Why 95% of AI Pilots Fail: SingularityNET's Path to AGI

How the Artificial Superintelligence Alliance Plans to Beat Big Tech to AGI

Janet Adams, COO of SingularityNET and founding member of the Artificial Superintelligence Alliance (ASI), delivered a striking presentation at the Global AI Show in the UAE. Her message: while the world obsesses over large language models, the real path to AGI—and enterprise ROI—runs through neurosymbolic AI. With a background in banking technology and AI regulation, Adams brings a practitioner's lens to what she calls "the winner-takes-all race" for artificial general intelligence.

On why AGI ownership matters: "If that technology is owned by a very few privileged big tech companies or government powers, it will be the most intelligent, the most powerful technology ever invented. It will have the ability to be a winner takes all race." SingularityNET's answer is decentralized, open-source development—a direct challenge to OpenAI, Google, and Microsoft's closed approaches.

On enterprise AI's dirty secret: "MIT recently did a very vast report on implementation of AI in industry. They concluded that 90% of companies are saying they have implemented some form of AI in their company. But 95% of the generative AI pilots show no measurable ROI." Adams has seen this firsthand: AI teams spun up with enthusiasm, then shut down when results don't materialize.

On why LLMs fail in high-stakes domains: "In anything which is high stakes—finance, education, healthcare, aviation, manufacturing, oil and gas—the industries in which you can't afford to make a mistake cannot effectively deploy LLMs for any serious processing." The hallucination problem isn't a bug to be fixed; it's fundamental to how neural networks work.

On neurosymbolic AI's advantage: "Neurosymbolic AI—they can reason and they can explain themselves in a way that regulators can approve, in a way that executives can understand that they are fulfilling their fiduciary duty towards their customers." This is the key differentiator: explainability that meets regulatory requirements, not black-box predictions.

On the pace of AI development: "AI is increasing in power at 10x a year right now. That means in three years it's going to be a thousand times smarter than today. We as humanity, our brains, we can't even get our heads around what this could mean." SingularityNET believes AGI arrives within 1-3 years—and they're building the decentralized infrastructure to ensure it benefits everyone.

On Web4 and AI agents: "Web 4 is the web of AI. It's where we will have AI intelligent agents interacting with each other to do much of the work that humans currently do. They're taking that basic programmable researchy stuff that we all do in our day jobs, and it's moving on to the web." The ASI Alliance's agentic layer (via Fetch.ai) is designed for exactly this multi-agent future.

5 Key Takeaways From Janet Adams on Enterprise AI and AGI

• ROI requires reimagination, not bolt-ons - Companies achieving results are going "AI first," reimagining their entire business in the world of AI, not asking "where can I put a chatbot"
• Neurosymbolic beats neural-only - Knowledge graphs hold richer semantic information than databases; neurosymbolic AI uses less compute and produces explainable outputs
• ASI Chain targets 100K TPS - Their AI-optimized blockchain competes with Visa's transaction speeds, enabling real-time messaging between AI agents at financial settlement scale
• Decentralization prevents winner-takes-all - The ASI Alliance (SingularityNET, Fetch.ai, Kudos) uses community governance to prevent any single party from controlling AGI
• OpenCog Hyperon enables multi-AI collaboration - Their framework lets different AI types (evolutionary, logical, neural) interact, learn, and evolve together toward AGI

What This Means for Organizations Deploying AI Agents

Adams' thesis cuts against the current AI hype: LLMs alone won't deliver enterprise value in domains where mistakes are costly. The path forward combines today's AI capabilities with neurosymbolic methods that can reason, explain, and pass regulatory scrutiny. Whether you buy SingularityNET's AGI timeline or not, her point about the 95% failure rate of generative AI pilots deserves attention. The organizations winning with AI aren't sprinkling chatbots on existing processes—they're fundamentally rethinking how humans and machines work together in what she calls the "Web4" era.