Bus Bunching and Bus Bridging: What Can We Learn from Generative AI Tools like ChatGPT?

Abstract

Regarding tools and systems from artificial intelligence (AI), chat-based ones from the area of generative AI have become a major focus regarding media coverage. ChatGPT and occasionally other systems (such as those from Microsoft and Google) are discussed with hundreds if not thousands of academic papers as well as newspaper articles. While various areas have considerably gone into this discussion, transportation and logistics has not yet come that far. In this paper, we explore the use of generative AI tools within this domain. More specifically, we focus on a topic related to sustainable passenger transportation, that is, the handling of disturbances in public transport when it comes to bus bunching and bus bridging. The first of these concepts is related to analyzing situations where we observe two or more buses of the same line following close to each other without being planned deliberately and the second is related to the case where buses are used to replace broken connections in other systems, such as subways. Generative AI tools seem to be able to provide meaningful entries and a lot of food for thought while the academic use may still be classified as limited.

1. Introduction

Artificial intelligence (AI) was a mouthwatering technique especially in the 1980s, with success stories in quite a few dimensions and related textbooks such as [1]. Recently, based on powerful improvements and advances in information technology and computer science as well as in information systems research, AI has gained popularity again. Media coverage is high and mostly addresses ChatGPT as a generative AI tool, or simply a chatbot which mimics communication based on text or voice interaction. Current applications work in natural language and may be used as an addition to communication over a webpage even in natural language (written or spoken). Usage of modern technology coined, e.g., deep learning and machine learning, seems most important in this respect. Besides spoken or written text, we also see the generation of software, images, or other types of documents. The technology is trained based on large language models (LLMs) in response to user questions, requests, and prompts. A mouthwatering general introduction can be found in, e.g., [2].

ChatGPT has been built by the company OpenAI (see, e.g., https://openai.com/blog/chatgpt; last accessed on 8 May 2023). We may also distinguish between two versions of ChatGPT, a free version and a commercial version, which differ based on the used LLM. The company Google propagates the system Bard (see, e.g., https://blog.google/technology/ai/bard-google-ai-search-updates/; last accessed on 8 May 2023). Note that Bard may not be available in many countries (see, e.g., https://bard.google.com/faq?hl=en; last accessed on 8 May 2023). The company Microsoft uses Bing and utilizes the LLM GPT4 from OpenAI. There is also a system available in China called Ernie which is (planned to be) available on Baidu [3].

In this paper, we aim to investigate the freely available versions of ChatGPT and the Bing chat and compare them based on user experience within specific requests in the transportation and logistics domain. All entries were conducted on 8/9 May 2023 using the ChatGPT release from 3 May 2023; see https://help.openai.com/en/articles/6825453-chatgpt-release-notes. With respect to Bing, the recently released version from 4 May 2023 is used. See, e.g., https://blogs.microsoft.com/blog/2023/05/04/announcing-the-next-wave-of-ai-innovation-with-microsoft-bing-and-edge/ (last accessed on 10 May 2023). Questions (Q) have been asked in English from an internet access point in Northern Germany. Answers will be commented and if the self-test goes wrong or produces wrong answers, this is indicated. In a sense, we utilize a narrative style of conducting interviews published in scientific journals; see, e.g., [4]. More specifically, we focus on certain topics within public transport.

Currently, we see a wealth of AI applications in almost any area of practical life, be it in education or—as is the focus in this paper—in public transport. The question is not whether there is an importance and/or necessity of using AI in transportation and logistics. Rather, we claim that it is already in use or it will be used sooner or later and we need to understand it. This is what the paper tries to support.

In real-world public transport systems, we may be forced to focus on quite a few issues related to robustness and disturbances. A comprehensive survey on this topic is provided in [5]. Among the specific areas dealt with in this respect, we can find bus bunching and bus bridging. Often these issues can be characterized as a problem setting in a public transport system (bus bunching) as well as regarding a solution concept (bus bridging). Using a bus-based public transport system, bus bunching relates to unplanned situations where two or more buses (usually of the same line) follow close to each other. This means that the headway approaches a value of (almost) zero, where headway is the distance or duration between vehicles in a transit system measured in space or time. The other case, bus bridging, is mostly applied in cases where buses are used to help overcome broken connections in other systems, such as trams, railways, or subways.

The remainder of the paper is organized as follows. In Section 2, we sketch some of the necessary literature to provide a proper introduction into the different areas touched on in this paper. This is followed by two sections on the application of generative AI tools regarding bus bunching (Section 3) and bus bridging (Section 4). Finally, some conclusions are drawn together with a few hints regarding the needs for future research.

2. Literature Review

We first give a very brief review with pointers regarding generative AI. After that, we discuss the recent literature on bus bunching and bus bridging.

2.1. Generative Artificial Intelligence

Q: “What is generative AI and what is the difference to conversational AI?”

Bing: “Generative AI is a category of AI algorithms that can create new and realistic content from the data they have been trained on. Conversational AI, on the other hand, is a type of AI that can simulate human conversation using natural language processing (NLP) and machine learning (ML). Conversational AI can use generative AI to produce responses that are appropriate and relevant to the conversation. For example, ChatGPT is a generative AI chatbot that uses NLP and ML to interact with humans. Conversational AI can be used for various purposes, such as customer service, virtual assistants, speech recognition and text-to-speech”.

ChatGPT: “Generative AI refers to a type of artificial intelligence that is designed to generate new, original content based on existing data or models. This can include generating text, images, music, or even entire virtual environments. The goal of generative AI is to create new content that is both original and compelling, and that can be used in a variety of applications such as entertainment, art, and advertising.

Conversational AI, on the other hand, is focused on creating intelligent systems that can interact with human users in a natural and conversational way. Conversational AI uses natural language processing, machine learning, and other AI technologies to understand and respond to human language, with the goal of creating a more seamless and intuitive user experience.

