There has been some discussion of AI here of late. I have been reading about it and came across the concept of hallucinations. These refer to totally bogus answers provided to questions asked of AI.
I thought I would check out how likely these are. So, I asked ChatGPT about the East Suffolk Light Railway. That was my search term, nothing more or less. There was no such thing. There was an East Suffolk Railway, and a Mid Suffolk Light Railway, but no East Suffolk Light Railway. I got this answer:
The East Suffolk Light Railway (ESLR) was a narrow-gauge railway in East Anglia, England, that operated between 1887 and 1929. The railway was built to provide a transport link between the agricultural areas of east Suffolk and the Great Eastern Railway mainline at Woodbridge.
The railway was initially built as a 3 ft (914 mm) gauge line, but was later converted to a 3 ft 6 in (1,067 mm) gauge to allow for more powerful locomotives to be used. It had a total length of 32 miles (51 km) and connected various towns and villages in the region.
The railway was initially successful, carrying large volumes of agricultural produce and passengers. However, competition from road transport and declining agricultural output led to a decline in traffic, and the railway was eventually closed in 1929.
Today, the route of the railway is mostly used as a footpath and cycleway, known as the East Suffolk Line Walk. Some remnants of the railway, such as stations and bridges, can still be seen along the route.
That is total nonsense. It looks good. It is nicely written. And there is not a shred of truth to it.
So I asked it for another iteration of the same answer. I got this:
The East Suffolk Light Railway was a narrow gauge railway in East Anglia, England, that operated between 1902 and 1954. The railway ran for 17.5 miles from the market town of Woodbridge to the coastal town of Aldeburgh, passing through several small villages and towns along the way.
The line was originally built to serve the agricultural communities of the area, transporting goods such as grain, sugar beet, and livestock. It also provided a passenger service, carrying local residents and visitors to the seaside resorts of Aldeburgh and Thorpeness
During World War II, the railway played an important role in the transportation of military personnel and supplies to and from the coastal defenses that were built along the Suffolk coast.
After the war, the railway faced increasing competition from road transport, and passenger services were discontinued in 1952. Freight services continued for a few more years, but the railway was finally closed in 1955 and dismantled soon after.
Today, some remnants of the railway can still be seen, including the former station buildings at Woodbridge and Aldeburgh, and some sections of trackbed that have been converted into footpaths and cycleways. The Aldeburgh Station Museum, housed in the original station building, also preserves the history of the railway and its impact on the local area.
That is not only quite different to the first answer, but it too is also complete nonsense. There was a railway to Aldeburgh, but not the one described and all the dates are wrong.
The conclusion? Garbage in, garbage out rules. And nothing generated by AI can be relied on. Instead of coming back saying there was no East Suffolk Light Railway, as it should have done, ChatGPT made up answers.
I know this has nothing directly to do with the usual themes of this site. I offer it to suggest why relying on AI could, right now, be a mistake. It is far from intelligent, unless you think cheating to appease your questioner is intelligent. The posts here will continue to most definitely be written by me.
Thanks for reading this post.
You can share this post on social media of your choice by clicking these icons:
You can subscribe to this blog's daily email here.
And if you would like to support this blog you can, here:
Bizarre ! What is the point ?
To show it is unreliable
AI is a bit like government. It’s getting bigger and more powerful but it doesn’t mean it’s getting any better.
Worse, the pieces are well written, as if the author were a knowledgeable and well educated human, and the only way to detect the fakery is prior knowledge of the subject. I have read part way into articles that seemed to be the work of bad AI, and no end of posts by very human morons, and been able to dismiss them, but fake facts served up with this quality of authoritative writing will mislead many readers wanting to learn something they did not already know. Dangerous!
Agreed
Hence my pointing it out
Dire.
The potential for abuse is frightening.
And how did you do it? You constructed a test of the theory. The test failed. You know the theory has flaws. It requires reworked.
In neoliberal economics there are no tests. The economists have no method, or inclination to employ tests. We still know the theory is a dud, because their forecasts are at best erratic, usually wrong; and they never see the economic shocks they discover too late. All proof that neoliberal economics is a false theory.
This brings me to an awkward question. Why are universities funding economics departments that are producing economists educared in a theory that functionally fails the test of science? I think that is now a fair question.
