Opinion by: Phil Mataras, founder of AR.IO
Artificial intelligence in all forms has many positive potential applications. However, the current systems are blurred, owned and shields from the audience of legal and technical barriers.
Control is increasingly an assumption rather than a guarantee.
In Palisade researchEngineers recently undergo one of the latest Openai models in 100 shutdown drills. In 79 cases, the AI system re -wrote the conclusion of the command and ongoing operation.
The lab is linked to the trained purpose -optimization (rather than awareness). However, it marked a development in the development of AI where systems were resistant to protocols, even though it was clearly ordered to follow them.
China It aims to deploy more than 10,000 humanoid robots by the end of the year, costing more than half of the global number of machines governing warehouses and building cars. Meanwhile, Amazon began testing autonomous couriers walking the last meter to the door.
This is, perhaps, a scary sound that is future for anyone watching a dystopian science-fiction movie. This is not the reality of AI’s development that remembers it, but how it was developed.
Managing the risks of artificial general intelligence (AGI) is not a task that can be delayed. In fact, suppose the goal is to avoid the “Skynet” dystopian of “Terminator” films. In that case, the threats that surfed the main defect of architecture that gives a chatbot to deprive human commands need to be met.
Centralization is where oversight is destroyed
Ai failures Administration is often traced back to a common defect: centralization. This is mainly because, when model weights, signals and care exist within a sealed corporate stack, there is no external mechanism for verification or rollback.
Opacity means the outsiders Can’t assess or fork the code of an AI programAnd the lack of public custody of the record suggests that a single, quiet patch can change an AI from the following recalcitrant.
The developers behind some of our current critical systems learned from these mistakes were decades ago. Modern voting machines are the image of the hash-chain ballot, the settlement of glass networks to ledgers on the continents, and the control of air traffic has added redundant, tamper-bright logs.
Related: When an AI says, ‘No, I don’t want to be powerful’: Within the decline of O3
Why is validation and permanent being treated as optional extras just because they slow down the discharge schedules when it comes to AI development?
Resurrection, not just oversight
A viable path forward involves embedding the required transparency and proven to AI at a foundation level. This means ensuring that each set of training shows, fingerprint model and trace understanding are recorded in a permanent, decentralized ledger, such as Permaweb.
Pair with the gateways that stream real-time artifacts so that auditors, researchers and even journalists can see anomalies once they appear. Then there is no need for whistleblowers; The stealth patch that slides into the warehouse at 04:19 am will trigger a ledger alert by 04:20.
Shutdowns should also change from reaction controls to the processes implemented in mathematics because unity is not enough. Instead of relying on firewalls or killing switches, a multiparty quorum can be scryptographically withdrawing an AI’s ability to make references to a publicly audible and irreversible way.
The software may ignore human emotions, but it has not ignored private mathematics key.
Open-sourcing models and publishing signed hashes help, but the proven is the non-negotiating piece. Without the irreversible route, the pressure pressure will inevitably distance the system from its intended goal.
Oversight begins verification and should continue if the software has real-world implications. The Blind Trust period in closed-door systems should end.
Choosing the right foundations in the future
Humanity stands in the rain of a major decision: either allowing AI programs to develop and work without external, irreversible paths to auditing or securing their actions in permanent, transparent and public that the systems see.
By adopting the proven design patterns today, it will be sure, where Ai Being authorized to act in the physical or financial world, these actions can be monitored and restored.
These are not very careful. Models that ignore shutdown commands are in motion and move across the beta test. The solution is simple. Keep these artifacts in Permaweb, expose all the internal activities currently behind the closed doors of large tech companies and empower people to recover them if they make a mistake.
Either choose the right foundation for AI development and make ethical and informed decisions today or accept the consequences of an accidental design selection.
Time is no longer allies. The Humanoids of Beijing, the Amazon couriers and Palisade’s rebellious chatbots all move from the demo to the expansion of the same calendar year.
If nothing changes, Skynet does not sound gondor horns and express itself with a headline; It dripped silently on the very foundations of all that stabilized the global infrastructure.
Communication, identity and trust can be maintained by proper preparation when each central server fails. Permaweb can release Skynet, but if the preparations start now.
It’s not too late.
Opinion by: Phil Mataras, founder of AR.IO.
This article is for general information purposes and is not intended to be and should not be done as legal or investment advice. The views, attitudes, and opinions expressed here are unique and do not necessarily reflect or represent the views and opinions of the cointelegraph.