While both generative AI and conversational AI use similar AI technologies, the main difference between the two is the type of output they produce. Generative AI is focused on creating new content, while conversational AI is focused on interacting with users in a natural and conversational way.

Some examples of generative AI include language models like GPT-3, which can generate realistic text based on input prompts, and deep learning models like StyleGAN, which can generate realistic images of people, animals, and other objects. Examples of conversational AI include virtual assistants like Siri, Alexa, and Google Assistant, as well as chatbots and other conversational interfaces that can be used to interact with customers and users in a natural way”.

Major companies are providing solutions and explanations in the realm of generative and conversational AI (see, e.g., https://www.ibm.com/topics/conversational-ai; last call 11 May 2023). A most comprehensive discussion of using conversational generative AI can be found in [6]. A related text regarding education is, e.g., [7].

As mentioned above, generative AI can be a system using AI techniques able to generate text, software, or other types of documents in response to questions or requests; see, e.g., [2]. ChatGPT has been widely and openly discussed in the public and, according to popular search engines, there are already thousand of articles with ChatGPT in their titles or mentioning it in their abstracts, as emphasized in the survey/review from [8]. This also includes quite a few more or less critical commentaries judging challenges and opportunities of ChatGPT; see, e.g., [7]. It seems important to both understand the technology as well as its limitations and even imagine possible misuses. A comprehensive study comparing human answers to ChatGPT answers can be found in [9]. That is, linguistic analyses of ChatGPT-generated content compared with that of humans can be found and is well-documented (the dataset, code, and models are publicly available at https://github.com/Hello-SimpleAI/chatgpt-comparison-detection; last accessed on 30 April 2023). Various areas of research received commentaries and initial studies regarding the use of ChatGPT, including, e.g., environmental research [10], medicine [11], and history [12]. The latter study is interesting as the author claims to see some political bias in what is presented by the system. While we do not comment on the specific result, we echo the need to be careful regarding possible misuses of ChatGPT [13]. Furthermore, ethical concerns are always apparent, as indicated, e.g., in [14,15].

The discussion of using generative AI is closely related to cognitive intelligence, which is referred to as human mental ability and understanding developed through thinking, experiences, and senses [16]. It is the ability to generate knowledge by using existing information (as is also a common and well-known issue in information management; see, e.g., [17]). We do not claim that ChatGPT and generative AI tools are able to successfully pursue this generation, but the fear of related claims is there.

Finally, we should note that comparative studies of different chatbots may also be conducted but are still rare; see, e.g., [18]. Some insights into related methodologies are briefly surveyed, e.g., in [19]. The use of different chatbots is seen in the context of what has happened during the last 20 years regarding the use of chatbots, especially in supply chain management, where simple tasks such as tracking shipments or managing inventory are accompanied by related tools. Marrying NLP and optimization is also on the horizon, with the first efforts being on the way [20].

Generative AI has potential applications across a wide range of industries and domains, especially in education (see, e.g., [21]). Current studies on ChatGPT mostly report on experiences by asking questions and providing as well as discussing the answers. Based on this, we perform some self-tests, as reported below.

ChatGPT has not yet received the same attention within logistics as within other areas; see, e.g., [13,22,23]. Examples addressing logistics in specific domains as well as supply chain management include [24,25]. In general, however, the transportation and logistics domain has not yet been comprehensively covered regarding the description of using generative AI systems. Moreover, while attempting to showcase the opportunities and future applications, the limitations regarding content also need to be exemplified. In [13], some insights regarding the use of ChatGPT in the logistics domain are discussed by investigating a specific problem in stochastic vehicle routing, which has not yet been comprehensively covered in the academic literature. In [26], the context is intelligent vehicles, an area that might have a major influence regarding autonomous vehicles in the future. Other examples touching on the logistics domain include [27].

An interesting and necessary twist in considering generative AI tools relates to discussing human aspects, which connect AI tools to ethical questions as well as human-centered technology (and society). Often, issues of a so-called brave new world with overabundance and issues of simplifying everyday life are often discussed with respect to new streams of technology. The usage of generative AI belongs to this realm. That is, in different words, from a societal perspective as well as human aspects, the area of generative and conversational AI touches on ethical issues. Regarding the area of education, this is documented, e.g., in [28]. Notions like dependable AI [29] and trustworthy AI [30] have been proposed and discussed. The proposal of the latter came in response to what the authors call mounting public criticism of AI systems, in particular with regard to the proliferation of such systems into ever more sensitive areas of human life without proper checks and balances. This is exactly what happens or is seen as a major issue or fear regarding conversational and generative AI.

In Europe, the High-Level Expert Group on Artificial Intelligence had presented some Ethics Guidelines for Trustworthy AI [31]. These guidelines may be seen as an important step for the governance of AI but they might also distract efforts from genuine AI regulation. The guidelines put forward a set of seven key requirements that AI systems should meet in order to be called trustworthy:

• Human agency and oversight;
• Technical robustness and safety;
• Privacy and data governance;
• Transparency;
• Diversity, non-discrimination, and fairness;
• Societal and environmental well-being;
• Accountability (incl. auditability).

Other places around the world have also seen the appearance of related issues to better human existence. Notable here might be Society 5.0, which may be called Japan’s concept of a technology-based, human-centered society [32]. It is assumed that technology such as AI will permeate all areas of life, including, e.g., healthcare, the environment, scientific research, and ethics. In our context, this may also call for connections to public transport as very roughly touched on by [33,34]. Without calling it Society 5.0, the fact that humans and machines coexist in harmony, e.g., based on proper use of data and information, can be traced back for decades, as indicated in information management approaches in traffic and transportation; see, e.g., [35]. Common sense may actually help to include concepts from other areas such as, e.g., mystery shopping out of the marketing domain into public transport, such as when designing bus stations [36].

