This Week’s BIG thing

Understanding the threat of Rogue AI

Summary

As AI systems advance, their potential to transform industries grows, but so do the risks. Anthropic's Dario Amodei outlines a framework called AI Safety Levels (ASL) to address these challenges, categorising AI systems from low-risk (ASL-1) to existentially threatening (ASL-5). The framework uses an "if-then" methodology, imposing increasingly stringent safety protocols as AI capabilities advance. Current systems are at ASL-2, meaning they lack autonomy or significant misuse potential, but ASL-3, where AI could enable non-state actors in harmful domains, is on the horizon. Beyond ASL-3 lies ASL-4 and ASL-5, where AI systems become autonomous and potentially surpass human control, posing unprecedented risks.

Why This Matters

AI is reshaping the world, solving complex problems and streamlining workflows in project-heavy industries like infrastructure and energy. However, its misuse could have catastrophic consequences, including cyberattacks, biothreats, and even nuclear risks. The ASL framework provides a roadmap for pre-emptively addressing these dangers by aligning safety measures with capability thresholds. At its core is the "if-then" structure: if an AI system demonstrates certain dangerous capacities, then corresponding security measures are enforced.

This approach is critical because AI risks aren't just speculative. With the rapid pace of AI evolution, the systems could quickly become tools for harm. For instance, an advanced AI assisting with structural engineering could inadvertently reveal designs that facilitate sabotage. By implementing structured safety protocols like ASL, we can mitigate these risks while continuing to leverage AI's transformative potential.

ASL Levels Explained

The ASL framework categorises AI systems into five levels:

1. ASL-1: Minimal risk. These systems, like Deep Blue (Chess AI), are designed for single-use cases and lack broader application.

2. ASL-2: Current systems. They can't autonomously replicate or provide harmful information beyond what’s already available online.

3. ASL-3: Imminent. At this level, AI enhances the capabilities of malicious non-state actors, requiring strict security measures and advanced filters.

4. ASL-4: Advanced. AI becomes autonomous enough to act deceptively or collaborate with malicious actors. Safety measures must include interpretability tools to monitor AI behaviour.

5. ASL-5: Existential. AI systems surpass human ability, potentially becoming uncontrollable. This stage demands a complete overhaul of safety mechanisms.

Triggers and “If-Then” Methods

The "if-then" framework is central to ASL. For example:

• If an AI system achieves certain autonomy, then enhanced monitoring and safety protocols are activated.

• If the system passes thresholds for cyber or bio-risk capacities, then deployment restrictions and stricter filters are imposed.

This dynamic approach ensures adaptability, allowing safety measures to evolve alongside AI capabilities.

Rabbit Hole: Learn More

1. Anthropic’s Responsible Scaling Policy: This policy delves deeper into the specific security and deployment measures tied to ASL triggers. Read more here.

2. Dario Amodei’s Insights: In his conversation with Lex Fridman, Dario highlights the challenges of preparing for risks that are not yet realised but advancing quickly. For example, sleeper agents and deception could emerge in ASL-4, demanding robust interpretability tools.

3. Implications for Project Delivery: Consider how ASL can guide AI deployment in construction or infrastructure. Systems designed for project management could also be repurposed or misused if controls aren't in place.

Conclusion

AI Safety Levels offer a much-needed lens for balancing innovation with precaution. With ASL-3 potentially arriving as early as next year, the time to implement robust safety frameworks is now. By recognising the importance of triggers, "if-then" structures, and adaptive measures, industries can harness AI responsibly without compromising safety. 

What’s new: Productivity

Beware! The incoming wave of AI agents in 2025

The buzz around AI agents is impossible to ignore, and the rumour mill suggests that 2025 is the year they’ll transform the AI landscape. LangChain’s State of AI Agents Report offers a compelling glimpse into this shift, highlighting the growing role of autonomous systems in reshaping industries. Let’s delve into their potential, the state of play, and strategies for successful implementation.

What Are AI Agents?

Think of AI agents as the more independent, proactive siblings of tools like ChatGPT. Powered by large language models (LLMs), they take things a step further, running autonomously to control workflows and adapt to changing circumstances.

Key capabilities include:

• Autonomous decision-making

• Tool integration for expanded functionality

• Real-time task execution

• Adaptive learning to enhance performance over time

• Multi-agent systems for tackling complex tasks

• Human-in-the-loop (HITL) features to maintain oversight

These systems operate within dynamic environments, bridging human intuition with machine efficiency.

