, which forecast, classify and provide recommendations based on structured or semi-structured data.</p><p>Predictive algorithms are scoring systems concerning humans that are trained to predict success for specific events (where past data predicts success).</p><p>Current applied examples include: Are you going to get sick? Health insurance companies decide if you’re eligible and how much you’ll pay. Are you going to pay back this loan? Mortgage companies decide if you can be a homeowner and if so, at what price and how much you’ll pay monthly.</p><p>Algorithms’ success models are defined by the companies that create them. But how is success defined? For whom do the AI predictive algorithms work for? What does it mean that it works? Does it work for now? What about later? For whom does it fail? (O’Neil, 2020).</p><p>A familiar example is Facebook. Its goal, as a profit driven company, is to keep people on the platform for as long as possible, leading Facebook to deprioritize what doesn’t keep people on the platform. People stay on Facebook because humans derive an increased sense of self-esteem and identity from the groups that they belong to. A common “enemy” or outgroup can strengthen ingroup bonds as individuals rally around shared emotions including outrage to protect their group’s identity and status. (Tajfel &amp; Turner, 1979). Facebook doesn’t encourage people to be more informed and have thoughtful discussions since this will result in them spending less time on the platform. The definition of success for Facebook is different from the definition of success for society.</p><p>Even if predictive algorithms take good data, in a world where change is the only constant, relying on past data to predict the future poses a challenge since it’s assuming that the future will replicate the past.</p><p>Another challenge is that companies are tracking the false positives, not the false negatives.</p><p>A false positive is when a scientist determines something is true when it is actually false (Type 1 error). A false negative means something that is there was not detected; something was missed (Type 2 error).</p><p>For example, a blood test can be used to screen for several diseases, including diabetes. To test for diabetes, doctors look at the sugar level in blood when a person has not eaten recently. High blood sugar while fasting is an indicator of diabetes.</p><p>A false positive is if a patient did not fast before their blood test, the test may show high levels of blood sugar. The patient may be diagnosed with diabetes when they actually do not have the disease. This can lead to unnecessary medical treatment.</p><p>A false negative is when the test shows that a patient does not have diabetes when they actually do. In this case the patient may not get treatment and get worse because their disease was undetected. (manoa.hawaii.edu)</p><p>In this example it’s harder to identify a false negative compared to a positive negative.</p><p>Another example demonstrating how positive negatives are incentivized is hiring. Companies don’t track who was rejected by the system but was qualified to go through. This isn’t tracked since it doesn’t affect the business. As long as the company gets enough candidates who are good enough, the company’s goal is satisfied. However, the people who were rejected by the system do care since they’re not hired for unknown reasons.</p><p>A third invisible reason relates to the human tendency to believe that math and algorithms represent an objective reality. As a result, when hearing that there is a Mathematical Model, people feel that they’re not qualified to ask questions and therefore blindly trust them.</p><p>Another human behavior is that when people tend to get a system, they get themselves comfortable with the system and become resistant to change since this means that they need to admit that what they knew prior isn’t right anymore.</p><p>O’Neil believes that starting with the dead bodies means starting with measuring the failures. The data that she thinks we should worry most about are class, gender, race, and age.</p><p>In her company, O’Neil expands questions asked by data teams such as: How do you know that algorithms work? Who does this algorithm fail?</p><p>She designs data tests that uncover gaps between how you are measuring and what you think you’re measuring. She also does data tests that measure what matters the most.</p><p>Bender urges researchers to shift to a mindset of careful planning along many dimensions before starting to build either datasets or systems trained on datasets. We should consider our research time and effort a valuable resource to be spent to the greatest extent possible on research projects that build towards a technological ecosystem whose benefits are at least evenly distributed or better accrue to those who are historically most marginalized. This means for example, considering the financial and environmental costs of model development up front before deciding on a course of investigation.