2.2. Bus Bunching and Bus Bridging

2.2.1. Bus Bunching

A comprehensive list of references on bus bunching can be found in [5] (see esp. Section 5.7 and Table 7 in that work). A most influential and most often cited earlier work in this area seems to be [37]. Most attractive for us, though for a different reason, is [38], i.e., due to the nice writing in self-coördinating as part of the title. While the original meaning of bus bunching usually refers to just one bus line, the consideration of multiple-origin bus operation can be found, e.g., in [39]. Defining bunching swings as repeating patterns of pairs of delayed and bunched vehicles can be found in [40]. Together with [41,42,43], the latter paper may also be classified as using machine learning tools.

Very recent papers beyond [5] include the following. In [44], the idea is that passengers are provided with real-time waiting time information and degrees of in-vehicle congestion. If properly applied, the passengers’ boarding choice behavior is adjusted rather than controlling the buses themselves. In [45], it is argued that there may be reasons to adequately adjust the control methods to let link and stop headway deviations counteract each other, so that the bus system can automatically mitigate the headway deviations.

2.2.2. Bus Bridging

Classical methods available from vehicle and crew scheduling can be used and modified to solve bus bridging problems. A comprehensive list of references on bus bridging can be found in [5] (see esp. Section 5.6 and Table 6 in that work). A most influential and most often cited earlier work in this area seems to be [46]. In [47], a bus bridging service is described integrating a routing strategy for express and short-turn routes and exemplified for a Melbourne (Australia) case study. An integrated optimization framework is proposed to jointly optimize the bus bridging routes and timetables under time-varying demand. A robust version considering uncertainty of travel time can be found in [48]. A queuing-model-based tool to support bus bridging cases together with a case study in Toronto (Canada) is provided in [49]. (It should be noted that the references from this paragraph have already been included in [5]).

More recent works have almost exploded in number, including [50], where coordination between passenger delay during the bus bridging and a metro short turning process is attempted. A case study is described for a metro line in Shenzhen (China). A bus bridging case study for the Hong Kong MTR is provided in [51]. Assuming that extra buses are taken from different depots, an interesting problem arises as to where to locate buses for eventually appearing bridging events. In [52], a covering location model is applied and a case study is provided for Tianjin (China). A case study for Nanjing (China) is described in [53], although the algorithmic concept includes a misconception. A recent consideration of short turning policies which can support disturbances as well as bus bridging situations with a case study in Istanbul (Turkey) can be found in [54].

3. Bus Bunching in Light of Generative AI

Q: “What is bus bunching?”

ChatGPT: “Bus bunching is a phenomenon that occurs when two or more buses that are scheduled to arrive at regular intervals on a particular route end up arriving at the same time or very close together. This results in a gap in service for passengers waiting for the next bus to arrive, followed by several buses arriving in quick succession. Bus bunching can be caused by a variety of factors, including traffic congestion, passenger loading and unloading times, and differences in bus driver behavior. When a bus experiences delays, it may become even more delayed as it picks up more passengers and falls further behind schedule, leading to a domino effect where subsequent buses are delayed as well. Bus bunching can be frustrating for passengers, as they may have to wait longer than expected for a bus and then face overcrowding and potential delays once the bus arrives. To mitigate the effects of bus bunching, transportation agencies may implement strategies such as dynamic scheduling, which adjusts bus headways in real-time based on traffic and passenger demand, or bus lanes and priority signals to reduce congestion and improve travel times”.

Bing: “Bus bunching is a phenomenon where two or more buses that are supposed to run at regular intervals along a common route end up arriving together at the same stop. This happens when a bus falls behind its schedule due to traffic congestion, heavy passenger load, or other factors, and the following bus catches up with it. Bus bunching reduces the reliability and efficiency of bus service, and increases the waiting time and frustration of passengers. There are different strategies to prevent or reduce bus bunching, such as holding buses at control points, skipping stops, or short-turning some trips”.

A more generic question may be asked regarding the importance of bus bunching for sustainable passenger transportation.

Q: “How important is bus bunching regarding sustainable passenger transportation?”

ChatGPT: “Bus bunching is an important issue for sustainable passenger transportation because it can impact the efficiency, reliability, and attractiveness of public transportation services. When buses bunch up, it can lead to longer wait times and increased frustration for passengers, which can discourage people from using public transportation. This can lead to an increase in single-occupancy vehicle use, which contributes to traffic congestion and air pollution.

In addition, when buses are not operating efficiently, it can lead to increased fuel consumption and greenhouse gas emissions, which can contribute to climate change. By optimizing bus operations and reducing bunching, transit agencies can improve the overall efficiency of their services, reduce emissions, and encourage more people to use public transportation.

Furthermore, efficient and reliable bus operations can also have positive economic impacts, by reducing the cost of operating transit services and making public transportation more attractive to businesses and residents. This can help to promote economic development and reduce the need for expensive new infrastructure projects.

Overall, bus bunching is an important issue for sustainable passenger transportation, and efforts to address it through the use of technology and data analytics can help to improve the efficiency, reliability, and attractiveness of public transportation services”.

Bing: “Bus bunching is a common phenomenon that undermines the reliability and efficiency of bus services. It refers to a bus arriving with a little headway to its predecessor. This can result in some buses being overcrowded with passengers while others carry only a few, which means inefficient utilization of the bus capacity and an uncomfortable service for passengers. Thus, reducing bus bunching is beneficial to both passengers and the transit agency.

There are various strategies being developed to alleviate bus bunching, such as holding buses at control points, short-turning strategy and multi-agent reinforcement learning. These strategies aim to improve the resilience of bus services and provide quality service to attract more passengers to the system.

In summary, bus bunching is an important issue regarding sustainable passenger transportation and various strategies are being developed to alleviate it”.

An interesting twist might arise if open research topics from the latest research are utilized to ask questions. For instance, we might ask whether it is possible to motivate a specific vehicle routing or vehicle scheduling problem by means of bus bunching. In a recent publication on bus bunching analysis for a bus rapid transit corridor in a major Mexican city by [55], the corridor has the functionality of a round-trip one. Based on that, only available buses leave the end station. Thus, in case of very severe situations based on bunching or other disturbances, there might not be enough buses available for departure according to the given schedule. This will eventually lead to subsequent stochastic problems with an uncertain number of vehicles. Related research in that context is specified per se and in terms of ChatGPT usage in the following works: [13,56,57].

