What Kind of AGI Will We Get?

The dawn of Artificial General Intelligence (AGI), the ability for AI to match human level’s ability to understand, learn and apply knowledge, is upon us:

📏 You can 𝐦𝐞𝐚𝐬𝐮𝐫𝐞 it in the ability of models to ace benchmark tests. GPT-4, Gemini and Claude have shown significant year-over-year gains on academic and professional benchmarks like MMLU and HumanEval.
🤖❤️ You can 𝐟𝐞𝐞𝐥 it in the multimodal, real-time experiences offered by projects like Google’s Astra, which demonstrate integrated perception, memory, and reasoning in a single conversational agent.
👀 You can 𝐬𝐞𝐞 it in robotics, where Amazon’s latest manipulation breakthroughs suggest more generalizable object-handling, long a key challenge in the field.
🧩 You can 𝐫𝐞𝐚𝐬𝐨𝐧 it in the emergence of agentic models that combine logical planning with perceptual intuition, echoing Daniel Kahneman’s 𝘚𝘺𝘴𝘵𝘦𝘮 1 (fast, intuitive) and 𝘚𝘺𝘴𝘵𝘦𝘮 2 (slow, deliberative) thinking working better in sync.

While there is still some way to go, whether it’s 2 or 10 years, AGI is beginning to enter our lives. The question is no longer 𝘪𝘧 it will arrive, but 𝘸𝘩𝘢𝘵 𝘬𝘪𝘯𝘥 𝘰𝘧 𝘈𝘎𝘐 we will get.

Consider the trajectories now taking shape:
– 𝐌𝐞𝐭𝐚 envisions AI agents that can sell anything to anyone 𝘢𝘯𝘥 serve as your therapist or companion, all optimized to your data for maximum engagement and monetization.
– 𝐆𝐨𝐨𝐠𝐥𝐞 is making AI the new interface to the web. AI Mode displaces traditional link-based results with synthesized answers. While convenient, this shift threatens the lifeblood of many websites by cutting off traffic, and merges ads with facts, raising concerns about manipulation. Studies show how a simple change in result order shapes behavior, let alone corrupting information.
– 𝐎𝐩𝐞𝐧𝐀𝐈 has drawn criticism for launching powerful models without full safety or system cards at launch.
– 𝐱𝐀𝐈 emphasizes “free speech” at the cost of alignment, releasing models that use offensive language or express politically charged views.
– In contrast, 𝐃𝐞𝐞𝐩𝐒𝐞𝐞𝐤 appears tightly censored, especially on topics like the Chinese government, often resulting in evasive or nonsensical responses.

So the question becomes:
𝐖𝐢𝐥𝐥 𝐭𝐡𝐞 𝐀𝐆𝐈 𝐰𝐞 𝐠𝐞𝐭 𝐞𝐦𝐩𝐨𝐰𝐞𝐫 𝐮𝐬, 𝐨𝐫 𝐦𝐚𝐧𝐢𝐩𝐮𝐥𝐚𝐭𝐞 𝐮𝐬 𝐢𝐧𝐭𝐨 𝐚𝐧 𝐮𝐥𝐭𝐫𝐚-𝐜𝐨𝐧𝐬𝐮𝐦𝐞𝐫𝐢𝐬𝐭, 𝐢𝐝𝐞𝐨𝐥𝐨𝐠𝐢𝐜𝐚𝐥𝐥𝐲 𝐛𝐫𝐚𝐢𝐧𝐰𝐚𝐬𝐡𝐞𝐝, 𝐜𝐨𝐮𝐜𝐡 𝐩𝐨𝐭𝐚𝐭𝐨-𝐛𝐨𝐮𝐧𝐝 𝐝𝐲𝐬𝐭𝐨𝐩𝐢𝐚?

Life with AGI will reflect the incentives, values, and accountability of those building it, and that consumers and businesses demand, not just the capabilities of the systems themselves.

(References available on demand)
#AI #AGI #AIEthics #AIEconomics #Leadership