ChatGPT Figures Out Huddersfield: Three AI-Guided Urban Explorations

Aled Singleton, Andrea Freeman, Tim O'Connor, David Upton

In 2023 the authors undertook several psychogeographical activities during a weekend in Huddersfield, UK. We based these around ChatGPT, the now well-known Artificial Intelligence (AI) Large Language Model (LLM) that was just becoming available to the general public. Firstly, we asked ChatGPT to write song lyrics based on Huddersfield locations and had these 'sung' in public to karaoke backings. Secondly, we used ChatGPT as a navigational tool to help us to play a game of Monopoly based on Huddersfield streets. Thirdly, we asked ChatGPT to help us by finding locations and writing itineraries in Huddersfield. This article discusses these three experiments as they were experienced by the four authors, and offers perspectives for other writers, artists, walkers, and scholars who may seek to also use LLM forms of AI to interpret urban space. 

From the song writing experiment we find that ChatGPT is very sophisticated in grammatical and verbal terms. For instance, it can produce song lyrics with metres that fit a given melody and passable rhyme schemes. However, the contents of the lyrics can be sadly misleading. From the second two experiments we find that even on factual issues, such as the location of streets where map information is readily available, ChatGPT can sometimes be inaccurate and make complete mistakes. An underlying finding is also that ChatGPT can be overly conversative in its curation. This is demonstrated by a walking route which pursues cultural diversity.

In this essay, we firstly explain some of the foundations of psychogeographical practice, such as the dérive and the importance of finding a line to follow. We then outline ChatGPT and propose how it offers possibilities for digital interpretations of psychogeography. In turn this leads into detailed accounts, supported by maps and images, of the three experiments that we pursued.

Psychogeography and walking experiments 

Psychogeography and writing from solo walks can be traced back to writers such as William Blake, Robert Louis Stevenson, and Daniel Defoe. However, it is perhaps best known in a twentieth century urban context and through exponents such as Guy Debord and the Situationist International. The latter saw psychogeography as a research tool, often with a specifically Marxist agenda, which stands to the side of physical geographers who consider the sky or the climate or more humanistic geographers who study society and economies. Debord states that:

Psychogeography could study of the precise laws and specific effects of the geographical environment, whether consciously organized or not, on the emotions and behaviour of individuals. The charmingly vague adjective psychogeographical can be applied to the findings arrived at by this type of investigation, to their influence on human feelings, and more generally to any situation or conduct that seems to reflect the same spirit of discovery.[1]

The focus on emotions and the principles of vagueness and discovery are embodied in a walking method called the theory of the dérive (or the drift). Such practice has inspired many writers, artists, and scholars. In this article we take a playful perspective on the routes created by ChatGPT and see them as dérives. Rebecca Solnit interprets the dérive as venturing into the unknown and Emma Arnold sees the walking routes produced by drifts as:

‘an attempt to get lost in the city.’[2] After looking at trends towards materialist psychogeography, occult and mythical psychogeography, and psychogeography and the walking arts, van Ratingen sees ‘the final frontier, of net-wandering, cyberflânerie [meaning to walk aimlessly in digital space], and game space’ as the future of the discipline.[3]

Emboldened by this vision of the future, the authors of this paper felt justified in basing our own psychogeographic interventions on activities that centrally involved ChatGPT. 

Introducing ChatGPT

Since it was released to the public in March 2023, the artificial intelligence functionality of ChatGPT4 has attracted a lot of attention. ChatGPT is ‘a large multimodal model capable of processing image and text inputs and producing text outputs’.[4] Its performance is, at least at first sight, almost magical. It can research and write essays, lyrics for songs, itineraries, or computer code. It can pass exams and write poems. Seen as a realistic AI robot, a report by OpenAI states that ChatGPT is able to do well in many tests that humans would find difficult, for example: 

GPT-4 was evaluated on a variety of exams originally designed for humans. In these evaluations it performs quite well and often outscores the vast majority of human test takers. For example, on a simulated bar exam, GPT-4 achieves a score that falls in the top 10% of test takers.[5]

It is also able to produce prose which is written to human standards – in other words, it passes the ‘Turing Test’,[6] by making us believe that we are reading the work of a human, not of a computer. In this mode, it can act as a prosthetic device for many students and writers without the time or the inclination to research and write for themselves. (Disclaimer: we did not use it to write any part of this essay.)