Q: “Is it possible to motivate a stochastic vehicle problem with an uncertain number of vehicles through bus bunching?” (This was also specified as follows: “stochastic vehicle routing problem” and “stochastic vehicle scheduling problem” without any significant change in the somewhat negative answer in the second-last sentence of the answer.)

Bing: “Bus bunching is a common problem in public transport systems and is characterized by a negative feedback loop between service headways, number of boarding passengers and dwell times. In uncontrolled bus systems, buses tend to bunch due to the stochastic nature of traffic flows and passenger demands at bus stops. It is widely acknowledged that bus bunching affects transit operations increasing passenger waiting times and variability. It is possible that a stochastic vehicle problem with an uncertain number of vehicles could be motivated through bus bunching. However, I couldn’t find any specific information on this topic. Would you like me to search for more information on this topic?”

Q: “Yes, please search for more information”.

Bing: “I’m sorry, but I couldn’t find any specific information on motivating a stochastic vehicle problem with an uncertain number of vehicles through bus bunching. Is there anything else I can help you with?”

ChatGPT: “It is possible to use bus bunching as a motivation for a stochastic vehicle routing problem with an uncertain number of vehicles, particularly in situations where the number of vehicles required to serve a particular route can vary due to unpredictable factors such as traffic congestion or passenger demand.

In a traditional deterministic vehicle routing problem, the number of vehicles and the routes they take are predetermined based on a fixed set of orders and constraints. However, in a stochastic vehicle routing problem, the number of vehicles and the routes they take are determined in real-time based on the changing conditions and uncertain factors.

Bus bunching can create uncertainty in the number of vehicles needed to serve a particular route, as delays and congestion can cause buses to fall behind schedule and become bunched together. This can result in a sudden surge of passengers that may require additional buses to be dispatched in order to meet demand.

By incorporating bus bunching into a stochastic vehicle routing problem, transportation planners and operators can better allocate resources and adjust their operations in real-time to respond to changing conditions. This can help improve the efficiency of the transportation network, reduce delays and wait times for passengers, and ultimately improve the overall quality of service”.

ChatGPT: “…A stochastic vehicle scheduling problem with an uncertain number of vehicles could be used to address these challenges by developing a flexible and responsive scheduling framework that takes into account the uncertainties of bus bunching. For example, the scheduling algorithm could adjust the number and timing of buses based on real-time data on bus locations, passenger loads, and traffic conditions. The algorithm could also incorporate statistical models that predict the likelihood of bus bunching based on historical data and current conditions.

By using a stochastic vehicle scheduling approach, transportation operators could more effectively manage their fleets, reduce operating costs, and provide more reliable and consistent service for passengers. In addition, the approach could lead to more efficient use of resources, reduced environmental impacts, and improved overall quality of service”.

So, we get meaningful answers without the research issue being solved (as expected and without hesitation). Regarding the methodology, we may ask for machine learning and deep learning applications.

Q: “Do you know of machine learning and deep learning applications to mitigate bus bunching?”

Bing: “Yes, there are applications of machine learning and deep learning to mitigate bus bunching. For example, a multi-agent deep reinforcement learning framework has been proposed to develop dynamic holding control strategies to address the bus bunching problem. By carefully designing the reward function, the agents (i.e., buses) can effectively learn global holding strategies by pursuing headway equalization and avoiding heavy traffic [42]”.

ChatGPT: “Yes, machine learning and deep learning techniques have been applied to mitigate bus bunching in several studies.

One common approach is to use predictive models that can forecast the likelihood of bus bunching based on historical data and current conditions. These models can be developed using machine learning techniques such as regression analysis, time series analysis, or neural networks. By predicting when and where bus bunching is likely to occur, transportation operators can take proactive measures to mitigate the problem, such as adjusting schedules, dispatching additional buses, or changing routes.

Another approach is to use real-time data from sensors, GPS devices, and other sources to detect and respond to bus bunching in real-time. Deep learning techniques such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs) have been used to analyze real-time data streams and identify patterns and anomalies that may indicate bus bunching. By using these techniques, transportation operators can quickly detect and respond to bus bunching events, reducing delays and improving service reliability.

Several research studies have demonstrated the effectiveness of machine learning and deep learning approaches for mitigating bus bunching. For example, a study by Li et al. (2019) used a deep learning approach to predict bus bunching in real-time, achieving a prediction accuracy of over 90 percent. Another study by Tian et al. (2018) used machine learning models to predict bus bunching based on historical data, achieving a prediction accuracy of over 80 percent.

Overall, machine learning and deep learning techniques have the potential to significantly improve the efficiency and reliability of bus services by mitigating bus bunching and other operational challenges”.

The next question, i.e., asking for specific details of those questions, reveals the same experience as before. From ChatGPT, we obtain references that do not seem to exist. That is, Bing provides an existing reference and ChatGPT seems to provide more food for thought derived from different papers, but to some extent provides ideas from a different context that might successfully be applied here.

Q: “Is bus bunching found in connection with GTFS?”

Bing: “Yes, bus bunching can be found in connection with GTFS (General Transit Feed Specification). GTFS-RT (Real-Time) data can be used to quantify transit performance and group bus delays into two categories: stochastic and systematic”.

ChatGPT: “Yes, bus bunching can be analyzed in connection with GTFS (General Transit Feed Specification) data. GTFS is a data format used by many transit agencies to share information about their public transportation services. GTFS provides information on routes, schedules, stops, and other transit-related data that can be used for planning and analysis purposes.

Bus bunching can be identified and analyzed in GTFS data by looking at the actual arrival and departure times of buses at stops along a given route. By comparing these actual times with the scheduled times, it is possible to identify instances of bus bunching and to analyze the factors that contribute to it.