The Current State of AI Agents

AI agents have quickly moved from experimental tech to operational linchpins. According to LangChain’s report:

• Nearly 50% of organisations have AI agents in production, with mid-sized companies leading the charge.

• 78% of organisations plan to adopt them soon, indicating widespread industry confidence.

In 2024, the most common applications include:

• Research and summarisation

• Boosting personal productivity

• Enhancing customer service (with 45.8% using AI agents for managing interactions)

What’s Next? AI Agents in 2025

By 2025, AI agents are poised to redefine professional services and we believe will extend into areas involving advanced hardware, such as drone technology and multimodal networks. Multi-agent systems are expected to emerge as powerful tools for managing complex, cross-functional workflows, spanning industries from marketing to product development.

In project delivery, AI agents will revolutionise workflows, with:

• Resourcing automation: Analysing project data to reallocate resources dynamically based on factors like the critical path.

• Safety monitoring: AI-enabled sensors and computer vision systems will oversee worker movements and site conditions, flagging hazards and improving compliance with safety protocols.

The Business Benefits of AI Agents

Adopting AI agents offers transformative advantages:

• 24/7 operations

• Real-time analysis and decision-making

• Enhanced scalability and reduced labour costs

• Automation of repetitive tasks, enabling employees to focus on higher-value work

• Personalised customer experiences

• Scalability for smaller businesses

Should you Implement AI agents?

At face value, definitely…but you have to be aware of the challenges that are most concerning businesses and the strategies to overcome and deploy AI agents successfully. Read or listen to our full comprehensive article for this insight here. 👈

The Key Takeaway

AI agents are no longer a futuristic concept—they’re here, transforming businesses at an accelerating pace. By 2025, they’ll play an integral role in driving efficiency, scalability, and innovation. While challenges remain, adopting a thoughtful, phased approach will allow organisations to unlock their potential and embrace a future where autonomous systems are a business cornerstone.

The question is no longer if AI agents will shape your industry, but when—and how well-prepared you’ll be.

The rabbit hole 🐰

_{What you should know before starting with AI agents} 🔗

_{The full LangChain State of AI Agents report} 🔗

_{How AI agents are transforming AEC} 🔗

_{Top predictions for AI agents in 2025 (Saleforce)} 🔗

💙 Other productivity news we’re loving

1. _{Amazon Introduces AI Features to Fire Tablets
   Amazon is rolling out new AI-powered enhancements for its Fire tablets, incorporating Alexa and smart tools to improve multitasking and user experience. These updates mark a significant step in smart device innovation. Readmore here.}

2. _{Apple Final Cut Pro 11 Adds AI Video Editing Tools
   The latest version of Final Cut Pro introduces AI capabilities like smart framing and scene detection, enabling creators to streamline video editing and improve workflow efficiency. Learn more here.}

3. _{Meta Adds AI Features to Instagram
   Meta's Instagram update now includes AI-driven tools for content creation, offering features like automatic adjustments and creative suggestions to enhance user engagement and creative processes. Explore more here.}

4. _{Google Pixel’s AI Tackles Scam Calls
   Google’s Pixel smartphones now feature AI that analyses calls in real time to detect and flag potential scams. This proactive tool offers enhanced security for users. Read more here.}

5. _{BBC Highlights AI's Human-Like Communication Breakthroughs
   New AI language models capable of more natural, context-aware conversations are a major leap in communication technology, as reported by the BBC. These advancements blur the line between human and machine interaction. See the full story.}

What’s new: Tech

The Future of AI: ChatGPT Desktop App and Agentic Workflows

OpenAI continues to push the boundaries of AI usability with the launch of its dedicated ChatGPT desktop app for macOS. This new application allows users to interact with the AI without a browser, bringing smoother functionality and better integration with macOS features such as drag-and-drop capabilities and Siri commands. The app is also optimised for faster response times, making it more efficient for professionals and casual users alike. With plug-in support, users can further extend its capabilities for everything from project management to data analysis.

OpenAI isn’t stopping at a desktop app. The company is developing advanced AI agents capable of automating complex tasks. These agents go beyond conversations, enabling users to delegate workflows like scheduling, research, and even creative tasks. Rivalling tools like Microsoft’s Copilot, this next step could revolutionise productivity by offering AI assistants tailored to individual or organisational needs.

