What you need to know about general artificial intelligence

Recently, there has been increasing discussion about generative artificial intelligence tools, especially after the release of several large-scale language models and image generators (such as DALL-E or Midjourney).

These inventions have once again put general artificial intelligence (GPAI) under the spotlight and once again raised hypothetical questions such as whether GPAI should be regulated.

Before one explores the possibilities further, first understand the concept of GPAI, what it means, when was it introduced, etc.

What is general artificial intelligence?

Two years ago, in April 2021, the European Commission launched general artificial intelligence. The original AI bill proposal exempted creators of general artificial intelligence from complying with a number of legal instruments and other liability standards.

The reason is that it only applies to high-risk artificial intelligence, which is clearly mentioned and explained in the bill based on their purpose and context.

Another regulation, Article 28, supports this assertion and recommends that AGI developers only make significant adjustments or developments to AI systems for high-risk uses. Responsible for regulations.

But according to recent reports, the European Parliament is also considering certain "obligations" for original general artificial intelligence providers.

The basic purpose of the EU Artificial Intelligence Act is to classify and categorize the various chains of actors involved in developing and deploying systems using artificial intelligence.

Here are 5 considerations to guide the regulation of general artificial intelligence

The Artificial Intelligence Act’s approach to general artificial intelligence is ideal for establishing a regulatory tone for addressing the harms of global artificial intelligence. With the recent increase in public interest in generative AI, there is also the risk that regulatory stances may be overly adapted to current issues.

Surprisingly, newer innovations like ChatGPT, dall-e2, and Bard aren’t even really a problem; in fact, they’re just the tip of the iceberg.

Recommended: Responsible Artificial Intelligence - How to Adopt an Ethical Model in Regulatory Environments

General Artificial Intelligence is a huge category

The first thing to understand is that General Artificial Intelligence Intelligence is a huge category, so it is logical to apply it to a wide range of technology areas, rather than limiting it to chatbots and LL.M.s.

To ensure that the EU Artificial Intelligence Bill is futuristic in nature, it must address a much larger scale. First, a proper description of GPAI should include many techniques ("tasks") that can serve as the basis for other artificial intelligence systems.

The European Council defines this as:

“The provider intends to perform generally applicable functions such as image and speech recognition, audio and video generation, pattern detection, question answering, translation, etc.; General artificial intelligence systems can be used in multiple scenarios and can be integrated into multiple other AI systems."

General artificial intelligence may cause a wide range of harm

Although these risks cannot be applied in layers are completely overcome, but we can deny the fact that they can have an impact on various applications and actors. We should consider the current state of AI technology, its applications, and how it works while developing a general approach to AI regulation.

For example, general AI models run the risk of generating anti-democratic discourse, such as hate speech targeting sexual, racial and religious minorities. The risk with these models is that they entrench constrained or distorted views into the data underlying them.

General Artificial Intelligence should be governed throughout the product lifecycle

In order for General Artificial Intelligence to take into account the diversity of stakeholders, it must be governed throughout the entire product lifecycle, not just It's at the application layer. The first stages of development are critical, and the businesses creating these models must take responsibility for the information they leverage and the architectural decisions they make. The existing architecture of the AI ​​supply network effectively enables participants to profit from remote downstream applications while avoiding any consequential liability due to a lack of oversight at the development layer. This includes the process of collecting, cleaning, and annotating data, as well as the process of creating, testing, and evaluating models.

A standard legal disclaimer is not enough

It is impossible for the creators of general artificial intelligence to use basic legal disclaimers to exclude their liability. This approach can lead to a dangerous vulnerability that releases all responsibility from the original developer and places the responsibility on downstream actors who do not have the ability to manage all risks. The council does have an exception to the general approach, which would allow AGI developers to absolve themselves of any liability as long as they exclude all high-risk uses in their instructions and ensure the system cannot be abused.

Recommended: Uncovering, addressing and debunking hidden AI bias

Engaging non-industry actors, society and researchers in wider consultation

A fundamental, unifying Documentation of practices to evaluate general AI models, and generative AI models in particular, across a variety of hazards is an ongoing area of ​​research. To avoid superficial tick-box exercises, regulation should prevent narrow assessment approaches.

General artificial intelligence systems must undergo meticulous vigilance, verification, and inspection before they are implemented or made available to the public. Recent proposals to bring general artificial intelligence models within the scope of the AI ​​Bill either defer the development of future specific standards (to be decided by the Commission) or attempt to do so in the wording of the AI ​​Bill.

For example, in a consensus society, the distribution of possible impacts may differ depending on whether a prototype is built and used by the entire community or by a small community.

The EU Artificial Intelligence Bill is about to become the first broad artificial intelligence law, and one day, it will become a unified standard for all countries. This is why it is crucial to take the field of artificial intelligence and translate it into a global template that everyone can follow.

The above is the detailed content of What you need to know about general artificial intelligence. For more information, please follow other related articles on the PHP Chinese website!