GTFS data can also be used to develop predictive models that can forecast the likelihood of bus bunching based on historical data and other factors such as weather, traffic conditions, and passenger demand. These predictive models can help transportation operators to take proactive measures to mitigate bus bunching, such as adjusting schedules or dispatching additional buses.

Several studies have used GTFS data to analyze bus bunching and to develop predictive models for forecasting it. For example, a study by Sun et al. (2019) used GTFS data from a bus route in Beijing, China, to develop a machine learning model for predicting bus bunching based on real-time data. Another study by Kumar et al. (2018) used GTFS data from a bus route in Bangalore, India, to analyze the causes of bus bunching and to develop strategies for mitigating it.

Overall, GTFS data can be a valuable resource for analyzing and addressing bus bunching, as well as for other planning and analysis tasks related to public transportation”.

The given references from Bing are as follows, while those given by ChatGPT could not be found: [58,59,60]. To clarify, while references Sun et al. (2019) and Kumar et al. (2018) are mentioned in the text, to the best of our knowledge, they do not exist in our context. This also holds for the specifications given in the detailed answer provided next. That is, asking for those references in detail reveals no information at Bing, as expected, and ideas and food for thought are revealed by ChatGPT for possible papers that could be written, but they most probably have not yet.

Q: “Can you please provide the details of Sun et al. (2019) and Kumar et al. (2018)?”

ChatGPT: “Sure, here are some details on the studies by Sun et al. (2019) and Kumar et al. (2018):

Sun, Z., Wang, F., Gao, Z., & Wang, H. (2019). Bus Bunching Prediction Based on Multitask Deep Learning with GTFS Data. IEEE Transactions on Intelligent Transportation Systems, 20(10), 3666–3675. In this study, the authors propose a multitask deep learning approach for predicting bus bunching based on GTFS data and real-time GPS data. The approach uses a combination of convolutional neural networks (CNNs) and long short-term memory (LSTM) networks to learn patterns in the data and to make predictions. The authors tested their model using GTFS data and GPS data from a bus route in Beijing, China, and achieved a prediction accuracy of over 90 percent.

Kumar, P., Roy, P. P., & Chakraborty, S. (2018). Bus Bunching and Propagation Analysis Using Smart Card Data in Bangalore City. Journal of Advanced Transportation, 2018. In this study, the authors analyze the causes of bus bunching and develop strategies for mitigating it using smart card data and GTFS data from a bus route in Bangalore, India. The authors use clustering and propagation analysis techniques to identify patterns in the data and to analyze the factors that contribute to bus bunching. The authors also propose several strategies for mitigating bus bunching, including improving schedule adherence, increasing the frequency of buses, and optimizing passenger boarding and alighting times. The authors show that these strategies can significantly reduce bus bunching and improve service reliability”.

Q: “How does GTFS-RT help with bus bunching?”

Bing: “GTFS-RT (General Transit Feed Specification Real-Time) is a feed specification that allows public transportation agencies to provide real-time updates about their fleet to application developers. It is an extension to GTFS, an open data format for public transportation schedules and associated geographic information”.

It should be noted that the given source from Bing at this place is (meaningful): https://developers.google.com/transit/gtfs-realtime?hl=de, accessed on 8 May 2023.

Bing (continued): “GTFS-RT can be used to quantify transit performance in a way that is fully generalizable to any bus transit network. Bus delays can be grouped into two categories: stochastic and systematic, which are quantified through several metrics. These metrics are then calculated at the segment-level across all routes [58].

This provides a network-wide screening tool that can be used to determine locations where reactive treatments (e.g., schedule padding) or proactive infrastructural changes (e.g., bus-only lanes, transit signal priority) may be effective at improving efficiency and reliability [58]. …”

A final issue relates to statistics on bus bunching events.

Q: “Does any public transport company provide statistics on bus bunching events?”

ChatGPT: “Yes, some public transport companies do provide statistics on bus bunching events. Many modern public transit systems use real-time data collection and analysis tools that allow them to track the location and status of their vehicles in real-time. These systems can also be used to identify instances of bus bunching and to collect data on the frequency and severity of these events.

Some public transit companies make this data publicly available on their websites or through other channels such as mobile apps or open data portals. For example, the Washington Metropolitan Area Transit Authority (WMATA) in Washington D.C. provides real-time bus arrival information and also publishes historical data on bus performance metrics, including on-time performance and bus bunching”.

We should note in passing that WMATA provides regular reports with the most recent one—https://www.wmata.com/about/records/upload/MetroPerformanceReport_FY23Q2_1Report_v20230207.pdf; last accessed on 10 May 2023—providing a nice quote on bus bunching: “In Q2, Metro focused on reducing bus bunching (when multiple buses on a route end up at the same stop at the same time) by placing managers at key stops along the routes to adjust spacing. Additionally, Metro made technological changes to no longer show schedule information of buses that are not sending real time information to public information feeds like transit apps, eliminating the “ghost bus” effect”. Actually, these managers may even be classified as mystery shoppers [36], as mentioned above.

ChatGPT (continued): “Similarly, the Toronto Transit Commission (TTC) in Toronto, Canada provides real-time information on bus locations and schedules through their mobile app, and also makes performance data available through their open data portal. This data includes information on bus bunching events, as well as other performance metrics such as service reliability and passenger volumes.

Other public transit companies may not make this data publicly available, but may use it internally to monitor and improve their services. In any case, the availability of real-time data and analysis tools has made it easier for public transit operators to identify and address instances of bus bunching, which can improve service reliability and passenger satisfaction”.

Bing: “Yes, some public transport companies do provide statistics on bus bunching events. For example, according to CTA (Chicago Transit Authority) performance metrics, only around 3 percent of bus trips experience bunching, which the agency defines as a gap of less than 60 s between buses at a stop. Of course, that percentage is greater during morning and evening commutes, as well as along the busiest routes”.

