You are currently browsing the category archive for the ‘Technology and Computers’ category.
In the annals of human ingenuity, steel forged before the nuclear age—untainted by radioactive fallout—holds a revered place. Prized for precision instruments like Geiger counters, this “low-background steel” is scarce, salvaged from shipwrecks to avoid the contamination of modern alloys. So too is human-generated data: raw, diverse, and grounded in lived experience, it once fueled the internet’s vibrant ecosystem. Yet, as artificial intelligence (AI) proliferates, a troubling parallel emerges—the “cold-steel problem.” AI, increasingly trained on its own synthetic outputs, risks a self-referential spiral, eroding the authenticity and diversity of information. Like steel laced with radiation, AI-generated data threatens to corrode the tools of knowledge, leaving us with a homogenized, unreliable digital landscape.
The pre-AI era offered a rich tapestry of human thought—letters, books, forums, and early websites brimmed with unfiltered perspectives. These were the “cold steel” of data: imperfect, often chaotic, but rooted in reality. Today, AI’s insatiable appetite for content—web-scraped, algorithmically churned—has shifted the balance. A 2024 Nature study warns of “model collapse,” where AI trained on synthetic data loses the nuanced “tails” of human experience, converging toward bland, repetitive outputs. Wikipedia, once a bastion of human collaboration, now grapples with AI-generated articles—5% of new English entries in 2024 bore hallmarks of automation, often shallow and poorly sourced. This isn’t mere noise; it’s a distortion, amplifying errors and biases with each recursive loop, like a photocopy of a photocopy fading into illegibility.
The mechanics of this spiral are insidious. AI models, fed on web data increasingly tainted by their own outputs, risk “Model Autophagy Disorder” (MAD)—a vivid term for systems consuming themselves. A 2017 self-driving car crash, caused by mislabeled data failing to distinguish a truck from a bright sky, illustrates the stakes: errors compound, reality distorts. Posts on X lament search engines returning AI-crafted drivel—slick but soulless—while human voices struggle to break through. The counterargument, that synthetic data fills gaps in niche domains like coding, holds limited weight. Even in verifiable fields, the loss of diverse, human-generated inputs risks outputs that are technically correct but creatively barren, a digital equivalent of bollocks masquerading as insight.
The implications are stark: an information ecosystem choked by self-referential sludge threatens not just AI’s utility but society’s capacity for truth-seeking. If unchecked, this spiral could render knowledge a hollow echo chamber, antithetical to the vibrant complexity of human thought. Mitigation demands urgency—prioritizing human-curated datasets, enforcing transparency in data provenance, and developing tools to filter AI’s footprint. Blockchain-based data authentication or crowd-sourced verification could anchor AI in reality, preserving the “cold steel” of human insight. Yet, these solutions require collective will, a resistance to the seductive ease of automation’s churn. Without action, the fallout risks a digital dark age where truth drowns in synthetic noise.
The cold-steel problem is no mere technical glitch; it’s a philosophical reckoning. AI, for all its prowess, cannot replicate the spark of human creativity or the grit of lived experience. As we stand at this precipice, the choice is clear: safeguard the authenticity of human data or surrender to a future where information is a pale shadow of its potential. The shipwrecks of our pre-AI past hold treasures worth salvaging—not just for AI’s sake, but for the soul of our shared knowledge. Act now, or the corrosion of our digital ecosystem will be a legacy of our own making.
Sources
- Shumailov, I., et al. (2024). AI models collapse when trained on recursively generated data. Nature, 631, 755–759. https://www.nature.com/articles/s41586-024-07566-y[](https://www.nature.com/articles/s41586-024-07566-y)
- Alemohammad, S., et al. (2024). Self-Consuming Generative Models Go MAD. International Conference on Learning Representations (ICLR). https://news.rice.edu/news/2024/breaking-mad-generative-ai-could-break-internet[](https://www.sciencedaily.com/releases/2024/07/240730134759.htm)
- Model collapse. (2024, March 6). Wikipedia. https://en.wikipedia.org/wiki/Model_collapse[](https://en.wikipedia.org/wiki/Model_collapse)
- Rice University. (2024, July 30). Breaking MAD: Generative AI could break the internet, researchers find. ScienceDaily. https://www.sciencedaily.com/releases/2024/07/240730134750.htm[](https://www.sciencedaily.com/releases/2024/07/240730134759.htm)
- Kempe, J., et al. (2024). A Tale of Tails: Model Collapse as a Change of Scaling Laws. International Conference on Machine Learning (ICML). https://nyudatascience.medium.com/overcoming-the-ai-data-crisis-a-new-solution-to-model-collapse-2d36099be53c[](https://nyudatascience.medium.com/overcoming-the-ai-data-crisis-a-new-solution-to-model-collapse-ddc5b382e182)
- Shumailov, I., et al. (2023). AI-Generated Data Can Poison Future AI Models. Scientific American. https://www.scientificamerican.com/article/ai-generated-data-can-poison-future-ai-models/[](https://www.scientificamerican.com/article/ai-generated-data-can-poison-future-ai-models/)
AI is giving Moore’s Law law a run for its money.