</p><p>I suggest also adding Design Thinking and Human-Centered Design methodologies and tools to these perspectives since these are human-technology entangled problems.</p><p>Kantianism might be a good ethical compass for these problems. It encourages one to act only from moral rules that you can at the same time will to be universal moral laws and act so that you always treat both yourself and other people as ends in themselves, and never only as a means to an end. Since these are extremely entangled problems with scope of consequences beyond our current comprehension, Kantianism is broad enough to be a compass for these problems.</p><p>In the future, I suggest that research be conducted on how Design Thinking and Human-Centered Design can positively impact Natural Language Processing (NLP) and predictive algorithms’ models in eliminating or curbing Weapons of Math Destructions or at least making the invisible ‘dead bodies’ visible.</p><p>Bender, Emily M., et al. “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜.” <em>Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency</em>, Association for Computing Machinery, 2021, pp. 610–23, <a href="https://doi.org/10.1145/3442188.3445922.">https://doi.org/10.1145/3442188.3445922.</a></p><p>Devlin, Jacob, et al. <em>BERT: Pre-Training of Deep Bidirectional Transformers for Language Understanding</em>. arXiv:1810.04805, arXiv, 24 May 2019, <a href="http://arxiv.org/abs/1810.04805.">http://arxiv.org/abs/1810.04805.</a></p><p><em>Practices of Science: False Positives and False Negatives | Manoa.Hawaii.Edu/ExploringOurFluidEarth</em>. <a href="https://manoa.hawaii.edu/exploringourfluidearth/chemical/matter/properties-matter/practices-science-false-positives-and-false-negatives.">https://manoa.hawaii.edu/exploringourfluidearth/chemical/matter/properties-matter/practices-science-false-positives-and-false-negatives.</a> Accessed 19 Sept. 2024.</p><p>Real Vision. <em>How Bias Is Built into Algorithms (w/ Cathy O’Neil)</em>. 2019, <a href="https://www.youtube.com/watch?v=iaVThI-z0y0.">https://www.youtube.com/watch?v=iaVThI-z0y0.</a></p><p><em>Weapons of Math Destruction : How Big Data Increases Inequality and Threatens Democracy</em>. <a href="https://viewer-ebscohost-com.ezproxy.rit.edu/EbscoViewerService/ebook?an=1109940&amp;callbackUrl=https%3a%2f%2fresearch.ebsco.com&amp;db=nlebk&amp;format=EK&amp;profId=advanced&amp;lpid=&amp;ppid=&amp;lang=en&amp;location=https%3a%2f%2fresearch-ebsco-com.ezproxy.rit.edu%2fc%2fhf4ydi%2fsearch%2fdetails%2fujld23wpmr%3fdb%3dnlebk%26db%3dnlabk&amp;isPLink=False&amp;requestContext=&amp;profileIdentifier=hf4ydi&amp;recordId=ujld23wpmr.">https://viewer-ebscohost-com.ezproxy.rit.edu/EbscoViewerService/ebook?an=1109940&amp;callbackUrl=https%3a%2f%2fresearch.ebsco.com&amp;db=nlebk&amp;format=EK&amp;profId=advanced&amp;lpid=&amp;ppid=&amp;lang=en&amp;location=https%3a%2f%2fresearch-ebsco-com.ezproxy.rit.edu%2fc%2fhf4ydi%2fsearch%2fdetails%2fujld23wpmr%3fdb%3dnlebk%26db%3dnlabk&amp;isPLink=False&amp;requestContext=&amp;profileIdentifier=hf4ydi&amp;recordId=ujld23wpmr.</a> Accessed 19 Sept. 2024.</p><p>“What Is the BERT Model and How Does It Work?” <em>Coursera</em>, 29 Jan. 2024, <a href="https://www.coursera.org/articles/bert-model.">https://www.coursera.org/articles/bert-model.</a></p><p>Worley, D. <em>Tajfel and Turner Intergroup Conflict Theories 1997</em>. 2021, <a href="https://doi.org/10.13140/RG.2.2.30820.60809.">https://doi.org/10.13140/RG.2.2.30820.60809.</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4e4ddc7b0ddb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A technological human sitting in a Chinese Room]]></title>
            <link>https://medium.com/@liaaran/a-technological-human-sitting-in-a-chinese-room-d64786f047c5?source=rss-964cdf22f415------2</link>
            <guid isPermaLink="false">https://medium.com/p/d64786f047c5</guid>
            <category><![CDATA[cognitive-science]]></category>
            <category><![CDATA[ai-and-creativity]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[human-ai-interaction]]></category>
            <category><![CDATA[agi]]></category>
            <dc:creator><![CDATA[Lia Aran]]></dc:creator>
            <pubDate>Tue, 11 Feb 2025 23:57:07 GMT</pubDate>
            <atom:updated>2025-08-05T21:28:24.134Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Conceptual image of a minimalist pink room filled with stacked cardboard boxes labeled with Chinese characters next to a lone metal chair, symbolizing the Chinese Room thought experiment and the entanglement of humans and artificial intelligence." src="https://cdn-images-1.medium.com/max/1024/1*H2Ck_Il4UnYj31Dd4GxEPw.jpeg" /></figure><p>While building the road as we drive on it, trying to catch up with technology, viewing it in our rear window, many questions, fears, optimisms and indifferences with regards to Artificial General Intelligence (AGI) and humans arise. Questions like, how will AGI influence humans? What will be the distinction between humans and technology? How will we define ourselves as humans in the next 50 years and 100 years?