Analysts have already pointed to several problems. For a technology still under development this is to be expected. It is also true that the humans it replaces make mistakes. However, we hypothesized that, as AI appeared to approach human intelligence, would develop an ‘uncanny valley’. Mori et al. define the latter term:

Once we realize that the hand that looked real at first sight is actually artificial, we experience an eerie sensation... When this happens, we lose our sense of affinity, and the hand becomes uncanny.’[7]

Mori et al. were talking specifically of robot hands or prosthetic hands, but we believe the concept can equally apply to a prosthetic intelligence, which ChatGPT might be said to be offering us. The nearer the copy gets to total realism, without quite achieving it, the stronger the uncanny feeling.

Some of ChatGPT's errors appear, on an anthropomorphic reading, to be deliberate acts of a mischievous personality. You ask yourself - why did it do that? - even though you realise why does not mean the same thing to the programme as it does to you. We wanted to test the degree to which ChatGPT's intelligence or personality, as it came nearer to that of a human, became uncanny. How do we react to this new personality working alongside us? How does it differ from what we would expect from a human companion?   

It seems that a clear pattern of errors is emerging, firstly in the academic literature, of which there is a growing body. Kabir et al examine the way ChatGPT provides help texts for computer programmers, and find that compared to human support, there are conceptual or contextual errors.[8] ChatGPT’s answers are more incorrect, significantly lengthy, and not consistent with human answers half of the time. However, ChatGPT’s answers are very comprehensive and successfully cover all aspects of the questions and the answers. Many answers are incorrect due to ChatGPT’s incapability to understand the underlying context of the question being asked. Whereas ChatGPT makes fewer factual errors compared to conceptual errors, ChatGPT rarely makes syntax errors for code answers. According to Kabir et al ‘The majority of the code errors are due to applying wrong logic or implementing nonexisting or wrong API [Application Programming Interface], Library, or Functions.’[9]

Other analysts have found cases where ChatGPT (and similar products such as Microsoft's Bard) are simply wrong, and that when they do not know the answer, they fabricate – that is, make something up without admitting they have done so. Ruochen et al state that they make:

• Claims that conflict with the reference sources;

• Claims that don’t exist in the reference sources; and 

• Claims that don’t have a reference source and are inconsistent with multiple web sources.[10]

It has also been shown that the system depends on information it picks up, and that it may pick up information non-systemically and uncritically. This means that its judgements are sometimes biased towards whatever it has been accessing and can lack breadth and balance. McGuffie and Newhouse write that ‘GPT-3’s ability to emulate the ideologically consistent, interactive, normalizing environment of online extremist communities poses the risk of amplifying extremist movements that seek to radicalize and recruit individuals.’[11]

Many writers agree that ChatGPT is very convincing and verbose. For example, according to Kabir et al’s test of computer programming advice:

[expert] users could successfully identify the incorrect answers only 60.66% of the time and failed 39.34% of the time. Users overlook incorrect information in ChatGPT answers... due to the comprehensive, well-articulated, and humanoid insights in ChatGPT answers.[12]

In other words, these systems are highly persuasive, and even experts can be deceived one time in three. There is a sense in which, if it doesn't know the answer, it seems to rely on being verbose and sounding confident. In the real world, it offers a shortcut for students, and is a threat to journalists. Ortiz stats that it ‘can write an article on any topic efficiently (though not necessarily accurately) within seconds, potentially eliminating the need for a human writer.’[13] 

Discussion Through Three Practical Experiments: Blending Psychogeography With Digital The four authors of this article were brought together by the 4th World Congress of Psychogeography, which has been held annually for over a decade and mostly based in and around Huddersfield.[14] Some of the present authors have attended the 4WCOP regularly since 2017. As a group of four we have started to work together collaboratively since COVID-19 drove the conference away from physical practice to a small sub-culture of Zoom based panels and discussions.

