The most interesting thing in tech: How will we regulate AI agents? It's not going to be easy, and we'll have to do lots of things. One important step will be data labeling: making sure that packets of information created by agents are labeled as such. Maybe we even need to add special labels for bots making lots of requests or passing across lots of routers, just as cars and trucks are required to have license plates but bicycles do not.
You keep saying "we", as if you and your organization will have any voice in how the tech lords design our future. You won't - you already sold your souls to them and they own you. So "we" is just a remnant of the agency you had but gave up.
So if we don't regulate it, will it go away faster? Please?
Sounds like a job for an investigable digital mark of authenticity, rights and provenance. Oh wait, here's one. https://mrkd.art/metamark
Those mini-taskers are going to be hard to regulate as they are already embedded or are being embedded in workflows, connectors, apis, and organisations. A bit like the early days of the Internet when one would saunter down to 'Cyberia' cafe to access cyberspace it was regulated. then we got past dial-up and into wifi when the web became ubiquitous. same for mobile, same for metaverse (insert joke here) and absolutely the same for AI - where is it? how can I see it, touch it, use it? labelling sure, regs sure. this is a hard, hard task to manage - humans are needed here, they really are.
Imho what Nicholas Thompson is essentially talking about is that #AI (#mindfulAI hopefully in future) has just emphasized the importance of two aspects of data: non-repudiable attribution to author, and management of metadata. Privacy management already called for it, but big i-net advertising businesses so far have been winning the lobby wars. The tech is in place for years already, and some more is being frantically built. Imho, it is time for regulators, esp., Consumers' Rights Protection offices, to pull their act together, and to establish safe boundaries of "public information space" (definition due) within which only information with known author and required other metadata can live.
Why is everyone still accepting these “pizza at 6” examples at face value? We’ve been having claims and fake demos about this for years and they all forget one main thing: there is no business case for this stuff for consumers. People do not care about having an agent doing menial tasks for them, and AI won’t get good enough any time soon to try and convince them by generating the still non-existent need.
If people in a echo chamber reinforce their own narrow view of the world, what will AI bots consuming each other's content be like?
Interesting, as always, but good luck with that! Does the licence plate help the police when the getaway car is registered stolen? Not really I imagine and so it is with AI - regulation is going to enormously complicate life for regular folks and AI creators, whilst just providing another couple of steps nefarious people will get used to circumventing. Not really to do with AI agents, but relevant to the regulation debate: https://www.bbc.com/news/articles/c72ver6172do I don't know what the answer is, but I fear nefarious uses are almost more prevalent than virtuous ones. 🤷♂️
CEO @ The Atlantic | Co-Founder, Keynote Speaker
3wHere's the article I mentioned: https://www.theatlantic.com/technology/archive/2024/07/ai-agents-safety-risks/678864/