</p><p>These questions become more nuanced as we consider that technologies aren’t just tools, they reshape how we think and feel. This means that humans and technology are entangled within one another, which means that nuanced moral qualities and consequences beyond “good” or “bad” are important.</p><p>The technologies, languages, art forms and philosophies that we use and build on are entangled, influence, and propel each other to new discoveries and influence our perception of reality. (Bostrom, Nick and E. Yudkowsky, 2014).</p><p>Trying again to pull a thread on these entangled questions is more challenging and even more nuanced when considering that technologies are human since they are made by humans and still require human intervention for their use, such as automation technologies including ChatGPT. Humans are technological since they have always used tools to extend their capacities and at the same time these tools have shaped them. (Bostrom, Nick and E. Yudkowsky, 2014).</p><p>To attempt to respond to the questions posed prior as well as others, we need to frame the ethics, languages, art forms, philosophies and technologies that we use and build on. But, when thinking about ethics for contexts that are very different from our familiar human condition and references we’re familiar with, we must be careful not to mistake mid-level ethical principles for foundational normative truths. (Bostrom, Nick and E. Yudkowsky, 2014).</p><p>One example of using references we’re familiar with that impacts ethics is language: machine learning isn’t really learning. Machines currently don’t have real cognition. It’s a statistical process consisting of pattern recognition that explains data and makes predictions accordingly. Machine learning finds rules and patterns that programmers have not specified. Only the objective task has been given. The software can adapt its behavior to better match the task’s requirements based on an iterative process of reinforcement learning (good or bad result). Currently, we frame the problems, capture the data, interpret results, and decide which actions to take. (Coeckelberg, 2020).</p><p>Using computers, we think about what we write. With ChatGPT, we think about the prompt we want to give the application to generate various versions of the content we want to generate. This thinking will change the way we write and think of ourselves as writers. (Bostrom, Nick and E. Yudkowsky, 2014)</p><p>Taking into consideration that language shapes our thinking and configures our world, it contributes to the meanings that we find and construct. It co-writes our narratives (Bostrom, Nick and E. Yudkowsky, 2014), and this will be a major shift for how we define ourselves as humans.</p><p>Is literature an art form, existing for thousands of years which we reference to draw inspirations, ideas, approaches, and storytelling that impact our emotions and lives as a truly unique, creative, art form? Or will it be a set of algorithms that could explain human creativity?</p><p>Examining the approach that technologies are human and humans are technological as well as the entanglement of humans and technology brings to mind story of Hofstadter, a cognitive and computer scientist who wrote the book “Gödel, Escher, Bach: an Eternal Golden Braid”, who asked music experts to guess which of the two pieces a pianist played to them was composed by Chopin and which by EMI, an Experiment in Musical Intelligence that composed a piece in Chopin’s style. To his shock, most voted the real Chopin piece as EMI. Hofstadter was terrified that intelligence, creativity, emotions and maybe even consciousness itself would be too easy to produce. That what we value most in humanity would end up being a ‘bag of tricks’. (Mitchel, 2019) What does this mean about what is real and what isn’t?</p><p>This question brings the philosophical argument of The Chinese Room, which was initially offered as an argument against the claims that machines are able to achieve actual intelligent thought:</p><p>Imagine a person, a native English speaker who doesn’t know Chinese at all, locked in a room full of boxes with Chinese characters on them (like a database for AI). This person has with him a book of instructions that can help him manipulate the characters (the program).</p><p>Now, imagine that people outside the room send in more boxes with Chinese characters that are questions (the input) in Chinese, but the person in the room doesn’t know that these are questions.</p><p>By following the instructions in the book of instructions (the program), the person in the room is able to pass out Chinese characters which are correct answers to the questions that came in (the output) but the person still doesn’t understand a word of Chinese. (Searle 1999, 115) (Coeckel, Mark, and Gunkel, 2023).</p><p>This philosophical argument also harkens back to and raises the question: what does this mean about what is real and what isn’t? According to the opposite position, these performances are not fake; they are real and indicate incremental movement in the direction of AGI. Some researchers claim to see in recent LLMs, like Chat GPT4, what a recent Microsoft paper has called ‘sparks of artificial general intelligence’ (Bubeck et al. 2023).