In 2023 we decided to meet physically for a weekend in Huddersfield, and to hold our own mini-Congress, with a scheduled set of activities. As the AI programme ChatGPT had been made widely available since mid-2023, we decided to use it as our theme. We chose three activities, which were experienced one after other, and are described in detail below: 

1.     Write song lyrics based on Huddersfield places and characters

2.     Set up and help us to play a game of Monopoly in real streets; and

3.     Suggest itineraries we could walk to see particular types of place in Huddersfield.

The preparation was done over two months in advance of the physical gathering. Although there was communication by email, and bimonthly Zoom meetings, it was principally David who developed the Chat GPT songs, Andrea who prepared the Monopoly Board, and Tim who prepared themed walking routes.

The framework of psychogeography, as stated earlier, would allow us to test ChatGPT’s functions playfully rather than rigorously. For example, writing lyrics requires intangibles like flair and originality: the ability to take facts about the subject and to build on them to make something original or interesting. Through the latter two approaches we would embrace psychogeography’s subversion of cartography and therefore find different ways for people to navigate urban space.[15]

Experiment #1 - Chat GPT Sings Huddersfield

For this first experiment we walked between specific locations in Huddersfield, for which David had asked ChatGPT to prepare some songs. The first centred on Harold Wilson, British Prime Minister (1964-1970 and 1974-1976), who was born in Huddersfield and for whom there is a prominent statue outside the railway station. David instructed ChatGPT to: ‘Write lyrics of a song about the statue of Harold Wilson in St George's Square Huddersfield, in the style of George Formby, to be sung to the tune of Leaning on a lamp post by George Formby’. 

Each lyric was written by ChatGPT to work with a melody, so karaoke tracks could be found for these melodies. For psychogeographic use, the lyrics were read aloud by a text-to-speech engine and recorded. The words and karaoke tracks were then mixed in a DAW (Audacity) and the resulting MP3 was played at the appropriate sites in Huddersfield, using a tablet to drive a portable speaker. In Figure 1 David played an amplified version of the song on the pedestal below Harold Wilson. Aled and Andrea both held signs explaining what we were doing and with some background information. The lyrics include:

Harold Wilson led with a vision so clear,

A champion of progress, erasing fear,

He stood for equality, justice, and peace,

He believed in a future where all wars would cease.

The lyrics produced were notable for their bland superficiality. Wilson's time as Prime Minister of the UK was in fact very controversial, but there was no suggestion ChatGPT had researched this.[16] Though the lyrics roughly met the meter and rhythm of the George Formby song, they could have been written about any politician anywhere. 

In the next case David asked the LLM to ‘Write a song about Wappy Nick, a street in Huddersfield, in the style of SOS by ABBA’, ChatGPT turned Wappy Nick into a person, a rebellious hero, as follows:

Long ago, in ages gone, a legend emerged, A local lad named Nick, his spirit surged.

With a heart so wild and a mind so free,

He roamed these streets, a rebel with glee.

It is interesting that Chat GPT ignored the instruction that Wappy Nick (Figure 2) was a place, not a person. In fact, Wappy Nick is local Yorkshire dialect meaning a nick or a shortcut, that is wappy or narrow and crowded. Looking more closely at the lyrics, Nick's rebellion is vague – ‘painting the street with colours so profound’ – whilst Huddersfield has a long tradition of rebellion and agitation. There are many examples freely available on the internet, such as Luddites in 1812, Chartist agitation for political reform in the 1840s, agitation for workhouse and public health reforms in the 1850s.

It was notable that more people stopped and paid attention to the Wappy Nick song than had Harold Wilson, potentially due to the location being a useful thoroughfare from the main market area. However, had the public stopped to listen people would have noted that vague wordiness replaces any research into actual people or places.  