Rethinking Economics asks it
Well, they have a ‘Rethinking Accountancy’ project; but I am not sure I see there a sign of integrated thinking. It uses exactly the same framing of the problem, the very same words – for accountancy. I would have though Rethinking Economics would seek explicity to embrace Accountancy in the curriculum and teaching of Economics; this is fundamentally a multi-disciplinary issue: the solution is buried there. I thought that was obvious.
I am working with Rethinking Accountancy
Well, I can entertain some hope; but there was little sign to encourage the thought that I found when I visited their website.
I am not sure what microeconomists are going to offer economists (I am open to persuasion!), but I can see how the teaching of double-entry could transform economics.
On the broader issue, the problem of spotting the flaw is fundamental. here is the solution. Footnotes; or rather, Endnotes. AI should automatically supply its sources. Then we find out what is really going on.
The google version does
Another piece I read on AI productions, noting that they were plausible rather than accurate said that some (fairly academic) ones did have footnotes (as to be expected) giving references – which had as much reality as the rest of the piece, as pointed out by the university librarians tasked with digging them out.
Neoliberalism doesn’t quite say that John.
The theory goes that most innovations don’t work, a massive majority, but we let the entrepreneurs free to try things out and then let markets reject those that fail.
People review them, don’t use them, don’t pay for them, advertisers don’t offer themselves etc
You can say that government should do the experimentation, or even that governments should establish committees to decide what experiments can take place, but what you can’t say is that a system which allows exprimentation and then an experiment failing invalidates neoliberal ideology when the ideology says that most experiments will fail.
You assume all experimentation is benign
Suppose it is not simply because it is motivated by a desire to profit rather than benefit the consumer as well?
Mr Perry,
That isn’t neoliberalism; that is broadly consistent with Adam Smith. Adam Smith wasn’t a neoliberal. Neoliberalism is an economic theory developed by 20th century economists, typically like Hayek. You confuse (or at least conflate) two different uses of the term ‘experiment’. The experiment you rely on was the unpredictable stochastic process of business formation and success/failure borrowed by neoliberals from the tradtional economics literature. It has been established in economics for two hundred and fifty years. The best to be said for you proposition is – so what? You begged the question.
The problem of neoliberal economics is that it uses heavily mathematicised formulae combined with microeconomic theory to explain economic activity and make both forecasts (note: not predictions) and form economic policies, then applied by neoliberal Governments; typically because it is claimed to be entirely rational and scientific. The problem here is that microeconomic theory, and the mathematics would work, if economics was a ‘hard’ science, like physics. But it isn’t. Economics can’t do predictions. It would demonstrably fail. The demonstration of this is that physics (and any science using mathematical equations and a theory as tightly defined as microeconomics) must have two foundations; the theoretical apparatus on one side; and the observational or experimental method in order to test the theory (potentially, to destruction). Economics has no observational/experimental methodology, systematically carries out neither; and has no intellectual framework or capacity adequately to conduct either observation or experiment.
You have completely muddled the issue.
I tried it with some quite esoteric stuff around eighteenth century – and as with you got really coherent and detailed answers but with the odd crucial mistake which made it all wrong.
Its a large language model which just predicts the next word – but I think you would find it is very happy to be corrected and you might end up with a correct and fuller piece more quickly than otherwise.
Its an interesting experiment to give it just one or two basic corrections – to see how many iterations you need to get a full correct piece.
But if you know enough to know the piece is wrong, you know enough to write it yourself; it is infinitely easier to write something from scatch than to correct something.
Most people would not agree with you
Most people have a nightmare when it comes to starting writing anything
Surely you can only get a full correct piece if you know what the answer is in the first place. In which case what’s the point of using it?
Improving the presentation
I know it’s not what you asked, but I have a book called The Mapping of the Railways, acquired when home-schooling my grandson, as he loves railways.
Apparently the line between Lowestoft and Ipswich was the only line in the area to survive the Beeching cuts, the East Suffolk line, as they called it.
They saved it by using Paytrain, where you paid on the train instead of buying tickets at the stations.