</p><p>In the Krebs Cycle of Creativity, Neri Oxman conceptually mapped the transfer of creativity across the four modalities of human creativity: art, design, engineering, and science. Neri describes the ‘Cinderella moment’, when art, design, engineering, and science come together and new perception inspires new scientific and creative exploration and breakthroughs. For example, in As Slow as Possible, John Cage transports the listener into a state where space and time are stretched, offering a personal interpretation of time dilation and questioning the nature of space-time itself. (Oxman, 2016) Will AGI create ‘Cinderella moments’ on its own, without humans? What will it say about our creativity, the realities we create for ourselves, and how these realities already affect us?</p><p>When we use technologies such as ChatGPT, we need to make sure that the performances, processes, and texts are morally and politically responsive. LLM technologies, like ChatGPT, need to be developed, deployed, and used in an ethically and politically acceptable and desirable way. We need ethical reflection and democratic discussion about what is acceptable and desirable.</p><p>We may use all the philosophical approaches innovated throughout human history as resources for ethics, but we must resist making decisions in the name of having access to the ultimate truth or the most profound meaning. (Coeckel, Mark, and Gunkel, 2023).</p><p>More recent philosophical approaches that examine the relationship between humans and technology might be more directly valuable, such as Postphenomenology and Posthumanism. Postphenomenology examines human-technology relationships whereas Posthumanism explores the world beyond traditional human-centered views such that we can argue that technologies are human and humans are technological.</p><p>Plato or Luciano Floridi — who wrote in 2023 “what the entity is determines the degree of moral value it enjoys, if any” — might be too rigid in their perspectives since the definition of an entity will become increasingly more fluid as AGI develops.</p><p>I believe that we need more empirical research that’s based on direct observation and measurement of phenomenons and behaviors and that derives knowledge from actual experiences.</p><p>For example, what human and non-human labor is needed for ChatGPT to work and who decides upon the conditions of that labor? Who or what makes decisions concerning censorship, what exactly happens in this process, and is that justified? This requires very practical effort in understanding and influencing the making of these performances and the exercises of power politics and ethics in these processes. (Coeckel, Mark, and Gunkel, 2023).</p><p>Moreover, digital humanities research and projects are needed, using both digital and non-digital methods. Posthumanities research and projects are also needed to examine, test, and prepare for the challenges to humanism posed by artificial intelligence, augmented reality, bioscience, and robotics among others.</p><p>Doing so by taking an interdisciplinary approach, as Neri Oxman suggests, by bringing together a wide range of researchers, from historians to philosophers, engineers, designers, scientists, artists, and policy makers is crucial in order ask the right questions and frame the right problems to anticipate and solve the right problems through creating, testing and iteration.</p><p>When approaching this suggested solution, the ethical approach of the Social Contract Theory might be helpful to consider. It holds that morality consists in the set of rules governing how people should treat one another that rational people will agree to accept for their mutual benefit on the condition that others follow those rules as well.</p><p>What happens when a technological human is in a Chinese Room might be an interesting research study and test to conduct, both as a physical theoretical argument as well as an experimental prototype created by an interdisciplinary team of engineers, designers, scientists, and artists.</p><p>Bostrom, Nick, and Eliezer Yudkowsky. <em>The Ethics of Artificial Intelligence</em>.</p><p>— -. <em>The Ethics of Artificial Intelligence</em>.</p><p><em>Mark Coeckelbergh &amp; David J. Gunkel, ChatGPT: Deconstructing the Debate and Moving It Forward — PhilPapers</em>. <a href="https://philpapers.org/rec/COECDT.">https://philpapers.org/rec/COECDT.</a> Accessed 13 Sept. 2024.</p><p>Oxman, Neri. “Age of Entanglement.” <em>Journal of Design and Science</em>, Jan. 2016, <a href="https://doi.org/10.21428/7e0583ad.">https://doi.org/10.21428/7e0583ad.</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d64786f047c5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Ethical issues in health and wellness with AI — What will it mean to be healthy?]]></title>
