using AI for pension projections

af1963

The analogy is more that you go to a travel agent and ask to go to a holiday to a nice hot beach in Europe, and they suggest booking a flight that doesn't exist. Then when you discover this, they say "Yes, I'm sorry you're right, try this one instead." How far would you trust them ?

Dead_keen

The analogy is more that you go to a travel agent and ask to go to a holiday to a nice hot beach in Europe, and they suggest booking a flight that doesn't exist. Then when you discover this, they say "Yes, I'm sorry you're right, try this one instead." How far would you trust them ?

The modern world for AI (since 2025 or so) has all been about giving it tools that it can use. My travel agent AI checks the flights first using Google Flights before suggesting one. Give it access to live train times and it will tell you when and what platform your train will come in on. Give it access to your debit card and, well, that's probably a step too far for me!

Nebulous2

https://forums.moneysavingexpert.com/discussion/comment/82137538#Comment_82137538

That is what happened with my first successful attempt in France, asking for a route from the South of France to Calais. I gave it a few rules - no more than 450km in a day for instance. It had a fairly good attempt the first time, but placed me very close to Calais for the last two nights. I was able to go back and say, our crossing isn't until afternoon, so we have plenty time to get there on the last day, can you find me somewhere up to 100km away. So shaping the response to suit myself.

The second attempt - the find a hill attempt was so far off that I gave up. Insisting from a main road that I had to turn left, when I knew the hill was to the right. I showed it google maps, but it doubled down, insisting that though the hill was to the right I still had to turn left and find some hidden gate to take me to the hill.

I thought it was too far off to ever get anywhere and gave up.

snowlaser

https://forums.moneysavingexpert.com/discussion/comment/82138294#Comment_82138294

If a human did that I wouldn't trust them at all - but that's because it's an AI type of mistake to make. Humans make other types of mistakes - typos, forgetting things etc etc.

The bottom line is that AI is the most powerful tool ever invented, and is getting better at an extraordinary pace. People scoffing at it today will be like those who thought in 1997 the internet wasn't something they'd ever use or who said in 2008 "I'll never buy a smartphone - just don't need it".

Cobbler_tone

https://forums.moneysavingexpert.com/discussion/comment/82145162#Comment_82145162

A bit like when people scoff at the development of self driving cars. They might point to an instance when there was/could be a collision, forgetting the multiple bumps and accidents they drive by every day due to poor human controlled driving. Nearly 300 reported accidents a day, without the hundreds/thousands that aren't reported.

AI is changing the world rapidly and it is well documented, not necessarily all for the better but you better get used to it.

Nebulous2

https://forums.moneysavingexpert.com/discussion/comment/82145731#Comment_82145731

I think self-driving cars is a very good analogy.

They were vastly overhyped, and have been massively underdelivering. Maybe that was necessary to raise funding. The 'fake it until you make it.' silicon valley approach.

We've been promised that self-driving cars were imminent for around 10 years.

I have a friend who became extremely excited about self-driving vehicles around 6 years ago. "There are a million professional drivers in this country, and they are all going to lose their jobs within the next few years."

Another friend, who works in IT said "It's still a long-way off. The technology isn't good enough, the processing doesn't happen as it needs to, and the sensors / cameras don't work in poor weather. They may get it implemented in a limited way somewhere that it doesn't rain, like California, but it's still a long way off for the rest of us."

I was fairly neutral at the time, but the second one was much more on the money than the first.

If you drive a recent model car with all the inbuilt safety features - don't you get frustrated by how often it gets it wrong?

I have adaptive cruise control, which I really like, but I can pull out to overtake a lorry, it thinks it is in my lane and brakes as I get closer, just when I expect it to pick up speed for the overtake. I had a hired jeep which didn't cope at all with fog, kept thinking I was out of my lane and pulling me, either towards the verge, or towards the other side of the road. A friend was telling me this week of coming round a corner on a narrow road, meeting a lorry taking a lot of room and steered towards the verge to avoid it. There was a white line at the side of the road, the car - a recent Chinese model- thought he was leaving his lane and tried to pull him back into the path of the lorry.

I've no idea if AI is going to go the same way, but the history of self-driving cars is littered with broken promises.

It feels a bit like commercial nuclear fusion, which has been 10-15 years away for my whole adult life.

george_jetson

There is a (paywalled) article in the Times - headline AI chatbots give savers bad advice ‘most of the time’

The subheader tells you all you need to know…

When put to the test, artificial intelligence tools got 88 per cent of complex financial questions wrong — and failed 46 per cent of the easy ones

17 AI models were tested 121 questions covering pensions mortgages saving and loans.

Bobziz

https://forums.moneysavingexpert.com/discussion/comment/82146104#Comment_82146104

Interesting. It would be good to understand how much of the bad advice was due to the actual capability of the models tested v's the ability of the testers to write good prompts.

LHW99

https://forums.moneysavingexpert.com/discussion/comment/82146148#Comment_82146148

Trouble is, if the testers can't get the prompts right, how do they expect the average punter to?

Bobziz

https://forums.moneysavingexpert.com/discussion/comment/82146205#Comment_82146205

Indeed, like any powerful/dangerous/useful tool, proper training is required. It is available but costs time and money. I'd suggest that many people are unlikely to invest either, but it won't stop them claiming the tools are rubbish.