“Public transport companies may measure bus bunching in different ways. For example, the Chicago Transit Authority (CTA) defines bus bunching as a gap of less than 60 s between buses at a stop. Other companies may have different definitions and methods for measuring bus bunching. ”

The given references from Bing are available and valid: [39,61,62].

4. Bus Bridging in Light of Generative AI

Q: What is bus bridging?

ChatGPT: “Bus bridging, also known as shuttle service, is a transportation solution that is used when a section of a transportation network, such as a subway or train line, is temporarily out of service due to maintenance, construction, or other reasons. Bus bridging involves the use of buses to transport passengers between the affected section of the network and nearby stations or stops, allowing them to continue their journey. Bus bridging is typically used when the disruption to the transportation network is expected to last for a significant period of time and there is no alternative route available. It is also commonly used when the affected section of the network is in a densely populated area, as it allows passengers to continue their journey without having to walk long distances or find alternative means of transportation. During bus bridging, buses are usually deployed at regular intervals and may operate along a designated route or follow the same route as the affected section of the network. They may also be accompanied by staff to provide information to passengers and ensure a smooth transition between the bus and the regular transportation network. Bus bridging can be a cost-effective and efficient way to minimize disruption to the transportation network and ensure that passengers are able to reach their destinations as quickly and easily as possible”.

Bing: “Bus bridging is a service that provides substitute bus routes for metro passengers when there is a disruption in the metro system. Bus bridging aims to minimize the inconvenience and delay caused by the metro disruption, and to restore the normal level of service as soon as possible.

Bus bridging service design is a complex problem that involves determining the optimal bus routes, frequencies, and passenger assignments under various constraints and objectives. There are different methods and algorithms to solve this problem, such as genetic algorithms, column generation, and robust optimization”.

While Bing already releases sources for a search, ChatGPT has to be asked explicitly.

Q: “Do you have some important references on bus bridging?”

ChatGPT: “Certainly! Here are some references that provide more information on bus bridging:

“Transit Service Interruptions: Bus Bridge Implementation Guide” by the American Public Transportation Association (APTA). This guide provides detailed information on how to plan, implement, and manage bus bridging operations, including best practices, case studies, and templates for communication and coordination with stakeholders. It can be found on the APTA website.

“Bus Bridging Operations in Transit Systems” by the National Cooperative Highway Research Program (NCHRP). This report provides an overview of bus bridging operations and discusses their effectiveness, challenges, and costs. It also includes case studies and recommendations for improving bus bridging operations. It can be found on the Transportation Research Board website.

“Transit Street Design Guide” by the National Association of City Transportation Officials (NACTO). This guide includes a section on temporary transit operations, which covers bus bridging as well as other temporary transit solutions such as bus detours and temporary bus stops. It provides design guidelines for temporary bus facilities and examples of successful temporary transit operations. It can be found on the NACTO website”.

While the links to the provided associations APTA and NCHRP are important and meaningful, the specifically given references are, unfortunately, not found even by an extensive search. The Transit Street Design Guide is found; see [63]. While it covers topics such as bridge plates and street design, bus bridging is not really a topic of interest in that source.

Q: “Unfortunately I did not find the first two references and the third one seems to cover bridge plates but no bus bridging. Could you kindly clarify?”

ChatGPT: “I apologize for the confusion. Here are some updated references on bus bridging:

“Transit Service Interruptions: Bus Bridge Implementation Guide” by the American Public Transportation Association (APTA). This guide provides detailed information on how to plan, implement, and manage bus bridging operations, including best practices, case studies, and templates for communication and coordination with stakeholders. It can be found on the APTA website at https://www.apta.com/research-technical-resources/transit-practice/transit-service-interruptions-bus-bridge-implementation-guide/~.

“Bus Bridging Operations: Guidelines for Planning and Implementation” by the Transportation Association of Canada (TAC). This report provides guidance on how to plan and implement bus bridging operations, including considerations for route planning, service frequency, passenger information, and safety. It also includes case studies and examples of successful bus bridging operations. It can be found on the TAC website at https://www.tac-atc.ca/sites/default/files/Bus%20Bridging%20Operations%20Guidelines%20for%20Planning%20and%20Implementation_EN.pdf.

“Temporary Bus Stops and Shelters” by the National Association of City Transportation Officials (NACTO). This guide provides design guidance for temporary bus stops and shelters, which are often used during bus bridging operations. It includes recommendations for location, accessibility, and amenities such as seating, lighting, and real-time information displays. It can be found on the NACTO website at https://nacto.org/publication/urban-street-design-guide/street-design-elements/transit/temporary-bus-stops-and-shelters/.

I hope these updated references are helpful!”

Checking the links led to some “404 errors”, i.e., something went wrong. We could not find those web pages. Therefore, we tried to regenerate the response with different links having the same type of problems.

ChatGPT: “I apologize for any confusion. Here are some additional references specifically on bus bridging:

“Bus Bridge Manual” by the Los Angeles County Metropolitan Transportation Authority (Metro). This manual provides guidance on planning, implementing, and managing bus bridging operations, including information on staffing, equipment, communication, and safety. It can be found on the Metro website.

“Bus Bridge Operations: A Review of Current Practice” by the Transit Cooperative Research Program (TCRP). This report provides an overview of bus bridging operations and includes case studies from several transit agencies. It also discusses best practices, challenges, and recommendations for improving bus bridging operations. It can be found on the Transportation Research Board website.

“Temporary Bus Operations” by Transport for London (TfL). This guidance document provides information on temporary bus operations, including bus bridging, as well as temporary bus stops and diversions. It covers topics such as planning, communication, staffing, and safety. It can be found on the TfL website.

I hope these references are more helpful. Please let me know if you have any further questions or if there is anything else I can assist you with”.

