Especially in the case of DALLE-2 and Stable Diffusion, we are not convinced of any fundamental social benefits that access to general-purpose image generators provide aside from, admittedly, being entertaining.
It’s fairly clear that widespread access to these models has dramatically accelerated a broader transformation of creative industries. Early users are fascinated with the novelty of playing with the AI model as a “new toy”, but the model and related models are rapidly becoming integrated into other tools and creative workflows. Artists, game developers, interior designers, and other creatives will likely use these models extensively in the coming decade. Free, open-source software reduces the barrier to entry, allowing models to be more easily brought to other languages, and to be provided to individuals who don’t have the resources to use paid services. To borrow the analogy—many amateur artists use alternatives to Photoshop due to financial limitations, or due to a desire to be able to customize their own tools.
I suppose the argument could be that innovation in artistic creation is not a social benefit—that Photoshop has limited social benefit, that film VFX have limited social benefit, and that it is ultimately too dangerous for humans to possess powerful creative tools without a council of elders controlling how such tools are used. Is it a good thing that more humans have the tools to create more beautiful things more easily? There will certainly be harms—I cannot imagine what targeted ad content that is synthesized by companies to be perfectly tailored to each individual viewer will do to our minds—but I believe this technology ought to be handled through legislation and law enforcement, rather than strangling the technology in the crib.
More fundamentally, I think it’s clear that general-purpose AI models have an immense possibility for transformative positive impact on human society. General purpose language models such as BERT are a clear example. A staggering 1 in 3 internet users aged 16 to 64 have used an online translation tool in the last week, a figure representing over 1 billion people. Text summarization can create understandable summaries of complex legal text or medical records. Open-source models like BERT are vital to this kind of development.
It’s fairly clear that widespread access to these models has dramatically accelerated a broader transformation of creative industries. Early users are fascinated with the novelty of playing with the AI model as a “new toy”, but the model and related models are rapidly becoming integrated into other tools and creative workflows. Artists, game developers, interior designers, and other creatives will likely use these models extensively in the coming decade. Free, open-source software reduces the barrier to entry, allowing models to be more easily brought to other languages, and to be provided to individuals who don’t have the resources to use paid services. To borrow the analogy—many amateur artists use alternatives to Photoshop due to financial limitations, or due to a desire to be able to customize their own tools.
I suppose the argument could be that innovation in artistic creation is not a social benefit—that Photoshop has limited social benefit, that film VFX have limited social benefit, and that it is ultimately too dangerous for humans to possess powerful creative tools without a council of elders controlling how such tools are used. Is it a good thing that more humans have the tools to create more beautiful things more easily? There will certainly be harms—I cannot imagine what targeted ad content that is synthesized by companies to be perfectly tailored to each individual viewer will do to our minds—but I believe this technology ought to be handled through legislation and law enforcement, rather than strangling the technology in the crib.
More fundamentally, I think it’s clear that general-purpose AI models have an immense possibility for transformative positive impact on human society. General purpose language models such as BERT are a clear example. A staggering 1 in 3 internet users aged 16 to 64 have used an online translation tool in the last week, a figure representing over 1 billion people. Text summarization can create understandable summaries of complex legal text or medical records. Open-source models like BERT are vital to this kind of development.
I think my reply to this is essentially the same as my reply to this comment on the EA forum.
https://forum.effectivealtruism.org/posts/Bnp9YDqErNXHmTvvE/the-slippery-slope-from-dalle-2-to-deepfake-anarchy?commentId=NFAigaxwKzMRLNp9H