Moreover, there seemed to be a sense of being overly upbeat or optimistic. We return briefly to the lyrics about Harold Wilson and find a similar mood.

So let us gather 'round the statue tall,

Remembering Wilson, who inspired us all,

Together we'll forge a future full of hope,

Standing strong, hand in hand, as we cope.

OpenAI's own report refers to the system as ‘committed to promoting positive and inclusive content’ and there is a strong bias towards Polyanna optimism.[17] Presumably for legal reasons there seems to be a lack of any criticism of any person. All the lyrics we asked ChatGPT to write had this same feeling, reminiscent of US motivational posters – just have faith and dream big and all will be well. Failure or negativity are not contemplated. Wappy Nick, the supposed local hero, has a similar nostalgic or optimistic treatment.

Though Nick is gone, his spirit remains,

In every laugh and every beat that sustains.

Wappy Nick, a name etched in our hearts,

A symbol of freedom, where creativity starts.

Words like 'freedom', 'creativity', 'spirit', 'hope' and 'inspired' are bandied around like tokens but never seriously located in place or examined.  

In the final example from ChatGPT sings, we asked for lyrics about: ‘The Girl from Aspley Basin, based on The Girl from Ipanema, which is a song about a man admiring a beautiful girl as she walks past.’ The results included the following lines:

Oh, the girl from Aspley Basin,

Where the waters flow with tales of time, Transhipping goods, a thriving trade, ah!

Oh, Through the heart of England, 

Britain’s highest waterway,

Standage Tunnel, longest in the land...

The disconnected 'ah!' is quite eloquent! It is the only shred of emotion in what otherwise sounds like a travel brochure, or an economic history text focused solely and unglamorously on barge traffic. Huddersfield was once the centre of the heavy woollen industry in the UK, and Aspley Basin, where two canals joined, was a major transport site. Of note, ChatGPT's lyrics ignore the request about the man gazing upon a beautiful woman. We know from an article on Wikipedia,[18] and presumably available to ChatGPT, that wolf whistles were made at the 'girl from Ipanema' at the time the song was being written. Since this would now be seen as harassment, we suspect ChatGPT leaves the whole relationship issue alone, misinterpreting the song and blanking an important area of human activity. It is too keen not to offend and goes too far towards blandness and self-censorship.

Playing the songs had involved stopping in specific locations for approximately quarter of an hour to prepare the soundtrack and play the song. The second experiment, taken after lunch, would involve more movement as dérives were based on walking lines set or inspired by ChatGPT.

Experiment #2 - Monopolising Huddersfield

For this second walking expedition we asked the website to follow the principles of Monopoly.  

User: If you were setting up a game of Monopoly for Huddersfield, which streets would you put on the board?

ChatGPT responded politely, as it often does, by providing a list of 20 street. These seemed to pay attention to a spread of both expensive and relatively cheaper locations. Andrea tweaked the list of locations so that some were in the very centre of the town. In the spirit of the Situationists, she then numbered 11 destinations and offered a pair of dice to those on the walk. The total of the two dice would decide where each team would go. Chat GPT would then come back into the equation as it would direct teams to the location. 

Here we relate the experience of following this gamified experience of Huddersfield. As Aled is not local to Huddersfield, and therefore needs to trust the instructions provided, his perspective was both critical and potentially more generous. The images shared here (Figures 3, 4 and 5) are taken from a WhatsApp chat, which was used by two walking teams as they shared their experience in real time. Aled was in Team A with Tim, whilst Tim Waters was in Team B, accompanied by Phil Wood and Andrea.

