news: Microsoft & AI 😱
About images they will make clear when they are generated, they are not stupid.
About text, it is more worrying since many people are already using #ChatGPT as it could be reasonable and factually accurate, while it is just meant to produce text that sounds plausible and is syntactically and grammatically correct.
news: Microsoft & AI 😱
@post For example, this search result claims Microsoft is abandoning Windows.
Imagine people using Windows doing a search, getting a GPT3 result in Bing that *looks* legitimate like this, but is quite wrong... about basically any topic, not just tech.
https://www.perplexity.ai/?uuid=00bc5419-c9e4-415d-acc4-768be1699507
news: Microsoft & AI 😱
@post I do think computer generated imagery and text *can* be useful, but there are so many problems with it
1. It can appear authentic when it isn't (mentioned already)
2. Training corpus can be problematic; results will reflect the (racist, sexist, ableist, etc.) material fed to the machine. https://hub.jhu.edu/2022/06/21/flawed-artificial-intelligence-robot-racist-sexist/
3. It'll sometimes replace skilled people with low quality generated content (attached screenshot with alt text). https://twitter.com/mer__edith/status/1612445135580631041
news: Microsoft & AI 😱
@post Then there's the whole old cloud bait and switch problem, where OpenAI, Adobe, and others hold a proprietary model and charge fees to use it. 😔
And people who might be using it (for example: image generation, for inspiration or inpainting) are locked-in.
(Stable Diffusion is at least no cost and self-hostable, whereas Dall-E, Midjourney, and Photoshop are not.)
news: Microsoft & AI 😱
On degradation of reliability and quality of content, we are already seeing it for example with the wave of articles automatically translated into various languages and they are not honest enough to admit what is artificially generated.
As a feedback, we could see the rise of platforms where your identity, profile pic and skills are verified to prevent automated replies. Basically this could be an incentive to give away privacy in the name of authentic interactions.
As regards unemployment, it depends more on the workers' incapacity to renegotiate working conditions, so any external shock (technological innovation, speculation, inflation, etc.) becomes an excuse for a sneaky shift of wealth and the costs are passed on to the weaker social class. That is, it is a structural problem.
If you ask me, the most incisive structural change is the Job Guarantee Programs as elaborated by the economists of the Modern Monetary Theory, which a globalized but uneven world that wants to continue to innovate technologically can no longer do without.
news: Microsoft & AI 😱
I'm generally opposed to the term "AI", they could be labeled as "artificially generated" or something like that, shown in their own category or even being generated after the user specifically ask for it.
Sadly it's not like there isn't already the risk to retrieve an artificial image from a website when using a search engine and take it for real.
news: Microsoft & AI 😱
@post Sure, but some people will look at the images & not understand that they're made up by a computer. How can you message that to someone who doesn't know what "AI" actually means? And if you hide the meaning behind a (?) icon that has a tooltip, will they actually find and read it?
I agree the text regeneration is more problematic for sure. There's a GPT3 search engine integration demo @ https://www.perplexity.ai/ — it has notes, but that makes it even more problematic.