Gemma ✨🏳️‍⚧️
LightNews LightNews
banner
gemmasparkle.bsky.social
Gemma ✨🏳️‍⚧️
@gemmasparkle.bsky.social
53 followers 84 following 280 posts
she/her
Green Party member
I type too fast and you can't edit so I apologise for the typos.
Posts Replies Media Videos
gemmasparkle.bsky.social
Hmm, maybe ask this guy?
A black and white image of Neville Chamberlain after he signed the Munich Agreement on 30 September 1938 and appeasement of Hitler.
January 5, 2026 at 10:05 AM
gemmasparkle.bsky.social
I would suggest not only cowards but submissives. They enjoy submitting to their doms.
October 8, 2025 at 5:53 PM
gemmasparkle.bsky.social
Even better in person
An image of Zack Polanski in front of a conference podium talking to other members before his speech later on today.
October 3, 2025 at 9:56 AM
gemmasparkle.bsky.social
Even an AI knows this is wrong:
Ethical Implications for Public Engagement: If the public knows their detailed, thoughtful responses are being processed primarily by a machine, it could erode trust in the consultation process and the government's commitment to genuinely listening to citizens. It could devalue the effort people put into crafting their responses if they feel their voice isn't being heard by a human.
August 1, 2025 at 9:14 AM
gemmasparkle.bsky.social
Also asked "Do you see any problems with using AI to automatically review responses?" long responses in alt-text
Yes, there are several significant problems and ethical concerns with using AI to automatically review consultation responses, even with human oversight. While AI offers immense benefits in terms of speed and efficiency for large datasets, these issues must be carefully addressed to ensure the integrity and fairness of the process. Here are the main problems: Bias in Training Data and Algorithms: Historical Bias: AI models are trained on vast amounts of existing text data. If this data reflects historical biases (e.g., certain demographic groups being underrepresented or stereotyped in language), the AI will learn and perpetuate these biases. For example, if past consultations predominantly featured responses from a certain demographic, the AI might unconsciously give more weight to their phrasing or perspectives. Algorithmic Bias: The way algorithms are designed and the parameters they prioritize can inadvertently introduce bias. Even if data is unbiased, the algorithm's structure might lead to discriminatory outcomes. Developer Bias: The assumptions and unconscious biases of the AI developers can seep into the model's design and training. Impact: This can lead to certain viewpoints being over-emphasized, others being downplayed, or even crucial "less frequent points" from minority groups being missed entirely because the AI doesn't recognize them as significant due to a lack of similar patterns in its training data. Lack of Nuance, Context, and Interpretation: Sarcasm, Irony, and Figurative Language: AI struggles to understand the subtle nuances of human language. Sarcasm, irony, metaphors, and other forms of figurative speech can be easily misinterpreted, leading to inaccurate sentiment analysis or thematic categorization. Contextual Understanding: A single word or phrase can have different meanings depending on the context of the entire response or the broader social issue being discussed. AI might miss this crucial context. Emotional Depth: While AI can identify basic sentiment (positive/negative/neutral), it often lacks the capacity to truly understand the depth of emotion, frustration, or passion conveyed in a response, which can be critical for policy-making. Implicit Meanings: People often imply things rather than stating them directly. AI might struggle to pick up on these implicit meanings. "Black Box" Problem and Lack of Transparency: Many advanced AI models (especially deep learning models) are "black boxes," meaning it's difficult to understand why they arrived at a particular conclusion or categorization. This lack of transparency makes it challenging to audit the AI's decision-making process, identify the source of errors or biases, and explain why certain responses were categorized in a particular way. For a public consultation, transparency is vital for public trust. Risk of Over-Simplification and Missing Outliers: While AI is good at identifying common themes, it has a tendency to categorize and generalize. This can lead to unique or outlier responses being dismissed as "noise" or forced into existing categories, even if they represent genuinely novel or important perspectives. The very strength of identifying "less frequent points" relies heavily on careful human training and oversight, as the default tendency of AI might be to prioritize statistical frequency. Accountability and Human Responsibility: If a consultation review process relies heavily on AI and a significant issue is missed or misinterpreted, who is accountable? While the AI provides the analysis, the ultimate responsibility for the consultation outcome rests with the human team. There's a risk of "automation bias," where human reviewers might overly trust the AI's output and fail to critically evaluate it, leading to errors or biases going unchecked. Data Privacy and Security: Consultation responses often contain personal or sensitive information. Feeding this data into an AI system (especially a third-party service) raises significant privacy and data security concerns. Robust data governance, anonymization, and adherence to regulations like GDPR are paramount. Ethical Implications for Public Engagement: If the public knows their detailed, thoughtful responses are being processed primarily by a machine, it could erode trust in the consultation process and the government's commitment to genuinely listening to citizens. It could devalue the effort people put into crafting their responses if they feel their voice isn't being heard by a human. Mitigation is Key: As discussed in the previous answer, these problems don't mean AI shouldn't be used. Instead, they highlight the critical need for: Robust Human-in-the-Loop Processes: AI as an assistant, not a replacement. Careful Training and Validation: Continuously testing and refining the AI with human-coded samples. Explicit Strategies for Outlier Detection: Actively looking for the less frequent and unique points. Transparency: Being open about how AI is being used. Diverse Human Review Teams: To catch biases the AI might miss. Without these stringent safeguards, automatically reviewing consultation responses with AI carries significant risks to fairness, accuracy, and public trust.