As shown in Figure 3, Team A rolled Park Drive and proceeded to get directions from ChatGPT. Team B had rolled Fitzwilliam Street and this was found within four minutes of leaving the starting point at Huddersfield Railway Station. Effectively this was an easy move as it was one road away. ChatGPT directed Team A in a broadly southerly direction to what it assumed was a roundabout. Teams A had to imagine themselves as cars as they found the third exit from the circular junction. Straight away they decided that they would fall back on Google Maps to take them to the next destination; a rule they subsequently followed should the directions not be workable. This was particularly important as the directions around the imaginary roundabout were not clear about going clockwise - or in the opposite direction depending on the local highway law. Indeed, Google Maps was critical to helping Team A find the first stop at Southgate. As shown on the GPS-tracked route (Figure 6), Southgate is east of the Railway Station. Team A was then given the instruction to find Chapel Hill. As shown in the chat, Team B had already arrived there through the use of local knowledge. Team A were asked to follow the ring road around the whole town. They started to feel that the routes offered to them were designed for cars rather than pedestrians. 

Figure 4 shows that both teams were pursuing the same target. Both teams met in person near Park Drive and discussed how ChatGPT was unreliable as a walking guide. As such, we started to doubt the logic. Was it thinking as if we were cars? Was it running a request on Google Maps and sharing an edited version of the response? As we had doubts Tim started to ask more specific question to ChatGPT. As shown in Figure 5, the website offered Google Maps as a potentially more reliable way to get the ‘shortest’ route. The experience for Team A then descended into a farce, especially as the non-native Aled wanted to trust the literal offerings of ChatGPT.

In the final 40 minutes of the experiment Team B were increasingly reliant on their local knowledge. Under time pressure, Team A asked Chat GPT for the ‘quickest’ walking route between two points. The response was apologetic, stating that its knowledge was based on information available up to September 2021. As such it recommended using a map and navigation service or a dedicated mapping app like Google Maps, Apple Maps, or a GPS device. With time on their hands, and an open mind, this proved to be pure psychogeographical pleasure for Tim and Aled. They followed limited instructions and were taken through an underground car park, up some stairs and passed their original starting point. This surrender to the absurd helped to attain a state of flow or pleasure in our urban explorations. Wilson writes about how the quality of ‘jouissance’ appealed to spatial theorist Henri Lefebvre.[19] 

Though Team A’s 6.8km walk (Figure 6) had many interesting twists and turns, we bring together a summary from both groups before considering the next technique. The first point is that Chat GPT gives incomplete directions for places some distance away and not on main roads. Effectively ChatGPT had turned a 0.9km walk for Team A from the Railway Station to Park Drive into something 2.9km long. Moreover, neither team would have found the majority of the destinations offered by ChatGPT aside from easy moves, such as from railway station to Fitzwilliam Street. There are implications for people who have restricted mobility, limited time or indeed people who invest trust in this form of artificial intelligence. Having made this criticism, it is worth commending Chat GPT for instructions written in sentences that can be easily understood. Related to this promising latter point, some form of syntax may need to be developed for the user to ask ChatGPT for directions.

This exercise concluded Day One of our activities. We retired for dinner and to refresh ourselves for the third activity. We were prepared for the types of mission that Chat GPT would send us on and so would be open minded. 

Experiment #3 - Chat GPT Writes Tourist Routes through Huddersfield

ChatGPT has been used to compose tourism itineraries for different locations. For example, Rivera et al. considered AI-generated walking visits to five locations in America.[20] In most cases the results were overly ambitious for the time specified in the requests, and sometimes ChatGPT made factual errors. In this final task, we asked if ChatGPT could harvest the internet to guide walking dérives through Huddersfield, a town a little less visited. On Sunday morning Tim brought a pre-prepared collection of printed laminated prompts. He had asked for walks in Huddersfield less than two miles long: firstly passing locations that exemplify cultural diversity and; secondly, following a route which would take in streets and roads named after famous people. 

In search of cultural diversity our group of five took a line from the railway station, passing the statue of Harold Wilson, and straight to the old covered market. Chat GPT had offered a fairly simple move down one street, with a right turn. So far, so good. We then had to find the Sikh Temple (Gurdwara Singh Sabha) on Prospect Street. Local knowledge confirmed that this place did not exist at that location. We then moved to the Hudawi Cultural Centre on Great Northern Street. We wiggled around to navigate the Tesco supermarket and to find a safe route to cross the A62 ring road. It started to rain as we rounded one corner, so we took shelter under a tree, where a local couple asked if we were lost. We proffered the laminated instruction to explain our Chat GPT mission, and they offered some suggestions. This human interaction differed a little from the previous day, when nobody had offered help. Perhaps we had looked more focused and busier on the Saturday? Or maybe there is something to be said for holding a piece of paper as you navigate space.