Asking Bing explicitly for references leads to almost the same answer as before with the following correct references: [46,49,64]. As a next step, Bing proposes a question with the following answer. (Note that one might choose between different conversation styles, i.e., creative up to precise). The references given are existing ones provided by Bing and explicated by ourselves. This also holds for other places, where references are part of the answer.

Q: “What are some challenges or limitations of bus bridging?”

Bing: “Some of the challenges or limitations of bus bridging are:

Bus bridging may not be able to replace the train service adequately in congested city alignments, where the capacity of the roads and stations is limited [65].

Bus bridging may cause indirect delays to bus riders on the routes from which shuttle buses are dispatched [49].

Bus bridging may require terminators at both ends of the main cable and additional devices that may slow down the network”.

The given reference from Bing for this sentence is https://www.geeksforgeeks.org/advantages-and-disadvantages-of-bus-topology/, accessed on 8 May 2023. It may be classified as not belonging to public transport but still be correct due to an underspecified question from our side with possibly different meaning due to a bus also being a term in information technology).

Bing (continued): “Bus bridging may be affected by uncertainties in commuter demand and spare capacities of existing rail and bus lines [66].

Bus bridging may have to consider factors such as bus service capacity limitations, bus stop parking capacity, and the maximum bridging time limit of a single bus [66].

These are some of the factors that may influence the effectiveness and efficiency of bus bridging services in response to rail transit disruptions”.

A more generic question asks about the importance of bus bridging regarding sustainable passenger transportation. Moreover, we are interested in success stories.

Q: “How important is bus bridging regarding sustainable passenger transportation?”

ChatGPT: “Bus bridging can have a significant impact on the sustainability of passenger transportation. When buses bunch up or experience delays, it can lead to longer wait times and increased frustration for passengers, which can discourage people from using public transportation. This can lead to an increase in single-occupancy vehicle use, which contributes to traffic congestion and air pollution.

In addition, when buses are not operating efficiently, it can lead to increased fuel consumption and greenhouse gas emissions, which can contribute to climate change. By optimizing bus operations and reducing delays, transit agencies can improve the overall efficiency of their services, reduce emissions, and encourage more people to use public transportation.

Furthermore, efficient and reliable bus operations can also have positive economic impacts, by reducing the cost of operating transit services and making public transportation more attractive to businesses and residents. This can help to promote economic development and reduce the need for expensive new infrastructure projects.

Overall, bus bridging is an important issue for sustainable passenger transportation, and efforts to address it through the use of technology and data analytics can help to improve the efficiency, reliability, and attractiveness of public transportation services”.

Bing: “Bus bridging is an important strategy to mitigate the negative impacts of urban rail transit disruptions on passenger transportation. Bus bridging can help restore the connectivity of the disrupted area, reduce passenger delay and cost, improve passenger satisfaction and mode choice, and enhance the resilience and reliability of the public transport system. Bus bridging can also contribute to sustainable passenger transportation by reducing greenhouse gas emissions, energy consumption, and traffic congestion compared to private car use”.

[51,67]

Q: “Do you know of success stories where bus bridging worked really well?”

Bing: “Yes, there are success stories where bus bridging has worked well. For example, a study assessing the effectiveness of bus bridging on the operational resilience of the subway service in Toronto found that along uncongested subway segments where the available street capacity is relatively high, allowing large bus volumes to serve the corridor, bus bridging could be a good alternative. Another study developed a capacity-constrained bus bridging optimization framework and found that it showed strong potential to produce efficient shuttle response plans that reduced transit user delays by more than 50 % while ensuring minimum queue formation at disrupted stations and maximizing the utilization of shuttle buses”.

ChatGPT: “Bus bridging is typically used as a temporary measure to provide service continuity when there is a disruption to rail transit services such as a track closure or an equipment failure. While it may not be possible to point to specific success stories related to bus bridging, there have been many instances where it has been implemented effectively to provide a seamless transfer between rail and bus services during periods of disruption.

For example, during the 2012 London Olympics, the city’s transport authorities used a combination of bus bridging and other measures to ensure that visitors could get around the city despite a high volume of traffic and significant disruptions to transit services. The authorities deployed hundreds of buses to serve as shuttle services between rail stations and other key destinations, and also provided additional staff to help guide passengers and manage crowds.

Similarly, in the aftermath of Hurricane Sandy in 2012, the New York City Transit Authority implemented a bus bridging plan to provide service to customers affected by the closure of several subway lines. The plan involved the deployment of over 330 buses to provide service on shuttle routes between affected stations, and also included additional staff to help manage crowds and ensure smooth transfers between bus and subway services.

While bus bridging can be challenging to implement in some situations, it has proven to be an effective strategy for providing seamless service continuity during periods of rail transit disruption. By deploying additional buses and staff and providing clear communication to passengers, transit authorities can minimize the impact of disruptions on their customers and maintain the overall reliability of their transit systems”.

Both answers are meaningful, even if they have a different twist. Specific references from Bing are meaningfully provided with the above text in this order: [49,64,65].

Regarding the methodology applied, as mentioned above, bus bridging seems to be a problem that can be treated by classical operations research methods [5]. What about more recent AI methods?

Q: How can deep learning and machine learning be used to solve bus bridging problems?

ChatGPT: “Deep learning and machine learning can be used to solve bus bridging problems by providing intelligent algorithms that can analyze data and make predictions about traffic conditions, bus schedules, and other factors that can impact bus operations. Some possible ways in which deep learning and machine learning can be applied to bus bridging problems include:

Predictive maintenance: By analyzing data on bus performance and maintenance history, machine learning algorithms can predict when a bus is likely to experience mechanical problems and schedule maintenance proactively to avoid breakdowns.

Real-time route optimization: By analyzing traffic data in real-time, deep learning algorithms can optimize bus routes to minimize travel time and avoid congestion. This can help prevent buses from getting stuck in traffic and causing delays.

Demand forecasting: By analyzing historical data on passenger demand, machine learning algorithms can predict how many passengers are likely to use a particular bus route at a given time, allowing transit agencies to adjust schedules and allocate resources more efficiently.

