Balaji Srinivasan calls AI the most economically useful invention of our time

Balaji Srinivasan, a prominent thinker in technology and crypto, calls AI the most economically useful invention of our time. He argues that AI is not a single all-powerful system but rather a "polytheistic" ecosystem of many strong AIs backed by different players.

“We empirically observe polytheistic AI… rather than a single all-powerful model.”

This perspective dispels the myth of one dominant artificial general intelligence (AGI) and instead highlights a balance of power between many human/AI combinations.

AI Works "Middle-to-Middle"

Currently, AI does not perform entire jobs autonomously. Instead, it requires humans at both ends: one to prompt the AI and another to verify its outputs. This shifts the main costs and efforts to the edges — prompting and verifying — even though AI accelerates the core process.

AI Amplifies Intelligence, Not Replaces It

Balaji prefers the term "amplified intelligence" over artificial intelligence because AI is not fully agentic. It does not set long-term goals or verify its own output. The usefulness of AI depends heavily on the user's intelligence and ability to prompt effectively. Poor instructions lead to poor results. AI helps users perform more tasks but generally only to an average level of quality. For example, it can enable someone to fake being a UI designer or game animator but not to produce expert-level work. Specialists remain essential for high-quality output. Interestingly, AI tends to replace previous versions of itself rather than human jobs directly. For instance, GPT-4 replaced GPT-3, and Midjourney displaced Stable Diffusion in workflows. Balaji also notes AI is better at visuals than text because humans can easily verify images, whereas verifying text or code is slower and more costly.

Crypto as a Boundary for AI

Balaji draws a clear distinction between AI and crypto:

• AI is probabilistic, guessing based on patterns.
• Crypto is deterministic, based on provable mathematics.
This makes crypto a boundary AI cannot easily cross. For example, AI might break captchas but cannot fake blockchain balances. Cryptographic equations remain a barrier to AI's full capabilities.

The Real AI Threat: Autonomous Weapons

According to Balaji, the truly dangerous AI is not chatbots or image generators but autonomous weapons like drones, which are actively pursued by many countries and have lethal real-world impact.

AI Decentralization

Contrary to fears of AI centralization, Balaji argues AI is decentralizing power. Many AI companies and open-source models exist, enabling small teams to build strong AI systems without massive budgets. This breaks up power rather than concentrating it.

The Ideal Amount of AI

Balaji rejects the notion that more AI is always better. He compares AI adoption to the Laffer Curve in economics, where there is a sweet spot between extremes. 0% AI is slow; 100% AI leads to poor quality. The best results lie somewhere in between.

Four Limits of Today’s AI Systems

Balaji outlines four constraints that keep AI from becoming godlike machines:
• Economic: Every API call costs money; scaling AI is expensive.
• Mathematical: AI cannot solve chaotic or cryptographic problems.
• Practical: Humans are still needed to prompt and verify AI output.
• Physical: AI cannot independently gather or interpret real-world data.
These limits might be overcome in the future by combining fast, intuitive AI (System 1 thinking) with logical, careful traditional computing (System 2 thinking), but this remains theoretical.

There is no all-knowing AI today, only expensive, limited, competitive tools that require constant human oversight.

---

Source: Originally published at Cryptopolitan on August 3, 2025.

---

Frequently Asked Questions (FAQ)

Understanding AI's Role and Capabilities

Q: What does Balaji Srinivasan mean by "polytheistic AI"? A: Srinivasan uses "polytheistic AI" to describe an ecosystem of numerous powerful AI systems developed by different entities, rather than a single, all-powerful artificial general intelligence (AGI). This highlights a distributed power dynamic among various AI models and their human collaborators. Q: How does AI currently augment human capabilities, according to the article? A: The article suggests AI works "middle-to-middle," requiring human input for prompting and verification of outputs. It amplifies human intelligence by accelerating core processes, enabling users to perform more tasks, though not necessarily at an expert level without specialized human oversight. Q: What is the difference between "artificial intelligence" and "amplified intelligence" as discussed? A: Balaji Srinivasan prefers "amplified intelligence" because current AI systems are not fully autonomous or agentic. They lack the ability to set long-term goals or independently verify their work, thus their effectiveness is amplified by human intelligence and skill in prompting. Q: Why is AI considered more effective with visual content than text? A: AI is noted to be better with visuals because humans can verify images more quickly and easily. Verifying text or code, on the other hand, is a slower and more costly process.

AI's Limitations and Future

Q: What are the four main limitations of current AI systems mentioned? A: The four limits are:
• Economic: The high cost of API calls and scaling AI.
• Mathematical: AI's inability to solve chaotic or cryptographic problems.
• Practical: The continued necessity of human prompting and verification.
• Physical: AI's lack of independent capability to gather or interpret real-world data.
Q: What is considered the "real AI threat" according to Balaji Srinivasan? A: Srinivasan identifies autonomous weapons, such as drones, as the true danger, citing their pursuit by multiple nations and their lethal real-world impact, rather than consumer-facing AI like chatbots or image generators. Q: Does Balaji Srinivasan believe more AI is always better? A: No, he rejects this notion. He likens AI adoption to the Laffer Curve in economics, suggesting there's an optimal "sweet spot" for AI usage, where too little or too much can be detrimental.

AI and Crypto Interaction

Q: How does crypto act as a boundary for AI? A: Crypto's deterministic nature, based on provable mathematics, creates a boundary that AI, which is probabilistic and relies on pattern guessing, cannot easily cross. For instance, AI may bypass captchas but cannot falsify blockchain balances.

Crypto Market AI's Take

Balaji Srinivasan's perspective on AI as an amplifier of human intelligence and a decentralized force resonates strongly with our platform's ethos. At Crypto Market AI, we leverage cutting-edge AI to create sophisticated trading bots and market analysis tools that empower users, rather than replace them. Our focus on responsible AI integration aligns with the idea of finding the "ideal amount of AI" – augmenting human capabilities for better decision-making in the dynamic crypto market. We believe AI's true value lies in its ability to process vast amounts of data and identify patterns that might elude human traders, thereby enhancing decision-making and risk management.

More to Read:

• AI Agents: The Future of Business Automation and Customer Engagement
• Understanding AI-Driven Crypto Trading Tools
• The Future of Cryptocurrency: What's Changing and Why It Matters