Weekend Briefing: 20 January 2024

This Week from Artificiality:

Read/Write: A Brief History of Intelligence

Evolution, AI, and the Five Breakthroughs That Made Our Brains by Max Bennett.

Key Points:

• Max Bennett, an AI entrepreneur, tells an engaging story of the brain's evolutionary journey towards intelligence and parallels for AI development. The core idea is that until AI systems replicate each part of our brain’s evolution, AI systems will fail to exhibit human-like intelligence.
• Bennett argues that to unlock AI's full potential, systems should mirror biological evolution including key "breakthroughs" like learning from emotional valence, forming temporal predictive models, simulating possible scenarios, attributing mental states, and accumulating knowledge through language.
• Biological intelligence progressed in 5 key "breakthroughs"—steering, reinforcing, simulating, mentalizing, speaking. These built progressively more advanced functions like imagination and theory of mind.
• Open questions remain around precisely which higher cognitive capabilities biological intelligence possesses that advanced AI still lacks. And how can we formally evaluate more complex facets like curiosity, imagination and metacognition beyond human benchmarks?
• He surmises a speculative "6th Breakthrough" of a decoupling of intelligence from biology.
• We question however whether the most revolutionary AI systems may manifest differently than human mimicry or biology all together—embracing alternate pathways reflecting the technology's non-biological essence.

Rodrigo Liang: SambaNova and Generative AI in the Enterprise

An interview with SambaNova Systems CEO, Rodrigo Liang.

Key Points:

• SambaNova Systems aims to help companies adopt AI technology, particularly in enterprise environments. It provides full-stack AI solutions including hardware, software, models, etc. to simplify adoption.
• The company’s offerings are designed to enable companies to leverage AI while maintaining data privacy and security. A modular approach provides the flexibility to adapt to diverse enterprise needs.
• SambaNova takes an "AI asset" approach focused on creating long-term value rather than just providing "AI tools." A focus on open source models provides diversity of technology while reducing vendor lock-in.
• The company’s software stack enables fine-tuning of granular models on customer data, creating a multitude of AI experts to serve the enterprise. Unlimited use encourages experimentation without the same cost challenges as public cloud AI.

Bonds with Bots

Does consciousness attribution hold the key to funding AI futures? The shift towards AI intimacy might reclaim our attention but at the potential cost of human-to-human connections.

Key Points:

• Research shows more ChatGPT usage correlates with higher attributions of consciousness and emotional connection.
• 2/3 of users attribute some subjective awareness to models—especially the capacity for sensation/emotion rather than raw intelligence.
• Increased AI interaction encourages anthropomorphism and perceived intimacy.
• This shifts basis of valuation from capability towards emotional bonding and relationships.
• Willingness-to-pay may rely more on simulated intimacy than technological capability alone.

Is the GPT Store AGI in development?

We speculate on how the GPT store might act a market mechanism for discovering the most valuable use cases for AGI.

Key Points:

• OpenAI aims to evolve GPTs to be more "agentic" and the GPT store provides a mechanism to discover valuable real-world use cases to guide this evolution.
• Conversation analysis shows only a few GPTs dominate usage, following a power law distribution where a few hubs have many links. Only relatively few GPTs will influence the store's ecosystem evolution.
• In a GPT store with agentic apps, competition and natural selection pressures could accelerate novel capabilities.
• As agentic GPTs interact, they generate data on dynamic human preferences and effective learning processes valuable for developing general AI.
• The network dynamics could mirror aspects of natural intelligence like learning from experience, crucial for advanced AI.
• The network topology itself could give rise to new learning algorithms.

How to Talk with Your Teens about AI

Today, we hosted a webinar on how to talk with your teens about AI.

Key Points: 

• Generative AI holds great promise with its ability to uniquely access and combine all of recorded digital human history, but it also fabricates information and cannot be fully relied upon for accuracy. Managing expectations is important.
• Bias is absorbed into AI from human data and is difficult to fully remove. Generative AI risks perpetuating stereotypes and inequalities if not addressed properly in system design and use.
• With the ability for generative AI to form an intimate understanding of human wants and needs, there are open questions around how this could be monetized in the future.
• For education, while fears initially focused heavily on increased cheating, rates have not gone up. Most students believe AI should only be used to generate ideas or edit small portions of work. Guidelines are needed on proper integration into academics to support, not hinder, learning.
• While risks exist in how generative AI could negatively impact society if misused, these systems also provide extraordinary new opportunities for creativity, combinatorial thinking, gaining new perspectives, and more. An open, ongoing dialogue is important for developing shared understanding.