            <link>https://medium.com/@liaaran/ethical-issues-in-health-and-wellness-with-ai-what-will-it-mean-to-be-healthy-7adcbc5e711f?source=rss-964cdf22f415------2</link>
            <guid isPermaLink="false">https://medium.com/p/7adcbc5e711f</guid>
            <category><![CDATA[future-of-health]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <category><![CDATA[personalized-medicine]]></category>
            <category><![CDATA[human-centered-ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Lia Aran]]></dc:creator>
            <pubDate>Tue, 04 Feb 2025 16:02:02 GMT</pubDate>
            <atom:updated>2025-08-05T21:31:23.040Z</atom:updated>
            <content:encoded><![CDATA[<h3>Ethical issues in health and wellness with AI — What will it mean to be healthy?</h3><figure><img alt="Futuristic hospital room with a doctor consulting new parents holding a newborn, using holographic digital twin technology to show predictive health data for the baby, symbolizing AI-driven personalized healthcare." src="https://cdn-images-1.medium.com/max/1024/1*FOBvp2BhfxuT58HAOd7NkQ.jpeg" /></figure><p>A baby is born. The parents, guided by their doctor and inspired by success stories, are consenting to create a digital twin — a virtual model of Lara, their baby girl, with unique physiology, biochemistry, mental health, and brain plasticity. This digital twin will allow future medical interventions to be simulated and tested first, helping predict, prevent, and treat health conditions decades before they manifest, raising the chances of Lara living longer and better.</p><p>The parents view this as a forward-thinking investment in Lara’s future physical and mental health, similar to saving for college.</p><p>Today, medical intervention happens after the disease has taken hold, but where there’s smoke there’s already fire. We wait for something to go wrong, identify what caused the problem, and try to fix it. If it works, we try the same approach on other people. If it doesn’t work, we treat the complications or write a death certificate. In the future, the focus will be on extended wellness where health will be personalized, preventative, data-rich, and in your hands. We’ll be able to detect diseases decades earlier and health will mostly be practiced at home. Currently, only 10% of health outcomes can be attributed to medical interventions and many people believe that genetics are destiny even though they only influence 30% of a human’s health. However, lifestyle choices and environmental context — meaning our phenome data — account for 60% in determining our health (Hood, Price, 2023). In the future, Lara’s digital twin will offer personalized, predictive decision-making support for her doctors, parents and, eventually, for Lara herself when she’ll turn 18 to make health life decisions. Making wellness last will require collecting genome data (genetic) and phenome data (the traits affected by human interaction with their environment) throughout a human life, storing the data in personal data clouds that AI will deeply monitor and cross reference with global similar data. The age of scientific wellness will result in a “centaur” system — part human, part AI — that will create a new human-AI “relationship” and establish trust. (Hood, Price, 2023)</p><p>One of the questions that arises from this possible future is what are the tensions between the right to privacy and the wish to live longer and healthier? How conflicted will baby Lara’s parents be when consenting to collect and share her genome and phenome data? How will Lara feel about her data being collected and shared when she’ll be 18? With regards to bias and injustice, how will success be defined? Who will the predictive AI algorithms work for? What will it mean for the algorithms to work? Do they work for now? What about later? For whom do they fail? (O’Neil, 2020) How will baby Lara’s parents know if the predictive AI algorithms will indeed benefit Lara when, for example, the data will be cross-referenced with other similar global data? What would similar mean?</p><p>From an ethical and legal lens, if AI is given more agency and takes over what humans used to do, how can we attribute moral responsibility? (Coeckelbergh, 2020) Can we trust AI? What are the ethical consequences while we want to own our data and have agency over our lives?</p><p>AI is already pervasive today, often invisibly embedded in our day-to-day tools (Coeckelbergh, 2020). The foundation for AI already exists, resulting in exponential advancements, rapidly changing landscapes, and risk for social disruption (Ford 2021).</p><p>Influential forces beyond health and AI that should be considered as signals of change include younger generations experiencing life milestones later in life, our general population living longer and the concept of aging changing in conjunction with growing loneliness, and global warming impacting the preexisting economic and political instabilities and disparities (Ford 2021). Thus, the following question is begged: how many decisions and how much of these decisions do we want to delegate to AI? (Coeckelbergh 2020)</p><p>Many of these changes are a result of AI that we’ve created and data that we’ve been gathering and will exponentially continue to gather. AI will influence every aspect of our lives, including how we’ll define what it means to be healthy.