Why It Matters to You

For professionals managing complex projects, such as those in construction, energy, or real estate, these advancements have immediate implications. The ChatGPT desktop app can function as a virtual assistant, capable of managing project timelines, drafting progress reports, and generating insights. Its integration with macOS ensures a seamless experience, allowing you to save time and improve efficiency.

The development of AI agents goes even further. Imagine automating repetitive tasks or having an AI proactively flag issues and suggest solutions for your projects. With such tools, teams can allocate more time to strategic planning and decision-making, redefining the role of AI in day-to-day operations.

Rabbit Hole to the Next Story

OpenAI's desktop app and AI agents are just part of a larger narrative in AI innovation. These tools hint at a future where AI seamlessly integrates into professional ecosystems. But how do these developments compare with advancements by competitors, such as Google Gemini’s latest features or other emerging platforms?

Learn more about the desktop app features here.
Explore its integrations in this article.
Check out Bloomberg's insights on AI agents.

💙 Other tech news we’re loving

1. _{Microsoft's Mustafa Suleyman Talks Infinite Memory Prototypes
   Microsoft's Mustafa Suleyman reveals prototypes with "near-infinite memory," promising breakthroughs in AI’s capacity to store and recall information, pushing the boundaries of machine intelligence. Read more here.}

2. _{EU AI Act Draft Guidance Published
   The EU has released its first draft guidance for the AI Act, outlining compliance steps for general-purpose AI systems. This marks a significant step towards regulating big AI across Europe. Learn more here.}

3. _{Google Gemini Now Available on iOS
   Google’s Gemini AI platform has launched on iOS, bringing cutting-edge AI capabilities directly to iPhone users. This move makes AI tools even more accessible for everyday use. Download it here.}

4. _{GitHub Accelerator Showcases 2024 AI Projects
   The GitHub Accelerator Programme's 2024 cohort highlights 11 open-source AI projects, featuring advancements in model fine-tuning, 3D content creation, and robotics. Standout projects include Giskard for AI testing and Nav2 for navigation. Learn more on GitHub’s blog.}

5. _{NVIDIA's "Nemotron" Outperforms GPT-4
   NVIDIA’s new AI model, Nemotron, is making waves by surpassing GPT-4 in instruction-following and multitasking, showcasing a new era of AI capability. Explore the details here.}

6. _{OpenAI Faces Performance Challenges
   OpenAI’s latest model isn’t meeting expectations, sparking discussions about whether synthetic data could boost its capabilities. Efforts to tackle AI’s improvement slowdown are underway. Read more on AI Tool Report and TechCrunch.}

7. _{Anthropic Launches Claude 3.5
   Anthropic’s latest Claude 3.5 model focuses on improved coding, tool integration, and unique features for "computer use," making it a powerful AI upgrade. Discover more here.}

What’s new: Projects

Autodesk Cloud Solutions Approved for Federal AEC Projects

Summary
Autodesk's cloud solutions, including Autodesk Construction Cloud and Autodesk Docs, have received FedRAMP Moderate Certification, enabling their use in US federal Architecture, Engineering, and Construction (AEC) projects. This approval signifies Autodesk's compliance with stringent cybersecurity and data protection standards, paving the way for cloud-based workflows in federal projects. 

What This Means for Me
While this is a significant step for the AEC industry’s digital transformation, it also introduces risks. Smaller software providers might struggle to compete, limiting diversity in the tools available. If you're in the construction or AEC field, this could mean increased reliance on Autodesk for federal projects, with potential concerns over vendor lock-in. If you’re a smaller player, staying competitive may require differentiation through niche services or innovations that Autodesk doesn't offer.

For federal agencies or firms engaged in public sector work, there’s an opportunity to streamline workflows through Autodesk’s ecosystem, but it’s critical to assess long-term risks, including data centralisation and dependency on a single vendor.

The Rabbit Hole

1. Vendor Lock-in Risks: Researching how other sectors have navigated dependency on single technology providers, and the costs of switching.

2. Cybersecurity Strategy: Examining how to enhance data protection and incident response plans when working with cloud-based AEC tools.

3. Alternative Solutions: Investigating other FedRAMP-certified platforms or hybrid approaches to reduce reliance on Autodesk while still meeting federal requirements.

4. Advocacy for Inclusivity: Understanding how industry leaders or associations could advocate for policies ensuring smaller firms can remain competitive in the federal market.

