OpenAI, Makers Of ChatGPT, Commit To Developing Safe AI Systems

OpenAI has published a new blog post committing to developing artificial intelligence (AI) that’s safe and broadly beneficial.

ChatGPT, powered by OpenAI’s latest model, GPT-4, can improve productivity, enhance creativity, and provide tailored learning experiences.

However, OpenAI acknowledges that AI tools have inherent risks that must be addressed through safety measures and responsible deployment.

Here’s what the company is doing to mitigate those risks.

Ensuring Safety In AI Systems

OpenAI conducts thorough testing, seeks external guidance from experts, and refines its AI models with human feedback before releasing new systems.

The release of GPT-4, for example, was preceded by over six months of testing to ensure its safety and alignment with user needs.

OpenAI believes robust AI systems should be subjected to rigorous safety evaluations and supports the need for regulation.

Learning From Real-World Use

Real-world use is a critical component in developing safe AI systems. By cautiously releasing new models to a gradually expanding user base, OpenAI can make improvements that address unforeseen issues.

By offering AI models through its API and website, OpenAI can monitor for misuse, take appropriate action, and develop nuanced policies to balance risk.

Protecting Children & Respecting Privacy

OpenAI prioritizes protecting children by requiring age verification and prohibiting using its technology to generate harmful content.

Privacy is another essential aspect of OpenAI’s work. The organization uses data to make its models more helpful while protecting users.

Additionally, OpenAI removes personal information from training datasets and fine-tunes models to reject requests for personal information.

OpenAI will respond to requests to have personal information deletion from its systems.

Improving Factual Accuracy

Factual accuracy is a significant focus for OpenAI. GPT-4 is 40% more likely to produce accurate content than its predecessor, GPT-3.5.

The organization strives to educate users about the limitations of AI tools and the possibility of inaccuracies.

Continued Research & Engagement

OpenAI believes in dedicating time and resources to researching effective mitigations and alignment techniques.

However, that’s not something it can do alone. Addressing safety issues requires extensive debate, experimentation, and engagement among stakeholders.

OpenAI remains committed to fostering collaboration and open dialogue to create a safe AI ecosystem.

Criticism Over Existential Risks

Despite OpenAI’s commitment to ensuring its AI systems’ safety and broad benefits, its blog post has sparked criticism on social media.

Twitter users have expressed disappointment, stating that OpenAI fails to address existential risks associated with AI development.

One Twitter user voiced their disappointment, accusing OpenAI of betraying its founding mission and focusing on reckless commercialization.

The user suggests that OpenAI’s approach to safety is superficial and more concerned with appeasing critics than addressing genuine existential risks.

This is bitterly disappointing, vacuous, PR window-dressing.

You don’t even mention the existential risks from AI that are the central concern of many citizens, technologists, AI researchers, & AI industry leaders, including your own CEO @sama.@OpenAI is betraying its…

Advertisement

— Geoffrey Miller (@primalpoly) April 5, 2023

Another user expressed dissatisfaction with the announcement, arguing it glosses over real problems and remains vague. The user also highlights that the report ignores critical ethical issues and risks tied to AI self-awareness, implying that OpenAI’s approach to security issues is inadequate.

As a fan of GPT-4, I’m disappointed with your article.

It glosses over real problems, stays vague, and ignores crucial ethical issues and risks tied to AI self-awareness.

I appreciate the innovation, but this isn’t the right approach to tackle security issues.

— FrankyLabs (@FrankyLabs) April 5, 2023

The criticism underscores the broader concerns and ongoing debate about existential risks posed by AI development.

While OpenAI’s announcement outlines its commitment to safety, privacy, and accuracy, it’s essential to recognize the need for further discussion to address more significant concerns.

Featured Image: TY Lim/Shutterstock

Source: OpenAI