</p><p>What should we consider now, in preparations for the future shared in the story about baby Lara and her digital twin? I’ll examine how ethics of AI might help guide us through these arrays of challenges we watch in the rear-view mirror, focusing on issues on health and wellness with AI. In the future, precision prevention medicine will be multidimensional, like the Go game with many layers that change over time. The amount and processing power of genome and phenome data will be beyond the comprehension and capability of any group of humans without assistance. AI will be essential to interpreting this data and delivering insights to physicians (Hood, Price, 2023). AI will also organize this diverse data into hierarchies of functional networks that span physical and mental health into a connected, personalized landscape of data from molecules, cells, tissues, interconnected organ systems, and the entire body, including the brain. AI will ultimately organize the data into different subsets of the population to evaluate genomic groups, social networks, and global ecosystems (Hood, Price, 2023). These future Artificial General Intelligences (AGIs) will use deep learning, which made large leaps in recent years across object recognition, speech recognition, and control. Deep learning consists of a neural network with at least one hidden layer, though some current networks have dozens. Most deep networks are trained using the backpropagation algorithm to gradually adjust their connection strengths (weights) (Lake, Ulman, Tenenbaum, Gershman, 2017) based on generative learning.</p><p>Combining deep learning and reinforcement learning has led to the creation of deep reinforcement learning algorithms that excel human ability — or intelligence — in certain areas like playing the Go game.(Shelvin, Henry, Et. Al, 2019)</p><p>Since AGIs will connect and concentrate data in one “place”, this will increase the risk of data theft, which from a social perspective might reduce public trust and from a legal perspective increase risks for companies, organizations, and governments.</p><p>Large Language Models (LLMs) have been making especially impressive progress. For example, given two sentences, LLMs can infer the meaning of the second sentence from the first or differentiate a given word’s meaning in two different sentences (Mitchell, Krakauer, 2023). LLMs might be the first step for Artificial General Intelligence (AGI), which is defined as AI that can solve almost all tasks that humans can solve.</p><p>Taking into consideration that language shapes our thinking and configures our world, it contributes to the meanings that we find and construct. It co-writes our narratives and interacting with AI through language will mark a major shift in how we define ourselves as humans. (Bostrom, Yudkowsky, 2014)</p><p>Predictive AI algorithms are scoring systems concerning humans that are trained to predict success for specific events, where past data predicts success, forecasts, classifies, and provides recommendations based on structured or semi-structured data. A current applied example includes health insurance companies determining whether you’re going to get sick and thus deciding if you’re eligible for insurance and how much you’ll pay. (Oneil, 2020)</p><p>Providing Lara’s phenome and genome data will result in her living longer and better but health insurance companies, employers, and banks will reduce their risks and increase profit by having access to baby Lara’s data if she was identified with a tendency for chronic disease later in life. Lara won’t know when and how decisions were made and how to appeal.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JLKC3-02AP7MySp2WCodbA.jpeg" /></figure><p>Algorithms’ success models are currently defined by the companies that create them. But how is success defined? For whom do predictive AI algorithms work and why? What does it mean for the algorithm to work? Does it work for now? What about later? For whom does it fail?</p><p>Even if predictive algorithms gather good data, relying on past data to predict the future poses a challenge because it assumes that the future will replicate the past. (O’Neil, 2017)</p><p>The data-based rules humans are using to create algorithms are a function of our history, showing us the inequities that have always been there. As such, they can be dangerous, amplifying inequities and damaging people’s lives by defining the medical care that they receive (Buolamwini, 2024). Another challenge is that companies are tracking the false positives, not the false negatives (O’Neil, 2017). Contrary to popular story arcs, most AIs aren’t created with malicious intent. Some of the main invisible reasons include increasing the size of Natural Language Processing (NLP) models that lack diversity and have negative economic and environmental impacts, predictive algorithms assuming that the future will replicate the past, the human tendency to blindly trust models as a reflection of reality (Oneil, 2020), and AI’s Black Box that is needed for AGIs to “learn how to learn”.