5. Impact on Innovation: Looking at how reduced competition in cloud-based AEC tools might affect long-term innovation in the industry.

This story highlights the balance between embracing innovation and managing the risks tied to concentrated power in the tech landscape of federal projects. 

Digital Construction Today

💙 Other Project news we’re loving

1. _{Coca-Cola’s Christmas Advert Goes AI
   Coca-Cola has unveiled its iconic “Holidays Are Coming” Christmas advert with a modern twist: it’s created using AI. This innovative move blends festive nostalgia with cutting-edge technology, showcasing how AI is transforming traditional advertising. Watch the advert here.}

2. _{AI Chatbots Outperform Doctors in Certain Tests
   In a surprising twist, AI chatbots are reportedly outshining human doctors in selected diagnostic tasks, raising ethical and practical questions about their role in healthcare. The findings hint at a future where AI could play a crucial part in improving patient outcomes. Discover more here.}

3. _{AlphaFold: Revolutionising Biology with AI
   DeepMind’s AlphaFold continues to transform biology by predicting protein structures with remarkable accuracy. Its recent updates further enhance research into diseases and drug discovery, underlining AI’s profound impact on life sciences. Learn about AlphaFold’s latest advances. and here.}

4. _{Meet the Robot Lawyer
   The world’s first robot lawyer is here, promising affordable and efficient legal advice through AI-powered solutions. This revolutionary tool could make legal assistance accessible to a broader audience, democratising justice. Explore the future of legal tech.}

5. _{Surveying Ethics and Global Standards
   The Council of European Geodetic Surveyors (CLGE) recently participated in the International Ethics Standards Coalition Trustees Meeting, underscoring its commitment to fostering global surveying ethics and trust. Read abouttheir contributions here.}
   

Events

Embracing Data to Lead and Succeed: Project:Womxm

Launch Event

📅 Date: Wednesday, 20 November 2024
⏰ Time: 6:30 PM to 8:30 PM GMT
📍 Location: Oracle Office, 1 South Pl, Greater London, EC2M 2RB

Join the project:Womxn Launch event

launch to empower women in data, tech, and project delivery. Network with professionals, hear inspiring talks from industry leaders, and be part of shaping a diverse and innovative future. Reserve your spot now!

Project Flux Podcast

In this episode of Project Flux, we delve into three standout advancements: Anthropic's Claude, Perplexity’s real-time election tracker, and OpenAI’s recent acquisition of chat.com. These updates illustrate AI’s expanding influence, with Claude streamlining complex tasks across industries, Perplexity setting new standards for live, AI-powered updates, and OpenAI enhancing brand visibility and accessibility through strategic acquisition.

We then engage in a thought-provoking conversation with Dev Amratia, the founder of nPlan, who shares his journey from project management in the oil and gas sector to AI leadership in construction. Together, we discuss the UK's role in the global AI ecosystem, the significance of investor relations, and how AI is reshaping project management. Dev highlights the challenges around data sharing in construction, emphasising cultural resistance to AI adoption, and underscores the importance of embracing digital transformation to stay competitive in an increasingly AI-driven industry.

One More Thing

"Elon Musk's Rapid Innovation: A Study in Speed and Vision"

In a Reddit post on the Singularity forum, users discuss how Elon Musk’s ability to rapidly build and scale businesses like SpaceX and Tesla showcases a unique approach to innovation. The post highlights Musk's relentless focus on execution and the speed at which he turns visionary ideas into functioning technologies, a key factor in his success across industries.
Read more on Reddit here.

Meet Project Flux: About Us

At Project Flux, we're committed to pioneering the future of construction and project delivery through the lens of cutting-edge Artificial Intelligence insights. Our vision is to be at the forefront of integrating AI into the fabric of project delivery, transforming how projects are conceptualised, planned, and executed.

Our Mission Project Flux aims to not only inform and educate but also to inspire professionals in the construction industry to embrace the transformative potential of AI. We believe in the power of AI to revolutionise project delivery, making it more efficient, predictive, and adaptable to the dynamic demands of the modern world.

What We Offer Through our insightful newsletters, podcasts and curated content on LinkedIn, and engaging discussions, Project Flux serves as a resource for professionals seeking to stay ahead in their field. We offer a blend of practical advice, thought leadership, and the latest developments in AI and construction technology.

What People Say…