Our journey then took us down a very overgrown old road, replete with used prophylactics which confirmed reputations of the nearby night-time attractions. Similar to the pursuit of Monopoly locations, this felt like proper psychogeography: we had a seemingly ridiculous mission and had the time and mental space to be playful. We came across the incongruous Shaks Cars, which was selling Ferraris and other expensive brands of vehicle. In writing this post-event account, we have been overlaying our walked line on to Google Maps and Streetview. In the digital world Shaks allow people to enter the glass-fronted building from Google StreetView to see close-up shots of the leather interiors as if we were inside the cars. 

The next instruction asked us to find Huddersfield University. From the street we could see a big sign for the university. Satisfied with bagging this destination, we would then seek out the Piazza Shopping Centre (closed after 2021) and then the Lawrence Batley Theatre. Along our way we saw Chickano’s (Figure 7) - which may have qualified under cultural diversity as it means a native of, or descendant from, Mexico who lives in the United States. 

As we walked to the theatre, we met another local who offered directions. Seeming to be lost, and in need of help, is often the starting point to some fascinating discussions. We had a surreal ending to our dérive (Figure. 8) as we were greeted at the Lawrence Batley by a lifesize dinosaur swooshing its tail about the ground.   

We regrouped over a coffee and considered how ChatGPT had shown us cultural diversity. Firstly, it had been inaccurate as many destinations were not actually in Huddersfield. There was also a reliance on cultural centres and particularly religious influences from the Indian subcontinent. We were not directed to the Irish or the Catholic Centre, nor Chickano’s. In the same vein, being taken to the theatre was perhaps an overly obvious destination. Though unintentionally, this mission did lead on an excellent adventure in urban ethnography and was a delight for those who are psychogeographically oriented.

After a pause we followed a second mission: ‘User: Tell me a route to walk in Huddersfield using roads named after famous people’. The first four lines could only be ignored, as they referenced a non-existent Patrick Stewart Street. The fifth instruction offered a small crumb of hope:

Head west along Somerset Road until you reach Edith Road. Edith Road is named after the suffragette Edith Key, who was a prominent figure in the women’s suffrage movement.

Consulting Google Maps, we found an Edith Key Building, rather than an Edith Road. We walked towards the campus of Huddersfield University, searching for the building inspired by the aforementioned suffragette. Of note, the reception building was open and the staff gave us a printed map showing the buildings. Glancing at this tangible paper document it felt clear that many buildings had been explicitly named or inspired by people. For example, there is the huge edifice adorned with a poem by Lemn Sissay [21] (Figure 9). We found the Edith Key building and also uncovered the drama department, which was indeed named after Patrick Stewart. Leaving the campus, and crossing the ring road for the umpteenth time, we looked at the instructions. We found one street named after the Queen and the adjacent Albert Yard.

Our journey concluded at this point and so we stopped for lunch. Reflecting on this last experience, it was clear that Chat GPT had mostly failed to give us anything particular useful. In fact, the only useful offering was Edith Key, which in turn led to some interesting findings on the university campus. However, Huddersfield is a town with a lengthy street named after a John Williams and with geographical references to the names Ramsden and Byram. Were these people not famous and relevant to the town? Maybe a future version of Chat GPT will reference this writing and try to assemble a route from this text, with a special note attributed to Wappy Nick and the Girl from Aspley Basin. Having described and analysed our three experiments we now offer some concluding thoughts.

