Initially hailed as a panacea for productivity, Generative AI is now falling out of favor for reasons many have been flagging as inevitable all along. According to analysts at Gartner, some 30% of GenAI projects will be abandoned before exiting proof of concept by the end of 2025. Issues cited include low data quality, an inability to control risks, and spiraling costs of up to $20 million for a single deployment.
Advocates of decentralized AI solutions have long pointed out the potential for Web3 technologies to overcome or mitigate some of these issues. On-chain transactions can open the black box of data and algorithms, supporting transparency and, thus, quality. This transparency also facilitates better risk management of common AI problems such as algorithm bias and AI hallucinations. As for costs - most costs associated with AI stem from the vast storage and computing requirements. In both areas, decentralized networks promise to unlock the power of otherwise idle resources, such as storage or computing, at substantially reduced cost.
However, while some decentralized data and resource networks such as Bittensor and Render have seen reasonable traction, their success has so far been dwarfed by the centralized tech giants, all of which are keen to make their mark in AI. As such, among enterprises, takeup of decentralized networks has been relatively lukewarm.
I recently had the opportunity to grill Michael Casey, veteran industry commentator and Chairman of the Decentralized AI Society, a community of organizations and KOLs, which was established to tackle the probability of an AI tech monopoly. With a helicopter view of the industry, Casey is all-in on decentralized AI but points to the work ahead:
"I think within the next three years, decentralized AI will prove to be the first mass-scale application of Web3 technology. For this to take off, we must first bring definition, structure, regulatory clarity, and a sufficient degree of standardization to the currently fragmented, ill-defined, nascent decentralized AI economy. When we bring a little more order to the chaos, people and institutions outside of the Web3 community can take advantage of opportunities posed by this technology."
From Zero to StandardizationIn the context of the current blockchain landscape, where does one start with bringing "order to chaos"? The very concept of standardization can seem almost alien in an industry where each project is striving for its own USP. However, sharing a set of common definitions and protocols is the only route to true interoperability. A functioning decentralized AI economy that can replicate its centralized counterparts will require the seamless flow of data, communications, and payments across platforms. Due to complexity, multiple layers will be needed to handle storage, processing, access management, and more. A prerequisite for user experience and, thus adoption is that all this complexity must be handled under the hood.
While the big tech firms battle against regulation, decentralized AI networks will have an even higher bar to reach. Being prepared to comply with any AI regulation on the one hand, but on the other, digital asset regulations such as the EU MiCA rules will apply to the issuance and sale of network utility tokens. Compliance with any existing rules covering data protection and user privacy is also a prerequisite for enterprise takeup.
Issues of trust, consensus, and governance must also be addressed. Democratic, decentralized governance are utopian ideals that contrast sharply with the frankly dystopian future offered by a monopoly of big tech controlling all our personal data along with the development of AI. However, decentralized governance at scale comes with significant challenges. Governance attacks such as the one suffered by Compound this summer are one example of the kind of risk that needs to be addressed.
Finally, it's worth noting that many decentralized AI projects have been long in development while nimbler AI startups have sped ahead in development terms. Ultimately, it will be down to the Web3 sphere to show, not tell, that the results of these efforts will have been worth the wait. And according to Casey, one factor will be the final decider:
"The tipping point for an open economy for AI agents and computation will come when large-scale enterprises see the money to be made in it. At some point, they'll recognize both the cost savings to be had in tapping into decentralized excess compute capacity and the market opportunities that trusted, customizable autonomous AI agents will help them seize."
So, while bringing order to chaos enables Web3 to compete by mitigating the worst risks, delivering the hard value that's currently missing from centralized solutions will be the winning ticket for decentralized AI.