Social

The social implications of AI are baaaaad. The broad categories I can think of: parasocial relations, cheapened human-looking output, and mental degeneration.

Parasocial relationships

I've only heard of this one by second degree connection, so it's not super prevalent, but it's prevalent enough. Some people get so socially attached to things that do not reciprocate. This draws these people away from reality and feeds their delusions. Reality is the ultimate source of all things, and it does not do us well to stray far from it.

I have no solution to this. Some people are just hopeless. We can't control it like drugs, because there are very powerful open models out there that will run locally on someone's computer. You can't control that. At least drugs are a physical thing.

Cheapened human-looking output

This is probably the broadest category. For the first time ever I think, it has become easier to produce than it is to consume "human-looking" output. This is why it feels AI slop is so prevalent. It takes 2 seconds to use AI to generate an essay that looks legit, but it takes 100x longer to read it and analyze it. That never use to be the case, it used to be it takes far longer to write something that passes as valid text. This has all kinds of downstream implications.

The most surface level issue with this is just that the internet is flooded with absolute trash AI images and AI articles that you can't really tell are slop until you waste your time and attention to read a couple of sentences in. This content can be useless or actively detrimental.

Another problem is the explosion of scamming. It used to be that scamming took some effort and had returns too low for most criminals people to consider. Now they can set up an AI scam farm with potentially very high returns.

Disinformation becomes much easier to spread because of collapse of the proof of effort model.

This represents a paradigm shift. We may have to shift to a verification/trust broker model, instead of the current implicit effort model. This a big problem because lower trust increases the cost to cooperation, and cooperation is the ultimate force multiplier for progress.

Mental degeneration

In some ways AI is like cars. Cars get us very far for no physical effort. AI gets us a lot of information with no mental effort. Cars made us lazy, lazy made us fat, and fat made us unhealthy. AI might do the same thing on a mental level, it'll make us lazy and stupid. We'll have to do deliberate mental exercise just like we do physical exercise to keep our minds sharp.

Is this different from what every generation has said about new technology? I've seen people quote ancient testaments (unconfirmed source), that say "writing is making us lazy because we no longer have to remember things with our minds". I think AI is different. I think some things free our minds up to do other tasks, where as others replace our minds altogether. For example, people were right about TV being brain rot. I think AI is more like TV than it is like writing.

Terence Tao has written about this, with the recent AI resolution to the Navier–Stoke millennium problem.

Philosophical

I've seen some discussion about the nature of AI self-awareness, and how human AI is.

I'm a naturalist, so I think it's possible for consciousness to arise as an emergent property out of purely physical material. So if I extend that framework, it is technically possible for AI to become self-aware. But again i don't think that's a big issue, as long as we're not programming the AI to suffer.

Consciousness is literally the hard problem of philosophy. I don't think we'll truly ever know if AI is self-aware or conscious. Just in case though, we shouldn't program AI with chronic pain. That shouldn't be hard. As long as we do that we should be in the clear morally.

I also don't think there's anything special to being human. I didn't feel any existential dread when machines beat chess or go. There's nothing special about an AI doing something or a human doing something, as long as the outcome is the same (I do have to distinguish that sometimes the process itself is a part of the outcome, maybe even the most important part of the outcome; an emphasis on outcome does not equate being narrowly results-oriented. This distinction is often lost in discussions on ends and means).

I am human, so I by the principle of propinquity, I put human interests first; but I don't go so far as to be a human chauvinist.

Confirming and dispelling some notions

Datacenters

One prevailing notion I see on reddit and lemmy that I would argue against is on datacenters. Most anti-data center arguments are just factually wrong. You don't like AI, that's a perfectly legitimate reason to be anti-datacenter; just say that, don't make things up. I'm not even some kind of datacenter stan. It's just that every time someone says "I'm anti datacenter because [some absurd BS]" it rubs me the wrong way.

Datacenters use power - everything uses power. If the datacenter builders think paying for the power is worth the money then they can spend that money. If anything this is a huge boon for greenfield build out of renewables due to the demand and money that's coming into the electrical generation sector. High demand is never the enemy if supply can be increased. If we had smart policy wonks (we don't) we would be co-opting AI money to build solar farms.

Datacenters use water - the amount of water data centers use is miniscule compared to its output density; depending on the cooling method, they could be using less water than a residential zone of the same area. I am computer hardware adjacent, I know for a fact how much water a server rack uses. The water consumption figures critics use are just absurd lies.

Noise pollution - datacenters output less noise than basically any other industrial complexes like powerplants, factories, transit hubs. All the criticisms here depend on absurd misuse of resources. But it's very simple to just not do that.

Any locality where datacenters are built, they become a cash cows like any other business -- as long as it follows basic, existing zoning rules regarding consumption and ejecta control.

Copyright

Another point I want to discuss is AI copyright violation. I disagree that training AI is inherently violating copyright, even if the outcome is similar to copyright violation.

All copyright comes with fair and transformative use clause. When AI gets trained on copyrightable text or an image, you will never find it again in the weights. It's identical to when a person reads and watches, then synthetizes that info and outputs a distinct piece. It is still possible for AI to violate copyright, just like it's possible for any person to violate copyright, but training and inferencing are not inherently violations of copyright. The issue is separate from copyright.

I mentioned in the social section, the issue is ease of production. If a giga-artist can output 1000000 paintings a day, regardless of how good those paintings are, they cheapen themself and everyone else. This is achieving the same effect as copyright violation, which is to cheapen competing work. When I'm buying art I'm not just buying some paint, I'm buying the story behind it.

We should retain the right to sue AI users, hosts, and trainers for actual copyright violation. But we can't be trapped in the framework of copyright. Like I mentioned in the social section, this may have to move in a whole different direction like some kind of trust/verification scheme, but I'm not sure.

Autonomous weapons

There are levels to this. We already have missiles with computer assisted seekers that commit fratricide. I don't have a huge problem with AI seekers. People often forget that science fiction is fiction. Terminator isn't real. I'm not concerned about rogue hostile AI for various technological reasons. What I think is important in the end, is that true blackbox AI should be trained with Asimov's 3 laws as hard guardrails. This should be done in a verifiable way, so we can practically implement it in arms control.

The 2 problems with any weapon is 1) concentration of power, and 2) abuse by rogue actors. An example of 1 is the colonial era, when Europeans gains domination over all others and enslaved and genocides them. An example of 2) is random crazies doing mass shootings. We've dealt with both. Not well, but we're dealing with them and I don't think AI will bring anything truly unprecedented.

Overall though I should say again, as a society we should agree to slow down both AI adoption and research until we can catch up our thinking on some of these social and economic issues. But once the cat is out of the box we should try to use and direct it in the best direction we can. The last thing we should do is to emulate nuclear fearmongers who derailed the nuclear energy train, and possibly delayed action against climate change for decades.