August 1, 2025 at 9:12 AM
gemmasparkle.bsky.social
I asked Gemini for help on how to do this fairly. "I'm responsible for reviewing the 50 thousand responses to the EHCR consultation. I want to use Google Gemini to to this. How can I ensure that all responses are treated fairly and even less frequent points are considered?" Long response in alt-text
Reviewing 50,000 responses to a consultation is a massive undertaking, and Google Gemini (or any AI) can be a powerful tool to assist, but it's crucial to understand its strengths and limitations to ensure fairness and capture all nuances. Here's a breakdown of how you can approach this, focusing on fairness and ensuring less frequent points are considered: How Google Gemini (and AI in general) can help: Initial Thematic Analysis and Categorization: Gemini's natural language processing (NLP) capabilities can quickly read through responses and identify recurring themes, keywords, and sentiment. This can provide an initial high-level overview of the most common points raised. Tools like Consult (used by the UK government for similar tasks), ATLAS.ti, and MAXQDA leverage AI for this purpose. Summarization: Gemini can generate concise summaries of individual responses or groups of responses sharing similar themes. This can save immense time for human reviewers. Identifying Sentiment: AI can help classify responses as positive, negative, or neutral towards specific proposals or aspects of the consultation. Flagging Anomalies/Outliers: While focusing on common themes, AI can also be trained to flag responses that are unique, express dissenting opinions, or introduce entirely new concepts not widely discussed. Cross-referencing: If the consultation has specific questions, Gemini can help group responses by question and then analyze themes within each question. Efficiency: Automating initial analysis significantly speeds up the review process, allowing human experts to focus on deeper interpretation and validation. Ensuring Fairness and Considering Less Frequent Points: This is where human oversight, careful planning, and specific AI strategies become paramount. Define Your Objectives and Categories Clearly (Human-led): Before you even start with AI, thoroughly understand the goals of the consultation. What information are you trying to gather? What decisions will this inform? Develop a robust coding framework or taxonomy of potential themes and issues before applying AI. This framework should be comprehensive and ideally developed by a diverse team to minimize unconscious human bias. This will serve as the "ground truth" for the AI. Consider creating categories for "other," "miscellaneous," or "new ideas" to explicitly capture points that don't fit existing themes. Pilot Testing and Iterative Refinement of the AI Model: Start small: Don't throw all 50,000 responses at Gemini at once. Take a representative sample (e.g., a few hundred or a thousand responses) and manually review and code them first. Train and evaluate: Use this manually coded data to train your Gemini model (or an AI tool integrated with Gemini's capabilities) to identify themes and sentiment. Iterate and refine: Compare the AI's output with your human-coded sample. Crucially, pay close attention to responses where the AI and human disagree. This is where biases or missed nuances might be present in the AI. Adjust the AI model's parameters, add specific keywords, or provide more examples to improve its accuracy in identifying less frequent or subtle points. Actively look for instances where the AI might be over-representing dominant views or missing minority opinions. Strategies for Capturing Less Frequent Points: Keyword Extraction beyond Themes: Instead of just looking for overarching themes, use AI to extract all significant keywords and phrases from responses. Then, review the less frequent keywords to see if they represent important, albeit niche, perspectives. Anomaly Detection: Configure the AI to flag responses that are statistically different from the majority. These "outliers" might contain valuable unique insights. Explicit "Other" or "New Idea" Tagging: If your coding framework includes an "Other" category, ensure the AI is trained to identify responses that don't fit established themes and tag them appropriately for human review. Prioritize Diversity in Training Data: Ensure the data used to train the AI model (if you're fine-tuning it) is diverse and representative of the expected range of responses, including those from less vocal groups if possible. Random Sampling for Human Review: Even after AI analysis, regularly conduct random sampling of responses (including those categorized as common themes and outliers) for human review to act as a quality control and ensure no important points were missed. Human Oversight and Validation (Non-negotiable): AI as an Assistant, not a Replacement: AI should augment, not replace, human analysis. Human experts are essential for interpreting nuance, context, sarcasm, and cultural subtleties that AI might miss. Dedicated Review Teams: Assign specific human reviewers to focus on responses that the AI flags as "other," "unique," or those with ambiguous sentiment. Diverse Review Team: Ensure your human review team is diverse in terms of background, perspective, and understanding of the subject matter to reduce individual biases. Cross-Validation: Have multiple human reviewers independently assess a sample of the AI's output to ensure consistency and identify potential misinterpretations. Ethical AI Principles: Adhere to ethical AI guidelines, ensuring transparency about how AI is being used, maintaining accountability for the outcomes, and protecting data privacy. Transparency: Be transparent about the use of AI in the consultation review process. This can build trust with the public. Data Security and Privacy: Consultation responses often contain personal or sensitive information. Ensure that any AI tool or platform you use complies with data protection regulations (e.g., GDPR in the UK). Anonymize data wherever possible before feeding it into the AI. Use secure, reputable AI services that prioritize data encryption and privacy. By combining the efficiency of AI with rigorous human oversight and a well-defined analytical framework, you can ensure that all 50,000 responses to the EHCR consultation are treated fairly, and even less frequent but potentially important points are carefully considered.
