Many Americans got their first glimpse behind the scenes of machine learning when details about Amazon’s “Just Walk Out” technology became public. Instead of pure technology that tallies customers’ purchases and charges them when they leave the store, sales were manually monitored by about 1,000 real people working in India.
But these workers were the human half of what most AI really is: a collaboration between reinforcement learning and human intelligence.
The human element is often ignored in discussions about the safety of AI, which is a bit troubling given the major impact AI is likely to have on our labor markets and ultimately our individual lives. This is where decentralization, the inherent reliability and security of blockchain technology can play a major role.
The Center for Safe AI identifies four broad categories of AI risks. For starters, there’s malicious use, where users can “intentionally deploy powerful AIs to cause widespread harm” through “new pandemics or [using them] for propaganda, censorship and surveillance, or [releasing AIs] to autonomously pursue harmful goals.”
A more subtle concern is the risk of an AI race, where companies or nation states compete to quickly build more powerful systems while taking unacceptable risks. Uncontrolled cyberwarfare is one possible outcome; another possibility is that systems can evolve on their own and possibly escape human control; or a more prosaic, but no less disruptive, outcome could be mass unemployment due to unchecked competition.
The organizational risks in AI are similar to those in any other sector. AI could cause serious industrial accidents, or powerful programs could be stolen or copied by malicious actors. Finally, there is the risk that the AIs themselves will go rogue, “optimizing imperfect objectives, straying from their original objectives, becoming power-hungry, resisting shutdowns, or engaging in cheating.”
Regulation and good governance can pose many of these risks. Malicious use is addressed by restricting searches and access to various features, and the legal system can be used to hold developers accountable. The risks of malicious AI and organizational issues can be mitigated through common sense and promoting a security-conscious approach to AI use.
But these approaches do not address some of the second-order effects of AI. Namely centralization and the perverse incentives left over from older Web2 companies. For too long we have traded our private data for access to tools. You can unsubscribe, but for most users this is difficult.
AI is no different than any other algorithm, in that what you get out of it is the direct result of what you put into it – and enormous amounts of resources have already been spent cleaning and preparing data for use for AI. A good example is OpenAI’s ChatGPT, which is trained on hundreds of billions of lines of text from books, blogs, and communities like Reddit and Wikipedia, but also relies on humans and smaller, more customized databases to refine its output.
Read more in our opinion section: What can blockchain do for AI? Not what you heard.
This brings with it a number of problems. Mark Cuban recently pointed out that AI will ultimately need to be trained on data that companies and individuals may not want to share in order to become more commercially useful than just coding and copywriting. And as more jobs are impacted by AI – especially as AI agents make customized AI applications accessible – the job market as we know it could ultimately implode.
Creating a blockchain layer in a decentralized AI network could alleviate these problems.
We can build AI that can track data provenance, maintain privacy, and allow individuals and companies to charge for access to their specialized data if we leverage decentralized identities, validation staking, consensus, and roll-up technologies like optimistic and zero-knowledge proofs. This could shift the balance from large, opaque, centralized institutions and offer individuals and corporations a whole new economic system.
On a technology level, you need a way to confirm data integrity, data ownership, and legitimacy (model auditing).
Then you would need a provenance method (to borrow a phrase from the art world), meaning you can see the audit trail of every piece of data to properly compensate for whoever’s data is being used.
Privacy is also important: a user should be able to secure their data on their own electronics and control access to their data, including the ability to revoke that access. This requires cryptography and a security protection certification system.
This is an advancement over the existing system, which merely collects valuable information and sells it to centralized AI companies. Instead, it enables broad participation in AI development.
Individuals can take on a variety of roles, such as creating AI agents, providing specialized data, or offering intermediary services such as data labeling. Others can contribute by managing infrastructure, operating nodes or providing validation services. This inclusive approach enables a more diversified and collaborative AI ecosystem.
We could create a system that benefits everyone in the system, from the digital clerics a continent away to the shoppers whose shopping cart contents provide them with raw data to developers behind the scenes. Crypto can ensure safer, fairer, and more human-centric collaboration between AI and the rest of us.
Sean is the CEO and co-founder of Sahara, a platform building a blockchain-powered infrastructure that is reliable, permissionless, and privacy-protective to enable the development of custom autonomous AI tools by individuals and companies. In addition, Sean is an associate professor of computer science and the Andrew and Erna Viterbi Early Career Chair at the University of Southern California, where he is the Principal Investigator (PI) of the Intelligence and Knowledge Discovery (INK) Research Lab. At the Allen Institute for AI, Sean contributes to machine common sense research. Previously, Sean was a data science consultant at Snapchat. He completed his PhD work in computer science at the University of Illinois Urbana-Champaign and was a postdoctoral fellow at the Stanford University Department of Computer Science. Sean has received several awards recognizing his research and innovation in AI, including Samsung AI Researcher of the Year, MIT TR Innovators Under 35, Forbes Asia 30 Under 3, and more.