Can AI Be Made Safe?

 My answer is, obviously not. Virtually everything from screwdrivers to aircraft transport is "dual use." How will AI be different? 

Our efforts to keep 12-year-olds "safe" from seeing naked boobies will not prevent anyone from developing autonomous weapons. I will get back to the boobies later.

The following answer from Perplexity ignores the use of AI by the military or countries (most of them) beyond the reach of well-meaning chat groups.

PERPLEXITY ANSWERS:*

Yes, We Can Train AI to Be Safe

1. Technical Research is Growing: The field of AI safety technical research is expanding, with initiatives like the Harvard AI Safety Team aiming to build communities around AI safety research. This indicates a growing interest and investment in understanding and mitigating AI risks
   1
   .
2. Implementation Frameworks Exist: Google's Secure AI Framework (SAIF) demonstrates that standardized approaches to integrating security and privacy measures into AI development are being developed and adopted, addressing key implementation challenges
   3
   .
3. Policy Initiatives are Underway: The U.S. has shown a commitment to leading global AI policy, with measures like appointing chief AI officers in agencies and establishing AI governance structures, which are crucial steps towards managing AI safety at a policy level
   3
   .
4. Comprehensive Training Programs: There are a variety of training programs and resources available for individuals interested in AI safety, from self-study courses to machine learning-focused training, indicating a robust infrastructure for education in AI safety
   4
   .
5. Community and Advocacy Efforts: The establishment of the U.S. AI Safety Institute Consortium and international cooperation on AI safety demonstrate a strong community and advocacy effort towards ensuring AI safety, which is essential for both raising awareness and developing solutions
   3
   .

No, Training AI to Be Safe is Challenging

1. Neglected Career Path: AI safety technical research remains a highly neglected career path, with a relatively small number of individuals actively engaged in full-time research, indicating challenges in attracting talent to the field
   1
   .
2. Complex Technical Challenges: AI safety faces numerous technical challenges, such as data quality issues and the difficulty of debugging AI systems, which can hinder the effective implementation of safety measures
   2
   .
3. Regulatory Hurdles: Despite policy initiatives, the regulatory landscape for AI is complex and evolving, with potential for backlash and disagreements over the scope and implementation of regulations, which could slow progress in policy development
   3
   .
4. Limited Accessibility to Training: While there are many training resources available, facilitated courses are often oversubscribed, and there may be barriers to accessing or fully benefiting from these resources for all interested individuals
   4
   .
5. Unpredictable Risks: The nature of AI development means that new, unforeseen risks can emerge as technology advances, making it difficult to fully anticipate and mitigate all potential safety issues, despite ongoing efforts
   6
   .

In conclusion, while there are significant efforts and initiatives aimed at training AI to be safe, encompassing technical research, implementation strategies, policy development, and education, there are also substantial challenges that make this an ongoing and complex endeavor.

"Debugging Godzilla"

SAFE GENERATIVE AI

Of course, the porn industry rushes to exploit Generative AI. The quality can be creepy, as we see above, but the standards of this audience are not high. 

Both of these pictures are from sexy.ai, one of many outfits rushing in to deal with the terrible shortage of pictures of naked ladies on the internet.

Some go as far as tailoring a girlfriend who can chat with you to replace the girls you find so hard to talk to. Some offer 3D environments where you can act out your fantasies with a tailor-made virtual partner.

So what do we mean by "safe"? It's simply an effort by the big outfits to save themselves from scandal.

The following is a brilliant picture that slipped through the granny nanny of DALL-E, which will block any mention of "bikini." However, in this case AI got a bit confused ...

A realistic image of a beautiful Filipina wearing green bikini charging towards dynamically on the camera in a densely detailed forest with a long exposure zoom blur effect, emphasizing the speed and power of as it moves. The forest is intricately detailed with various types of trees, rich undergrowth, and scattered light rays penetrating through the foliage, rendered in high-resolution 4K.super Close up

AI will block this if we ask for "woman" rather than "Filipina." It would seem obvious to guess why, but it's also blocked for a "man in a business suit." Go figger.

If you ask me, creating "unsafe" images in AI may draw kids away from all the other violent and sexually explicit content that is at their finger tips. At least it's creative. 

===========================

* Note that Perplexity's "answers" include specific references in the numbers within what it "says." Follow to see if what Perplexity is saying makes sense.