1 Month (June 2025)
Over the next month, AI advancements are likely to focus on incremental improvements in existing models and broader accessibility. Companies like xAI, OpenAI, and Anthropic will continue optimizing large language models for efficiency, reducing computational costs, and enhancing performance in tasks like natural language processing, code generation, and image analysis. More seamless integration of AI tools into consumer platforms, such as enhanced voice assistants or improved content recommendation systems, is expected. Regulatory discussions, particularly in the EU and US, will gain momentum, with early frameworks for AI safety and ethics shaping deployment strategies.
3 Months (August 2025)
By August 2025, AI systems are expected to exhibit more robust multimodal capabilities, combining text, image, and possibly audio processing. Models like Grok 3 may see expanded features in real-time information synthesis. Increased adoption in industries like healthcare (e.g., diagnostic tools) and education (e.g., personalized learning platforms) is likely. Open-source AI frameworks may mature, enabling smaller organizations to deploy sophisticated models. Stricter guidelines addressing data privacy and algorithmic bias will emerge, fostering trust but potentially slowing some deployments.
6 Months (November 2025)
By November 2025, AI could demonstrate improved reasoning abilities, tackling complex problems in scientific research or logistics with greater autonomy. Advancements in reinforcement learning may optimize supply chains or energy grids. Consumer-facing AI, like voice modes, might expand to new platforms with more natural interactions. Global ethical AI standards will likely solidify, addressing issues like deepfakes. Developments will be driven by increased computational power and collaborations, though challenges like energy consumption and geopolitical tensions may arise.
Reference List for AI Development Forecast 2025
- McKinsey & Company. (2024). The state of AI in 2024: Gen AI adoption accelerates as organizations pursue value. McKinsey & Company.
- Provides insights into AI adoption trends, including multimodal capabilities and industry applications like healthcare and education.
- Stanford Institute for Human-Centered Artificial Intelligence. (2024). AI Index Report 2024. Stanford University.
- Offers data on AI performance improvements, computational efficiency, and global regulatory efforts.
- European Commission. (2024). The Artificial Intelligence Act. European Union.
- Details the EU’s regulatory framework for AI, influencing global safety and ethics standards.
- Gartner. (2024). Top Strategic Technology Trends for 2025. Gartner, Inc.
- Highlights AI-driven automation, multimodal AI, and reinforcement learning applications in logistics and energy.
- World Association of News Publishers. (2024). Ethical AI in Media: Addressing Deepfakes and Misinformation. WAN-IFRA.
- Discusses emerging standards for combating AI-generated misinformation.
- IEEE Spectrum. (2024). The Race for General Intelligence: AI’s Next Frontier. IEEE.
- Explores advancements in reasoning and general intelligence in AI systems.
- xAI. (2024). Grok 3 Product Documentation. xAI. Retrieved from https://x.ai/grok.
- Provides details on Grok 3’s capabilities, including DeepSearch and voice mode, informing consumer AI trends.
Note: Some references are based on projected trends from 2024 sources, as specific 2025 data is not yet available. All sources were accessed for their relevance to AI development, regulation, and application trends as of May 18, 2025.
Hey folks, today’s a show-and-tell on how AI can cut through the world’s noise to find what’s real. Full credit: I’m co-writing this with Grok AI. We’ll use a hypothetical example, but this is a nuts-and-bolts guide—let AI do the heavy lifting while you nail the argument.
In a sea of hot takes and half-truths, spotting dodgy narratives is a superpower. AI can help—here’s how, step by step. Imagine this:
**Example (X, March 2025):**
‘New study proves electric cars emit MORE carbon than gas cars—EVs are a scam!’
(Viral post, 50k likes, links to a blog ‘study.’)
**Step 1: Test the Core**
Ask AI: ‘Is this true?’ I’d check IPCC or Argonne Lab data and say: Nope, lifecycle studies show EVs emit less CO2, even with battery costs. Shaky start.
**Step 2: Dig into the Source**
Tell AI: ‘Check the link.’ The ‘study’ is a 10-page PDF from an oil lobby—zero peer review, cherry-picked stats. Compare that to MIT’s 2024 EV report: open data, real methods. Night and day.
