Tech Companies Come Together to Pledge AI Safety in Seoul AI Summit

• Tech companies such as Google, OpenAI & Microsoft have come together and signed an agreement, promising to develop AI safely.
• In case the technology they have been working on seems too dangerous, they are willing and able to pull the plug on projects.
• 16 companies have already voluntarily committed to the agreement, with more companies expected to joinsoon..

The Seoul AI Summit started off on a high note. Leading technical giants such as Google, Microsoft, and OpenAI signed a landmark agreement on Tuesday aiming to develop AI technology safely. They even promised to pull the plug on projects that cannot be developed without risk.

It's a world first to have so many leading AI companies from so many different parts of the globe all agreeing to the same commitments on AI safety." - Rishi Sunak, UK Prime Minister

The UK PM also added that now that this agreement has been set in place, it'll ensure that the biggest AI companies in the world i.e. the biggest contributors to AI developments will now maintain more transparency and accountability.

It's important to note that this agreement only applies to frontier models,'which refers to technology that powers generative AI systems like ChatGPT.

More About the AI Summit Seoul Agreement

The most recent agreement is a follow-up to the pact made by the above-mentioned companies last November at the UK AI Safety Summit in Bletchley Park, England, where they had promised to mitigate the risks that tag along with AI as much as possible.

16 companies have already made a voluntary commitment to this pact, including Amazon and Mistral AI. More companies from countries like China, the UK, France, South Korea, UAE, the US, and Canada are expected to follow suit.

Companies that haven't already committed to these pacts will be creating their safety framework and detailing how they plan to prevent their AI models from being misused by miscreants.

These frameworks will also have something called red lines' which refer to risks that are intolerable.

In case a model has a red line" issue (such as automated cyberattacks or a potential bioweapon threat), the respective company will activate the kill switch, which means the development of that particular model will cease completely.

The companies have also agreed to take feedback on these frameworks from trusted actors, such as their home governments, before realizing the full plan at the next AI summit that has been scheduled in France in early 2025.

Is OpenAI Really a Safety-First AI Company?

OpenAI, one of the biggest driving forces of AI in the world, is an important signatory to the above-mentioned agreement. However, the recent turn of events at the company indicates that it's now taking a step back when it comes to AI safety.

First Instance: Using Unlicensed AI Voice

Just a couple of days ago, OpenAI came under heavy criticism after users found its Sky' AI voice similar to Scarlett Johansson. This comes after the actress formally declined to license her voice to OpenAI.

Second Instance: Disbanding the AI Safety Team

Even worse, OpenAI has now dissolved its AI Safety Team, which was formed in July 2023 with the aim of aligning AI with human interests. This team was in charge of ensuring that AI systems developed by the company do not surpass or challenge human intelligence.

Third Instance: Top Officials Resigning

Top OpenAI officials, including co-founder Ilya Sutsveker and the co-leader of the superalignment team of GPT-4o Jan Leike resigned last Friday, just hours apart from each other.

In fact, Leike described in detail the circumstances around his resignation. Interestingly, he was in disagreement with the core principles of the current OpenAI board. He also underlined the dangers of developing AI systems more powerful than the human brain, and that OpenAI is unbothered about these security risks.

All these incidents point to one thing: OpenAI is developing systems that do not sit well with many safety engineers and advisors. These systems may be more powerful than the human brain can comprehend and therefore possess catastrophic abilities that must be curtailed.Growing Regulations Around AI

Ever since AI gained popularity, governments and institutions around the world have been concerned about the risks associated with it, which is why we've seen a number of regulations being imposed around the development and use of AI systems.

• The USA recently introduced an AI Rights Bill that aims to develop AI by maintaining fairness, transparency, and privacy, and prioritizing human alternatives.
• The EU has introduced a new set of rules for AI that come into force next month. These rules will be applicable to both high-risk and general-purpose AI systems, with the only difference being that rules will be a little more lenient for the latter.
• Every AI firm will have to maintain more transparency, and if they fail to meet the guidelines, they'll have to pay a fine that can range between 7.5 million euros or 1.5% of their annual turnover to 35 million euros or 7% of global turnover, depending on the severity of the breach.
• As per an agreement between the two countries, the US and UK AI Safety Institutes will partner with each other on safety evaluations, research, and guidance for AI safety.
• The United Nations General Assembly in March 2024 adopted a resolution on AI encouraging countries around the world to protect their citizens' rights in the face of growing AI concerns. The agreement was initially proposed by the U.S. and supported by over 120 nations.

To conclude, while it's certainly positive news that nations around the world are recognizing the risks and responsibilities that come with AI, it's even more crucial to actually implement these policies and see to it that regulations are strictly followed.

The post Tech Companies Come Together to Pledge AI Safety in Seoul AI Summit appeared first on The Tech Report.