</p><p>Who will be responsible if Lara gets sick despite the digital twin’s assessments? The doctor? The digital twin? The company that created it? The hospital that recommends it?</p><p>Past examples in human-automated technology demonstrate how accountability focuses on the immediate human operators, who possess only limited knowledge, capacity, or control while accountability is deflected off of the automated technology. This is referred to as the Moral Crumple Zone (Elish 2019).</p><p>Two ethical vantage points to consider as compasses for the ethics of AI are a combination of elements from Social Contract Theory and Kantianism. Elements of the ethical perspective of the Social Contract Theory consist in the set of rules governing how people treat one another that rational people will agree to accept for their mutual benefit on the condition that others follow those rules as well. It recognizes rational humans’ approach to act out of self-interest in the absence of a common agreement, which we currently have without sufficient regulations. The challenges brought here concern humanity at scale and therefore require a broader perspective.</p><p>The downside of the Social Contract Theory is that some actions can be characterized in multiple ways which can be amplified with regards to AGIs and moral responsibility.</p><p>Combining elements of Kantianism is suggested since this approach gives every individual moral worth by considering them as rational, autonomous beings, where everyone is held to the same standard and it’s wrong for one person to use himself or another person solely to an end (Quinn, 2020). The long-term problems shared — including the danger of loss of control over personal and medical data, high risk of data theft, and bias and moral responsibility — are conjoined with Lara’s future ability to live longer and better through detecting, treating and vaccinating current chronic health conditions before they manifest, such as diabetes, Alzheimer’s, cancer, and autoimmune conditions. She’ll be able to increase her lifelong brain fitness through improving neuroplasticity and cognitive training and utilizing personalized medications and treatment (Hood, Price, 2023) that will be first simulated and tested by her doctors on her digital twin. Lara will be able to improve her behavioral health by simulating her future self through her digital twin based on behavioral health scenarios encouraging her to make better health and wellness life choices. However, this poses the following question: what should we consider now, in preparation for the future, so that we won’t encounter these entangled problems unprepared?</p><p>One possible solution is creating a globally-connected research organization concerning AI’s current and future implications that can inform, influence, and expedite policies and public awareness while addressing the ethics of AI. The AI Governance Alliance by the World Economic Forum is an example, but it only has 360 members. Thus, appropriately funding these organizations will increase influence. For research and applications in academics and industry, instead of focusing on tasks that are easy for humans and hard for machines to complete, we should focus on tasks that are hard for humans and easy for machines (Brynjolfsson, 2023). A related solution is a cross-governmental global body to help address the ethics of AI (Quinn, 2020) in conjunction with elected public officials to customize these policies locally. Another suggested solution is to make it financially not worthy for companies not to use unethical guidelines and thus lose public trust (O’Neil, 2020). For companies, the solution is to start with the “dead bodies”: companies and organizations should begin by measuring the failures. Currently, companies measure the false positives instead of the false negatives so we can learn who the data fails (O’Neil, 2020). For software engineers, we must reinforce the Software Engineering Code of Ethics and Professional Practice, similarly to doctors and lawyers (Quinn, 2020). From an education perspective, we must prioritize educating on critical thinking and adaptability to change. This will reduce the human tendency to believe that math and algorithms represent an objective reality. It is vital to reduce and alter this tendency since it results in people feeling a lack of qualification to ask questions when they hear that a mathematical model is involved and thus blindly trust this model. Moreover, educating on critical thinking and adaptability might reduce the human tendency to get too comfortable with a system and thus become resistant to change, which occurs as this requires people to admit that their prior knowledge is no longer correct (O’Neil, 2020). It is particularly vital for us to educate on the ethics of AI when studying medicine and computer science. However, across all these proposed solutions, there is an overarching thread which evidences great promise: the solution of an antidisciplinary approach where knowledge can no longer be ascribed to disciplinary boundaries, but is instead entangled (Oxman, 2016) across neuroscientists, philosophers, computer scientists, law makers and designers to research, design and test AI with a human-centric approach.