The Robot Doctor Will See You Now

Not in that sense.

AI needs input that matches the training data. If it hasn’t trained on something, it can’t know how to deal with it (it will guess (badly) in some cases). This is why AI often fails at very niche edge cases - it wasn’t covered in its training.

This is because AI can’t create information. It can only reorganize information that it has been given to train on. If a self driving car AI hasn’t trained on humans wearing plastic bags, it will not be able to infer that it’s both a plastic bag and a human. It will pick one or the other. Silly example, I know.

 

https://twitter.com/user/status/1886610132567777514



The largest medical #AI randomized controlled trial yet performed, enrolling >100,000 women undergoing mammography screening, was published today @LancetDigitalH
The use of A.I. led to 29% higher detection of cancer, no increase of false positives, and reduced workload compared with radiologists without A.I.. thelancet.com/journals/landi…

https://www.thelancet.com/journals/landig/article/PIIS2589-7500(24)00267-X/fulltext

 

Sweden and Norway!
TL;DR: Standard procedure is to have two people check the images. They used AI to classify low and high risk. Low risk was checked once manually, high risk was checked twiced.

This resulted in overall better performance and 44 % fewer manual checks in total in the AI+Human group compared to 2xHumans.

 

If today's AIs are non-adaptive, then I don't see why people are worried about AIs taking over the world. A non-adaptive AI is no more threatening than a 4-banger calculator. AIs that can change their own goals could be a threat.

 

Yes. The science fiction— AI becomes concious and kills everyone is pretty far from reality.

But AI is a tool, a possibly dangerous tool — and if people with bad intentions get their hand on it — it can do a lot of damage (through spreading misinformation, or linking online actions to people and violating privacy, etc.)

 

We’re already at a point where we have to assume that any media online can be fake.

I’m just waiting to see if the first giant disaster will be because of someone using AI for evil, or because of someone using AI stupidly. It’s 50/50.

The problem is that AI for evil probably has happened already. We just don’t know about them yet.

Also, I’m not concerned about sentient AI or General AI. GAI is decades out, and might never be possible.

 

The ultra-rich&powerful could use it to make themselves even richer and more powerful. What will criminal organizations use it for?

Those aren't valid reasons for not using this tool for good, such as improving health care. The misuse will occur even if we banned AIs, since that wouldn't prevent it from being used in secret by people who don't honor the ban. Banning AIs would make that even worse, because we wouldn't be learning how to detect misuse.

 

What’s the difference between the criminals and the ultra rich focusing only on increasing their own wealth while millions starve?

 

None, but AI could be a more effective tool for them to do this. The problem isn't new: just think of the Pharoahs and the rest of the population. The rich&powerful have just made use of newer tools as they get developed (with good intentions). They haven't properly exploited AIs yet, AFAIK, but that is something to watch for.

Darn, I can't think of any SF stories about misuse of AIs.