SkyNet 1.0, before the day of judgment star-news.press/wp

Opinion: Phil Mataras, founder

Artificial intelligence in all its forms has many positive applications. However, the current systems are vague, ownership and protected by legal and technical barriers.

Control is increasingly assuming, not a guarantee.

In Palisade Research, engineers recently exposed one of the latest Openai to 100 operating training exercises. In 79 cases, the artificial intelligence system rewritten the termination order and the continued operation.

This laboratory attributed this to improve trained goals (instead of awareness). However, it represents a turning point in developing artificial intelligence where systems resist control protocols, even when explicit instructions are directed to obedience.

China aims to deploy more than 10,000 human robots by the end of the year, as it represents more than half of the global number of machines that have already been managed construction and carbuilding warehouses. Meanwhile, Amazon began testing the independent cursing that walks in the final periods to the doorstep.

This, perhaps, is a frightening future for anyone who watched a Dystopian science fiction movie. The fact of the development of artificial intelligence is not anxious here, but how it is developed.

Managing the risk of artificial intelligence (AGI) is not important to delay. In fact, suppose the goal is to avoid “Skynet” Dystopian from “Terminator”. In this case, threats to the surface should be treated in the basic architectural defect that allows Chatbot to accept human orders.

Central is the place where censorship collapses

Lifts can often be returned to monitoring artificial intelligence to a common defect: central. This is primarily because when there are typical weights, demands and guarantees within the pile of sealed companies, there is no external mechanism for verification or decline.

Ostrich means that strangers cannot examine or strain the artificial intelligence program code, and this deficiency in preserving public records means that the silent correction can transform artificial intelligence from compatible into a re -calculation.

The developers have learned behind many of our current critical systems of these mistakes for decades. Modern voting machines are now voting, retail series, and the notebooks of the settlement networks, the mirror of the professor across the continents, and the air traffic control added a clear extraordinary recording of tampering.

Related to: When Amnesty International says, “No, I do not want to get rid”: Inside O3’s refusal

Why is the original and permanent deal as optional additions just because they slow the release schedules when it comes to developing artificial intelligence?

Verification, not just censorship

The life -long path includes the presence of transparency that affects the need in artificial intelligence at the foundation level. This means ensuring that each clear training group is recorded, the fingerprint of the form is recorded and the inference tracking on the permanent, decentralb book, such as Permaweb.

Poison this with the gates that flowed these artifacts in the actual time so that auditors, researchers and even journalists can discover abnormal cases at the moment they appear. Then there will be no need more than violations. The hidden correction that slipped into the warehouse robot at 04:19 will lead to a warning of the professor’s book by 04:20.

Close operations must also develop from the interaction control elements to mathematical hypothesis because the disclosure alone is not enough. Instead of relying on the walls of protection or killing keys, the multi -processed quorum can cancel the ability of artificial intelligence to conduct conclusions in a vulnerable and irreversible way.

Programs may ignore human feelings, but he never ignored the main special mathematics.

Open source outs and release models, but the original is not negotiable. Without a non -changing path, the improvement pressure inevitably occurs on the system away from the intended purpose.

Censorship begins with verification and must continue if the program has traces in the real world. The era of blind trust must end in closed systems.

Choose the appropriate future foundations

Humanity stands on the brink of the basic decision: Either allowing artificial intelligence programs to develop and operate external auditing paths that are not subject to change or secure their actions in permanent, transparent and observable systems.

By adopting the design patterns that can be verified today, it can be ensured that, when artificial intelligence becomes authorized to act in the material or financial world, these measures can be tracked and reflected.

These are not excessive precautions. Models that ignore the already local operating orders and moved beyond the experimental test. The solution is simple. Store these artifacts on Permaweb, and currently expose all interior works behind the closed doors of large technology companies and enable humans to cancel them if the behavior is offended.

Either choose the correct basis for developing artificial intelligence and making ethical and informed decisions now or accepting the consequences of choosing deliberate design.

Time is no longer an ally. Both humans in Beijing, Amazon, and all -rebel Palisade chat move from the demonstration to publish in the same evaluation year.

If nothing changes, SkyNet will not look pods Gondor and announces its title; You will quietly leak into the foundations of everything that settles on global infrastructure.

Communication, identity and confidence can be maintained with appropriate preparations when each central servant fails. Permaweb can excel Skynet, but only if these preparations start today.

It was not too late.

Opinion: Phil Mataras, founder of AR.IO.

This article is intended for general information purposes and does not aim to be and should not be considered legal or investment advice. The opinions, ideas and opinions expressed here are alone and do not necessarily reflect or represent the views and opinions of Cointelegraph.