Conclusions

Our experiments are described frankly and honestly as a work in progress, taken with an open mind, and in the spirit of the walking dérive never being completely futile. As our experiments demonstrate, ChatGPT is approachable in a way which is new to many people. In theory you can ask it a question about anything, as long as it happened before 2021, simply by typing your question in to a text box, using plain English, and expect to get the sort of answer an intelligent human might provide. You can ask it to compose songs, demand directions as you walk, or you can ask wider questions based on places and its reaction to them. No programming skills are required, though it is sometimes necessary to 'game' the system to get good results. 

In our tests, ChatGPT was impressive, but not always for the right reasons. Its fluency and its verbal ability were sometimes excellent, even overpowering. However, as these skills drew it nearer and nearer to the edge of the ‘uncanny valley’, as defined by Mori et al. it became more and more clear that it often did not think like a human would.[22] It was not good at seeing things in context, particularly how streets connected with each other, and not good at understanding the practicalities of humans walking between different points. If ChatGPT had been a friend walking round with us, we would have been confused by its lack of empathy, and of a sense of humour. We would be in one of Huddersfield’s fine pubs and it would still be wondering the ring road in the dark. 

Chat GPT could not see the wider picture, and instead seemed to produce ‘satisficing’ results, going so far but no further. It has clear ‘guard rails’ with controversial subjects it will decline to discuss. It is extremely verbose and produces a torrent of words that can seem to carry conviction. In one case, where we were trying to work out how long a walk would take, we asked ‘How can Chat GPT help me with time?’ The response offered ten options, including one related to scheduling. In reality none of the options helped us get from Quay Street to Water Street in Huddersfield. None of us felt that we could develop a relationship with ChatGPT.  

Criticisms aside, ChatGPT is a brilliant achievement. This is demonstrated precisely because we feel an uncanny valley when we see it in action. Just as a prosthetic body part has to look realistic before the uncanny valley effect can form, here we have a prosthetic intelligence which is in many ways so like our own that uncanny valleys form – in anthropomorphic terms we feel we can't understand what our friend is doing, because we try to see it as a human intelligence when it actually works in a very different way to our own. Speaking as psychogeographers we enjoyed how ChatGPT provided us with walking lines through urban spaces in Huddersfield. As such this form of AI allowed us to experience what Rebecca Solnit values from the dérive - namely to explore space in a creative sense – and certainly fulfilled Emma Arnold’s desire to get lost in urban space. For example, through a pursuit of cultural diversity we were taken to the sites where prostitutes worked and pondered the cost of expensive cars.

Finally, it is worth stating that we did not require ChatGPT to help us with any life-or-death situations. Moreover, we were never going to be properly lost as we had an array of navigational technologies, paper maps, local knowledge and benefited from the people of Huddersfield offering directions – as they did on the cultural diversity tour. We trust that our experiments have revealed the limitations of ChatGPT and shown that overly-relying on AI systems runs risk of debasing our own view of the world, leading to blandness, lack of humour and wit, focus on only approved emotions such as bland optimism, and a refusal to contemplate the more serious or complex aspects of our lives.

Authors

Aled Singleton researches and teaches geography at Swansea University, Wales, UK a.m.singleton@swasnsea.ac.uk. 

Andrea Freeman is a psychogeographer and performance artist https://pinkandgray.co.uk/. 

Tim O'Connor is a psychogeographer based in the Colne Valley. Yorkshire, UK. 

David Upton is a computational artist www.codedwalls.com.

Notes

1      Debord Guy. ‘Introduction to a Critique of Urban Geography’

https://theanarchistlibrary.org/mirror/g/gd/guy-debord-introduction-to-a-critique-of-urban-geography.pdf (1955: 1). 

2      Solnit, Rebecca. Wanderlust. London: Granta (2014); Arnold, Emma. ‘Aesthetic practices of psychogeography and photography. Geography Compass 13, no 5.: e12419. https://doi.org/10.1111/gec3.12419 (2019: 2)

3      van Ratingen, Witold Jerzy. ‘Loitering with Intent: The Histories and Futures of 'Psychogeography'.’ Master of Arts Thesis. New York City: The New School for Social Research Department of Liberal Studies (2014: 216).