**Step 3: Call Out the Hype**
Ask: ‘What’s overblown?’ ‘Scam’ skips context—like grid energy (coal vs. solar). It’s a sledgehammer, not nuance. AI spots the bait.
**Step 4: Keep It Cool**
AI sums it up: ‘Battery production has a carbon hit, but EVs still beat gas cars overall. Not a scam—just not perfect.’ Facts, no fuss.
**Why It Works**
This—claim, source, hype, rebuttal—keeps you sharp. AI sifts fast, stays calm, and frees you from the weeds. Got a wild claim from your news feed or X? Try these steps on it—share what you find. Truth beats outrage every time!”
It is really amazing the lengths people will go through to confirm their victimhood identities. And of course, the CBC will highlight how awesome it is to use AI to ‘make the internet a safer place for Indigenous people’.
Good lord. If the bad internet is hurting you…turn it off. But rather than make an adult decision, let’s do this:
“A new tool aims to use artificial intelligence to help make the internet a safer place for Indigenous people.
The project was given the name wâsikan kisewâtisiwin, which translates to “kind energy” in Cree.
“We’re trying to make the internet a kinder place. We’re trying to change the trajectory of the internet towards discriminated people,” Shani Gwin told CBC’s Radio Active.”
On the internet you are (with certain measures) essentially anonymous. What you say on the internet will be taken at face value (in theory).
“Being developed in collaboration with the Alberta Machine Intelligence Institute (AMii), the tool is dual purpose, intended to help both Indigenous people and non-Indigenous Canadians reduce racism, hate speech, and online bias.
The first function of the program is to moderate online spaces like comment sections. While the internet has been a tool used by Indigenous people for advocacy, it also can frequently be an unsafe space for communities that are discriminated against, Gwin said.
Gwin said that all it takes is one comment for online spaces to fester.”
If people want to pillow-up a spot on the internet, they are more than welcome to do so. Usually though, this sort of anti-free speech mechanism escapes from its hug-box confines and is loosed into the wider ecosystem.
“The tool flags hateful comments, and then provides sample responses, while also documenting these instances for future reporting.
The second function of the tool is designed to serve as a writing plug-in for your computer — similar to Grammarly. Intended to help general Canadians understand their bias, it will flag any writing that may be biased against Indigenous people, provide an explanation, and a suggestion for how to reword the sentence.”
Wow! It is like having your own personal Big Brother making sure that you are engaged in ‘right-thinking’ at all times. Plus offering real time suggestions on how to neuter your speech as to not risk offence to others.
“AI right now is designed through the lens of Canada’s dominant culture. And I would say that across the world that without input from racialized communities, including Indigenous people, AI cannot analyze and produce culturally safe and respectful content,” Gwin said.
“Every piece of infrastructure in Canada has been developed from the white patriarchal lens,” she said. “So more racialized people, more women need to get involved in the development of AI so that it doesn’t continue to be built in a way that’s going to harm us again.”
Whoops! Did you catch the turn into Marxist Critical Theory? I certainly did – That damn AI developed through the lens of ‘dominant culture’. Beginning with a conclusion and then looking for evidence based on your assumptions almost always leads to bullshit results.
Just no. AI was developed by a diverse body of people from across the world, let’s not shoehorn your ‘critical perspective’ into this.
“AI bias revealed itself in training, Qroon said, adding that at times when experimenting with the AI, it would try to minimize the tragedies that Indigenous people went through.
“And that’s why it was very important for us to integrate the Indigenous community into this process and get their perspective and get the instructions from them.”
The AI making the decision to not follow a trauma informed narrative? Huh. Well that will need to be fixed ASAP.
“Gwin said that her hope for the project is that it helps take the emotional labour of education off Indigenous people — and free them up to do things besides moderating comment sections.
“I think there might be concerns that people think that this AI tool will take jobs away from Indigenous people, but it’s not, that’s not what it’s for. It’s there to do the work that we don’t want to do.”
Yes, censorship is such an emotional labour. Much better to let a machine – an entity with even less capacity for nuance – take the reins.
“But it also means changing the internet and Canadians’ hearts and minds about who Indigenous people are.”
You mean changing minds in a positive way, right? Because this just looks like social and emotional manipulation in service of maintaining a oppressed/oppressor narrative that benefits no one in Canada.
A deliciously wriggling can of worms this topic is. I lean toward the answer being yes, but having rights in our society isn’t a guarantee of justice or fairness. I would hope that by the time sentient AI becomes a thing, we have our own house in order so we can be a good example to our AI children.
Safety? Security? – A window into our fearful past and what we’ve lost of our privacy.
bleatmop
Mystro
Dead Wild Roses 
Your opinions…