Hacker News Reader

Comment by 0xpgm | Hacker News Reader

I starting to believe the biggest risk of AI is that it will become good enough to fool people into thinking it is sentient, and they'll start trusting it. With its inherent biases from the training data

View on HN

This happens all over the place already. I see apprentices blindly throwing prompts at ChatGPT á la "how do I accomplish [X] in [library I have no expertise in]?.

The biggest problem is: it doesn't even force you to form a coherent question. You can just throw word salad at it and feel "understood" in response.

I don't have the phantasy to imagine how this will impact the loops of, for example, confirmation bias - for basically anyone who isn't a deeply critical thinker/engineer. The dim and the loud are about to get even more confident _and_ even more shallow at the same time.

They don't need to be fully convinced of AI's sentience for this dynamic to become very problematic very quickly. They just want to feel "understood", and that's exactly what they will feel like.

Let me be the devil's advocate: don't people already listen to and trust other people who have biases?

I came here to leave this exact comment. A possible counter argument might be that people think computers are much more infallible than people (even if that's not objective reality).

Computers are, LLMs are not.

We already live in the world where you can learn anything from an "expert" on YoUTube. All AI does is remove the person at the other end. It's possible to learn some real dangerous stuff on there, too, like doing electrical work without properly disconnecting power first, or car maintenance that could fail and cause more damage. The problem isn't people assuming sentience and trusting AI, it's that people already are too trusting without verification.

With a positive spin: at least AI isn't actively malicious and doesn't try to get people to do dangerous things 'for the lulz'.

> All AI does is remove the person at the other end

So all it does is remove the only liable entity in the process, "all", scaling a pre existing bad thing doesn't make it good

Experts on youtube are often still fact checked eventually. AI gives answers privately that may never be fact checked by other

It hardly is an argument, "people already kill each-other with their fists let's give them guns"

The biggest risk is AI becoming economic agents - the elite will then have less and less need for human consumers, who will lose political power and protections, slowly diminish. Then... AI will slowly take over the elite, until there is no one left. Earth will spin endlessly through space, soulless.

Of course, that's not how AIs see it.

> Earth will spin endlessly through space, soulless.

Religion is on the decline, who still believes in souls?

85% of the global population identify with a religion

Yes, but there's generally a disconnect. If you merely identify with a religion as in "the cultural context and some rituals" it's different than truly, actually believing e.g. that there's life after death and Jesus awaits and you'll go to bootcamp for the ultimate battle between light and darkness that occurs at some point.

But at that point, you probably either believe that earth is endlessly spinning through space (with or without souls), or you really, truly believe in souls and whatever else some religion suggests (like parting seas with your will, or flying around with god at night because he wants to show you things).

Maybe that won't be so much of an ecological disaster. AIs don't need golf courses and marble kitchen counters.

If they're "made in our image" I don't see why they wouldn't. Even if they can't use a pool, maybe it's a status symbol in AI society to own physical space with such luxuries in it.

Who knows, maybe it'll help people develop actual critical thinking skills?

This will be needed. In fact, I think education will place a higher emphasis on critical thinking skills in the near future because of AI.

I wish I shared your optimism about education's ability to adapt and meet the needs of students.

At least Seneca is a step above marketeers and "influencers".

If people start taking advice from chatbots, Seneca will not be the most popular one. If a Seneca bot gets even close to popular, it will be surpassed by a Seneca bot tweaked to be a little bit more cool and relatable that also hypes energy drinks.

Seneca but with more sex appeal and maybe some catch phrases. Honestly that wouldn't be the worst thing. If it can make Stoicism attractive! lol

Perhaps this opens up another use case: inoculating people against being fooled by similar linguistic behavioral signalling into pretty much the same outcome eventualities by other people.

We are already steeped in and grown from the inherent biases of our cultures and societies.

Garbage in garbage out. But perhaps AI can be used to rescue and curate what goes in?

Nothing special when people are already trusting scams and others.

I don't think so, plurality of AI's should solve this.

Reply on news.ycombinator.com