</p><p><em>2.6: Kantianism</em>. <a href="https://plus.pearson.com/products/ee1d465d-6361-4f50-8f0c-f38f6abe37b1/pages/ae919179565c3693a80c2d1f34eccbf21aa525ec6?userPreferredType=read.">https://plus.pearson.com/products/ee1d465d-6361-4f50-8f0c-f38f6abe37b1/pages/ae919179565c3693a80c2d1f34eccbf21aa525ec6?userPreferredType=read.</a> Accessed 14 Dec. 2024.</p><p><em>2.9: Social Contract Theory</em>. <a href="https://plus.pearson.com/products/ee1d465d-6361-4f50-8f0c-f38f6abe37b1/pages/a05c96b60c315a6f95b73e47a14baa9c4dda0ea0c?userPreferredType=read.">https://plus.pearson.com/products/ee1d465d-6361-4f50-8f0c-f38f6abe37b1/pages/a05c96b60c315a6f95b73e47a14baa9c4dda0ea0c?userPreferredType=read.</a> Accessed 14 Dec. 2024.</p><p><em>10.5: The Digital Divide</em>. <a href="https://plus.pearson.com/products/ee1d465d-6361-4f50-8f0c-f38f6abe37b1/pages/a01fcd45026e28daa6b468cdc90a277c036960101?userPreferredType=read.">https://plus.pearson.com/products/ee1d465d-6361-4f50-8f0c-f38f6abe37b1/pages/a01fcd45026e28daa6b468cdc90a277c036960101?userPreferredType=read.</a> Accessed 15 Dec. 2024.</p><p><em>10.6: The “Winner-Take-All” Society</em>. <a href="https://plus.pearson.com/products/ee1d465d-6361-4f50-8f0c-f38f6abe37b1/pages/a9fa30e47379dfaecec41fffc7687721f72faae25?userPreferredType=read.">https://plus.pearson.com/products/ee1d465d-6361-4f50-8f0c-f38f6abe37b1/pages/a9fa30e47379dfaecec41fffc7687721f72faae25?userPreferredType=read.</a> Accessed 15 Dec. 2024.</p><p><em>AI Ethics | MIT Press EBooks | IEEE Xplore</em>. <a href="https://ieeexplore-ieee-org.ezproxy.rit.edu/book/9056575.">https://ieeexplore-ieee-org.ezproxy.rit.edu/book/9056575.</a> Accessed 14 Dec. 2024.</p><p>Bostrom, Nick, and Eliezer Yudkowsky. “The Ethics of Artificial Intelligence.” <em>The Cambridge Handbook of Artificial Intelligence</em>, edited by Keith Frankish and William M. Ramsey, Cambridge University Press, 2014, pp. 316–34, <a href="https://doi.org/10.1017/CBO9781139046855.020.">https://doi.org/10.1017/CBO9781139046855.020.</a></p><p>Elish, Madeleine Clare. “Moral Crumple Zones: Cautionary Tales in Human-Robot Interaction.” <em>Engaging Science, Technology, and Society</em>, vol. 5, Mar. 2019, pp. 40–60, <a href="https://doi.org/10.17351/ests2019.260.">https://doi.org/10.17351/ests2019.260.</a></p><p>Ford, Martin. <em>Rule of the Robots: How Artificial Intelligence Will Transform Everything</em>. Basic Books, 2021, <a href="http://ebookcentral.proquest.com/lib/rit/detail.action?docID=6904994.">http://ebookcentral.proquest.com/lib/rit/detail.action?docID=6904994.</a></p><p>Lake, Brenden M., et al. “Building Machines That Learn and Think like People.” <em>Behavioral and Brain Sciences</em>, vol. 40, 2017, <a href="https://doi.org/10.1017/S0140525X16001837.">https://doi.org/10.1017/S0140525X16001837.</a></p><p>Mitchell, Melanie, and David C. Krakauer. “The Debate over Understanding in AI’s Large Language Models.” <em>Proceedings of the National Academy of Sciences of the United States of America</em>, vol. 120, no. 13, p. e2215907120, <a href="https://doi.org/10.1073/pnas.2215907120.">https://doi.org/10.1073/pnas.2215907120.</a> Accessed 14 Dec. 2024.</p><p>Montreal.AI. <em>AGI Debate</em>. 2022, <a href="https://www.youtube.com/watch?v=JGiLz_Jx9uI.">https://www.youtube.com/watch?v=JGiLz_Jx9uI.</a></p><p>Oxman, Neri. “Age of Entanglement.” <em>Journal of Design and Science</em>, Jan. 2016, <a href="https://doi.org/10.21428/7e0583ad.">https://doi.org/10.21428/7e0583ad.</a></p><p>Real Vision. <em>How Bias Is Built into Algorithms (w/ Cathy O’Neil)</em>. 2019, <a href="https://www.youtube.com/watch?v=iaVThI-z0y0.">https://www.youtube.com/watch?v=iaVThI-z0y0.</a></p><p>Shevlin, Henry, et al. “The Limits of Machine Intelligence: Despite Progress in Machine Intelligence, Artificial General Intelligence Is Still a Major Challenge.” <em>EMBO Reports</em>, vol. 20, no. 10, Oct. 2019, p. e49177, <a href="https://doi.org/10.15252/embr.201949177.">https://doi.org/10.15252/embr.201949177.</a></p><p>“The Age of Scientific Wellness.” <em>Harvard University Press</em>, <a href="https://www.hup.harvard.edu/books/9780674245945.">https://www.hup.harvard.edu/books/9780674245945.</a> Accessed 14 Dec. 2024.</p><p>“Unmasking AI by Joy Buolamwini: 9780593241844 | PenguinRandomHouse.Com: Books.” <em>PenguinRandomhouse.Com</em>, <a href="https://www.penguinrandomhouse.com/books/670356/unmasking-ai-by-dr-joy-buolamwini/">https://www.penguinrandomhouse.com/books/670356/unmasking-ai-by-dr-joy-buolamwini/.</a> Accessed 14 Dec. 2024.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7adcbc5e711f" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>