Bus bunching detection: Machine learning algorithms can be used to detect patterns in bus arrivals and predict when buses are likely to bunch up. By doing so, transit agencies can adjust schedules and dispatch additional buses to prevent bunching and reduce delays.

Passenger flow optimization: By analyzing data on passenger behavior and flow patterns, deep learning algorithms can optimize bus stop locations and schedules to reduce passenger wait times and improve overall efficiency.

Overall, deep learning and machine learning can provide powerful tools for transit agencies to optimize bus operations, reduce delays, and improve the overall passenger experience”.

Bing: “I’m sorry, but I couldn’t find any information on the use of deep learning or machine learning to solve bus bridging problems. Would you like me to search for more information on this topic?”

Q: “How do you use NLP and/or ML to solve bus bridging problems?

Bing: “Bus bridging problems are related to the design and operation of temporary bus services that substitute for disrupted rail transit services. NLP and ML can be used to solve bus bridging problems by optimizing the route and frequency of bridging buses, minimizing passenger delay and cost, and incorporating uncertainty in bus travel time. For example, Luo et al. (2021) proposed a stochastic programming model for the design of bus bridging services plan within bus resource limit, considering uncertainties in commuter demand and spare capacities of existing rail and bus lines. NLP and ML can also help analyze the impact of bus bridging services on passenger satisfaction, mode choice, and travel behavior”.

[46,48,50,68]

5. Discussion

In this paper, we have undertaken a self-test, asking popular generative artificial intelligence tools ChatGPT and Bing some questions from the area of public transport. Based on an extensive survey paper [5], we felt qualified to judge the quality of the answers. One of the lessons learned is as follows. The advantage of Bing seems to be that sources are explicitly named with and without asking for them. On the other hand, ChatGPT provides sources that cannot be found and must have been invented by the system. With this in mind, we would like to re-specify and stress the focus of the paper and its motivation. In the current discussion of generative AI tools, there is and will be a lot of dynamics, and various science domains will have intensive considerations in due time. The underlying large language models of these tools will change. Currently, we can have a snapshot of what works well and what does not. We can see this from a method developer’s point of view and we can see it from an applicant’s point of view. Our intention here is to use the latter view. This bears the narrative style of the paper. Few review types possess prescribed and explicit methodologies and yet they deserve their place in academia even if they do not seem to be systematic; see, e.g., [69,70]. Nevertheless, our exposition allows a valuable reference point for those commissioning, conducting, supporting, or interpreting the use and the usability of chatbots in academia and especially in public transport research (and practice). Currently, the narrative style seems a possible method of choice.

We have used a narrative style to explore the ChatGPT system and also compare the answers to those from the Bing chatbot. Regarding transportation, this is new and innovative and especially the comparison among applying the two systems is meaningful. As mentioned above, one of the results of the comparison is that ChatGPT needs more care as there are possibly fake references, while Bing seems more serious on that. The specific exploration of bus bunching and bus bridging seems a reasonable focus. In a time where it is not clear how to handle the use of generative AI tools, it seems necessary to increase the understanding of these tools by empirical considerations as we did. OpenAI will possibly change the handling of references in due time and then it might be a good documentation to see which stages we went through with the application of these tools. The issue that ChatGPT “invented” references is something we need to document and learn. Furthermore, large language models will be trained and used in a different way in the future.

That is, on a specific scale, we looked at two important aspects within public transport, bus bunching and bus bridging. Beyond the above-mentioned survey paper, we even provided a somewhat meaningful update with very recent references and future research questions not yet answered, e.g., related to an uncertain number of vehicles in stochastic vehicle routing and/or vehicle scheduling problems and to which extent this may be motivated by bus bunching effects. We have shown that generative AI tools may be meaningful support tools if one needs additional ideas for possible research actions (e.g., through details of as yet unwritten papers invented by ChatGPT and through detailed information with existing references from Bing). Questions were based on our personal thinking and might have been biased. While questions were asked separately for bus bunching and bus bridging, the answers occasionally connected the two topics.

The last answers in Section 4 give the impression that bridging and bunching are to some extent connected. We did not check whether deleting our browser history would have influenced the answers. This might be explored in future research.

The writing within the answers was usually done in good English and mostly without language errors. We only discovered some mistakes regarding the use of hyphenation in compound adjectives without correcting them to keep the original writing.

6. Conclusions

In this paper, we have undertaken a self-test regarding the use of two popular generative artificial intelligence tools (i.e., ChatGPT and Bing) within the area of public transport. We have taken a narrative style and we intended that the reader “observes” us without the need to copy. As an observer we rather report on disruptive issues (or findings; the dawn of the usage of generative AI may be called disruptive) than developing them. The latter is carried out in other papers with a different focus. Having a careful use of quotes from Bing and ChatGPT, quite some attention is paid to chatbot conversations. A significant part of it is presented in the paper itself.

To conclude, we see that generative AI tools have become widely applicable in an intuitive and easy-to-use way. Without any specific effort, the user is able to apply tools whose underlying methodology seems untransparent. Therefore, it is important to critically use those tools and to avoid misuse. Future research on a general scale needs to investigate the ethical issues behind using these tools and possibly giving impressions that do not hold common standards in many places around the globe.

An overall conclusion that can be drawn from our specific chats is that the tools used can enhance research productivity and many provided ideas are given that may be the motivation for further research. If researchers thought about a handwaving policy—You don’t tell mushroom spots—these tools tell them. For young researchers, if they or their supervisors do not have enough ideas, here they may be acquainted with many of them. Transparency seems more an issue with ChatGPT rather than with Bing. Future research might check for more standardized questions for testing generative AI tools.

Data and technology seem to be available to enhance bus bridging and bus bunching decisions. The inclusion of these data into current databases available online (see, e.g., [58,71]) and a comprehensive connection to generative AI tools to further improve decision-making abilities of individual users of public transport seems a most suitable area for future research, too.