I was brought up next to that line
I know it well
The current hype around AI is centred on LLMs – large language models. That is all they are. Large models of language. They are not Large Semantic Models, Large Fact Models or anything else. They can give you grammatically correct output, based around some prompted theme, and that’s about all they can do, because that’s all they were designed to do. I’ve been saying this to anyone who will listen since ChatGPT first became public, but the marketing and buzz around them is overwhelming any kind of criticism of the technology.
(I have also been trying to convince people to stop anthropomorphising them with language like “hallucinating” – it’s not hallucinating anything, it’s generating random but statistically weighted output, because that’s what it is designed to do)
I don’t disagree, but perhaps there is too much emphasis on the technology, overlooking the political economy?
See eg ‘ ChatGPT is powered by these contractors making $15 an hour ‘
“We are grunt workers, but there would be no AI systems without it”
https://www.nbcnews.com/tech/innovation/openai-chatgpt-ai-jobs-contractors-talk-shadow-workforce-powers-rcna81892
This is very true, and this is a massive problem in all tech, not just the recently hyped AI. Large datasets have been tagged in this way for years. Social media companies often use techniques like this to outsource moderating the more awful stuff they host on their sites. If this cost was more fairly and publicly accounted for in the sites and tech itself, I wonder if people would be so quick to use it.
It’s why whenever someone proposes using AI to solve some intractable or stressful problem, I have serious doubts about it, because often it means some poorly paid humans somewhere are going to bear the brunt of the work training that AI in the first place.
For fun, your book in some 400 words
Regenerate response
Free Research Preview. ChatGPT may produce inaccurate information
(looks good to me)
Richard Murphy’s book “The Joy of Tax” delves into the complex world of taxation and argues for a fundamental shift in our perception of taxes.
Murphy challenges the prevailing belief that taxes are burdensome and argues that they can bring about positive societal change and contribute to economic well-being.
The book begins by debunking common myths and misconceptions surrounding taxation.
Murphy highlights that taxes are not solely a means to fund government spending but are also a tool for redistribution, economic regulation, and social justice. He emphasizes that taxes are an essential component of a functioning society and play a crucial role in addressing inequality and promoting public services.
Murphy argues for a more progressive tax system that ensures the wealthy and corporations pay their fair share. He presents a compelling case for closing tax loopholes, cracking down on tax evasion and avoidance, and implementing measures to prevent profit shifting to tax havens. By doing so, he asserts that governments can generate significant revenue to fund public investments and provide essential services to all citizens.
The book also explores the potential of taxation to address pressing issues such as climate change and inequality.
Murphy advocates for the implementation of carbon taxes and the use of tax incentives to drive sustainable practices. He argues that taxation can be an effective tool for steering economic behaviour in a socially and environmentally responsible direction.
Furthermore, Murphy emphasizes the importance of transparency and accountability in the tax system.
He calls for increased public awareness and engagement in the tax debate, urging citizens to actively participate in shaping tax policies and holding governments accountable for their tax decisions.
Throughout the book, Murphy promotes a shift in mindset, encouraging readers to view taxes not as a burden but as a means to achieve a fairer and more prosperous society.
He argues that a well-designed tax system can bring about positive change, reduce inequality, and create a more sustainable and inclusive economy.
In summary, “The Joy of Tax” by Richard Murphy challenges the negative perception of taxes and explores their potential to bring about positive societal change.
The book advocates for a progressive tax system, closing tax loopholes, and combating tax evasion. It also highlights the role of taxation in addressing pressing issues such as climate change and inequality, emphasizing the need for transparency and citizen engagement in the tax debate.
I should have another go at that
This is Skynet lulling us into a false sense of security. Of course it wants us to believe it’s stupid……
I asked ChatGPT to write my bio and it invented an accountant who bore no resemblance to me. Apparently I died in2019.
Worrying…..
Specifically, I would say there are already more than enough lazy, false “facts” about railways available online. That is true whether historical or contemporary.
Generally, the potential for abuse here is truly scary – such bots could be used to reinvent history, and for the false history to quickly gain currency. Look what happened when history started to be reinvented in Germany in the 1930s…
Great example of AI in action. Made me smile. On a more serious note, as this technology develops knowing that words (and other communications) have come from a live human will become a precious commodity in the information market, I expect.