August 1, 2025 at 9:09 AM
gemmasparkle.bsky.social
Accurate:
The profile page for MP Wes Streeting which has been blocked by "A Very British Bigotry" for Dehumanizing Rhetoric and TERF.
July 25, 2025 at 11:10 AM
gemmasparkle.bsky.social
When you are voting the same was as Enoch Powell you know you really messed up....
July 23, 2025 at 3:30 PM
gemmasparkle.bsky.social
Seems that "The Trail of Political Consciousness" from Andor is just as relevant to being trans as it was to being a rebel in Star Wars.
There will be times when the struggle seems impossible. I know this already. Alone, unsure, dwarfed by the scale of the enemy. Remember this. Freedom is a pure idea. It occurs spontaneously and without instruction. Random acts of insurrection are occurring constantly throughout the galaxy. There are whole armies, battalions that have no idea that they've already enlisted in the cause. Remember that the frontier of the Rebellion is everywhere. And even the smallest act of insurrection pushes our lines forward. And then remember this. The Imperial need for control is so desperate because it is so unnatural. Tyranny requires constant effort. It breaks, it leaks. Authority is brittle. Oppression is the mask of fear. Remember that. And know this, the day will come when all these skirmishes and battles, these moments of defiance will have flooded the banks of the Empire's authority and then there will be one too many. One single thing will break the siege. Remember this. Try.
July 3, 2025 at 11:53 AM
gemmasparkle.bsky.social
Is this photo unmanipulated? It looks like its been run through a Mad Magazine filter. I'd want my money back from any surgeon that did this to me.
An image of Alfred E. Neuman, the mascot and cover boy of the American humour magazine Mad
July 2, 2025 at 1:09 PM
gemmasparkle.bsky.social
Open Source right? I'm reminded of this from xkcd.com/2347/ and the fact that so many open source projects are changing their license simply because people don't contribute enough to make it anything more than a hobby and its a thankless job to be a maintainer. Most are kept alive by corp interests.
[A tower of blocks is shown. The upper half consists of many tiny blocks balanced on top of one another to form smaller towers, labeled:] All modern digital infrastructure [The blocks rest on larger blocks lower down in the image, finally on a single large block. This is balanced on top of a set of blocks on the left, and on the right, a single tiny block placed on its side. This one is labeled:] A project some random person in Nebraska has been thanklessly maintaining since 2003
July 2, 2025 at 9:19 AM
gemmasparkle.bsky.social
Slightly different set of storm names... A bit more personal...
A form filled out suggesting surnames of the CEOs of oil companies as storm names.... to celebrate the storm makers.
June 26, 2025 at 12:07 PM
gemmasparkle.bsky.social
Is is Tiamut? #marvel #marveleternals
June 25, 2025 at 11:52 AM
gemmasparkle.bsky.social
I finally finished the last two episodes of "What it feels like for a girl" last night. The prison episode is extremely tough viewing but 10/10 for the ending and overall show. I had a little cry when it finished and am so happy for her.
Marketing photo for "What is feels like for a girl" on BBC. It features the lead character, Byron, in a club with under green lighting. She is fairly androgynous with short hair but feminine makeup and a large spiked collar.
June 22, 2025 at 2:10 PM
gemmasparkle.bsky.social
I think you are talking about a closed sandwich and I don't think its possibly in the ideal flat shape since its not possible to flatten against itself. You could create one with two strips though. Tried to visualize but concept of a Möbius strip seems too complex for Gemini. Not bad though.
An image of a sandwich where the bread is in the shape of a Möbius strip so as to have the bread slices each forming both sides of the sandwich. This version is a tightly wrapped curve with slices of pride colour dyed meat. An image of a sandwich where the bread is in the shape of a Möbius strip so as to have the bread slices each forming both sides of the sandwich. This is version is a figure-eight with pride coloured fillings. An image of a sandwich where the bread is in the shape of a Möbius strip so as to have the bread slices each forming both sides of the sandwich. This version is more bagel shaped with pride coloured fillings.
June 20, 2025 at 11:11 AM
gemmasparkle.bsky.social
Is it wrong for someone in their 50's to want one of these because of what it represents?
Image of Ikea's BLÅHAJ shark plushie, a symbol for trans folk.
May 12, 2025 at 7:41 PM