4      OpenAI ‘GPT-4 Technical Report’ (2023: 1) accessed via arXiv:submit/4812508. 

5      ibid

6      The phrase ‘The Turing Test’ refers to a proposal made by Alan Turing (1950) as a way of dealing with the question whether machines can think.

7      Mori, Masahiro, Karl F. MacDorman, and Norri Kageki. ‘The uncanny valley [from the field].’ IEEE Robotics & Automation Magazine 19, no. 2 (2012: 98): 98-100. DOI:10.1109/MRA.2012.2192811

8      Kabir, Samia, David N. Udo-Imeh, Bonan Kou, and Tianyi Zhang. ‘Who Answers It Better? An In-Depth Analysis of ChatGPT and Stack Overflow Answers to Software Engineering Questions.’ arXiv preprint arXiv:2308.02312 (2023).

9      Ibid: 6

10   Zhao, Ruochen, Xingxuan Li, Yew Ken Chia, Bosheng Ding, and Lidong Bing. ‘Can chatgpt-like generative models guarantee factual accuracy? on the mistakes of new generation search engines.’ arXiv preprint arXiv:2304.11076 (2023: 1).

11   McGuffie, Kris, and Alex Newhouse. ‘The radicalization risks of GPT-3 and advanced neural language models.’ arXiv preprint arXiv:2009.06807 (2020: 10).

12   Kabir et al: 9 

13   Ortiz, Sabrina. ‘Need an AI essay writer? Here's how ChatGPT (and other chatbots) can help’ accessed via https://www.zdnet.com/article/how-to-use-chatgpt-to-write-an-essay/ (2023: 23)

14   In the general spirit of playfulness that characterises psychogeography, it is held as the ‘fourth’ congress every year, and since it is never formally closed, resumes as the ‘fourth’ the year after https://www.4wcop.org. 

15   Pinder, David. ‘Subverting cartography: The situationists and maps of the city.’ Environment and Planning A 28, no 3 (1996): 405-427. https://doi.org/10.1068/a280405

16   Wilson struggled with the left wing of his own party and his successful attempt to maintain a government on an eight-seat parliamentary majority led many to see him as a tactician rather than an ideological purist.

17   OpenAI: 90

18   The Girl from Ipanema was written in 1962 and based on a person called Helô Pinheiro. The authors of the song noticed that men who wolf whistled at her as she walked on the street https://en.wikipedia.org/wiki/The_Girl_from_Ipanema

19   Wilson, Japhy. ‘Review: Henri Lefebvre, Toward an Architecture of Enjoyment’ Antipode: A Radical Journal

(2014) https://antipodeonline.org/wp-content/uploads/2014/10/book-review_wilson-on-lefebvre.pdf

20   Rivera, John, Eve Chen, Nathan Diller, Zach Wichter, and Kathleen Wong. ‘Make travel easy: We tested ChatGPT itineraries in 5 US tourist spots’. USA Today, July 9, 2023. https://eu.usatoday.com/story/travel/2023/07/09/ai-plan-travel-chatgpt/70380395007/

21   Let There Be Peace by Lemn Sissay

Let there be peace

So frowns fly away like albatross

And skeletons foxtrot from cupboards,

So war correspondents become travel show presenters And magpies bring back lost property, Children, engagement rings, broken things.

Let there be peace

So storms can go out to sea to be

Angry and return to me calm,

So the broken can rise up and dance in the hospitals.

Let the aged Ethiopian man in the grey block of flats

Peer through his window and see Addis before him, So his thrilled outstretched arms become frames For his dreams.

Let there be peace

Let tears evaporate to form clouds, cleanse themselves And fall into reservoirs of drinking water.

Let harsh memories burst into fireworks that melt

In the dark pupils of a child’s eyes

And disappear like shoals of silver darting fish,

And let the waves reach the shore with a

Shhhhhhhhhhhhhhh Shhhhhhhhhhhhhhhh Shhhhhhhhhhhhhhhh

22 Mori et al.: 98