Potential Applications Of ChatGPT In Facilitating Vehicle Inspections For Accident Management | Inspektlabs
Integrating AI-based damage and repair assessment with ChatGPT can be a game changer. While AI can detect damages and generate repair estimation, we need customer interaction/inputs for complete automation.
Outline
In this blog, we talk about the capabilities of ChatGPT and the use case in vehicle inspection automation. We begin with an introduction to ChatGPT and a brief overview of its general applications. Then, we will discuss Inspektlabs' experiments using ChatGPT, and subsequently, we will discuss the potential workflow for its application. Lastly, we will go over the limitations of ChatGPT, followed by a conclusion.
General Applications of ChatGPT
Inspektlabs' Experiments with ChatGPT
Potential Workflow for Using ChatGPT
Introduction
There's been a lot of excitement about ChatGPT over the last few months. Much of the discourse revolves around the potential capabilities of this new technology and how it can impact various industries. The automotive and motor insurance industries are also grasping the opportunity to experiment with integrating ChatGPT into their workflows. Integrating AI-based damage and repair assessment with ChatGPT can be a game changer. While AI can detect damages and generate repair estimation, we need customer interaction/inputs for complete automation. Let us explore what role ChatGPT can play in facilitating vehicle inspections for accident management.
What is ChatGPT?
Chat Generative Pre-Trained Transformer, famous as ChatGPT, is a new artificial intelligence-powered chatbot that can generate intelligent dialogue. OpenAI, located in San Francisco, launched ChatGPT on 30 November 2022, a machine-learning chatbot that can answer queries in a conversational tone. [Source: Forbes]
The system employs a Generative Pre-trained Transformer (GPT) to create a textual response based on user input. It is helpful in various domains, such as producing essays, emails, and poems, responding to customer queries, and much more. While its primary use is to converse with humans, it may also be helpful for many other purposes. For example, it may propose blog post subjects and keywords and write code for you. [Source: NY Times] Depending on the model training, it might provide both correct and false results.
General Applications of ChatGPT
OpenAI has created the artificial intelligence (AI) model ChatGPT. It is meant to respond to human queries in natural language, and these replies may be helpful for various purposes. ChatGPT's uses include customer service, data analysis, and language translation. Businesses may use the model to categorize text or produce personalized suggestions. Companies may save time and money using this technology on human customer service representatives. [Source: McKinsey]
ChatGPT can analyze vast quantities of text and transform unstructured material into a more organized framework. This is particularly handy for firms that often need to generate vast amounts of data. ChatGPT has furthermore been used for patent application preparation and patent portfolio analysis. The model may be trained on a particular subject or area and deliver correct information.
Other ChatGPT uses include the development of language translation services and the enhancement of data analytics. In addition, developers may leverage this technology to create code more effectively. Now that we have covered the general applications of ChatGPT let us go over Inspektlabs' experiments with the chatbot.
Inspektlabs’ Experiments with ChatGPT
We ran multiple experiments with ChatGPT and used various prompts to test its capabilities. For our discourse, we will cover three of the experiment results.
Experiment #1:
Upon accident (or First Notice of Loss), ChatGPT can guide the user to take the correct steps, including prompting the customer to capture photos/videos of the car.
Experiment's Result in Plain Text:
"Client: I have had a car accident. What should I do?
ChatGPT: If you have been in a car accident, it is important to stay calm and assess the situation. If anyone is injured, call 911 immediately. If the accident is minor and there are no injuries, move the cars to a safe location, if possible, and exchange contact and insurance information with the other driver. Take photos of the damage to both vehicles and the surrounding area, if possible Report the accident to your insurance company as soon as possible. It is also important to seek medical attention, even if you feel fine, as some injuries may not appear until later."
Experiment #2:
Based on the specific damage type, ChatGPT can explain to the customer whether a repair vs. replacement decision on the windshield needs to be taken. This recommendation depends on the crack's type, size, and location.
Experiment's Result in Plain Text:
"Client: I have a crack on my car windshield. What should I do?
ChatGPT: If you have a crack on your car windshield, it's important to address it as soon as possible. A small crack can quickly turn into a larger one, and a damaged windshield can impair your visibility and make your car less safe to drive.
Depending on the size and location of the crack you may be able to have it repaired. Many auto glass repair shops can fix small chips and cracks in the windshield, and in some cases, the repair may be covered by your insurance. If the crack is too large or in a location that makes it difficult to repair, you may need to replace the entire windshield. This can also be covered by your insurance, but you should check your policy to be sure. It's important to note that if the crack is in the driver's line of sight, it might prevent you from passing the vehicle inspection."
Experiment #3:
Based on the dent type, ChatGPT can guide the person to the correct type of repair, such as PDR (Paintless Dent Removal), traditional body repair, or suction cup method.
Experiment's Result in Plain Text:
"Client: There is a dent on the door of my car. What should I do?
ChatGPT: If there is a dent on the door of your car, you have a few options for repair:
1. Paintless Dent Repair: This method involves massaging the dent out from the inside of the panel using specialized tools. This is a quick and efficient way to remove dents and is often used for small to medium-sized dents.
2. Traditional Body Repair: This method involves removing the damaged panel and replacing it with a new one or patching it. This method is used for larger dents or dents that are not suitable for paintless dent repair.
3. Suction Cup Method: This is a DIY method that you can use at home. A suction cup is placed on the dent and pulled to remove the dent. This method is only suitable for small dents, and it may not work if the paint is cracked or damaged.
It's important to note that, if the paint is cracked or damaged, a traditional body repair will be the best option. Also, if the dent is on a structural part of the car, it might affect the car's integrity, so it's better to have it checked by a professional."
Potential Workflow for Using ChatGPT
Following is the potential workflow for using ChatGPT for vehicle inspections for accident management:
1. The customer has an accident. The customer contacts the insurer/car leasing player/car rental player backed by ChatGPT and Inspektlabs through customer care, SMS, email, or WhatsApp.
2. ChatGPT answers the customer's queries on what they should do next. ChatGPT asks the customer to capture photos/videos of the car using the Inspektlabs web app.
3. The customer captures photos/videos with the Inspektlabs web app. Inspektlabs API analyses the photos/videos to detect damages, estimate the $ value of repair, and detect potential fraud. This outcome is communicated to ChatGPT using API integration.
4. Using Inspektlabs API assessment, ChatGPT could take several paths. The insurer would define these paths. Let us consider a few examples:
a. Based on the damage type detected by Inspektlabs, the customer could be guided to the correct location for repair. For example, windshield-only damage should be directed to the nearest glass repair network, and very minor damage (e.g., front bumper minor dent) can be directed to the nearest PDR (paintless dent repair) shop.
b. Based on the repair estimate, the customer could be recommended whether or not they should make an insurance claim or avail of a no-claim bonus. In many cases, the customer may be offered on-account or instant settlement.
c. In case a potential fraud is detected, ChatGPT may ask the user to get a human inspection rather than an AI-based inspection to verify the integrity of the claims.
5. Once the car goes under repair, ChatGPT integration can update the customer and answer customer queries. Some of these queries could be, for example, "By when would my car be repaired?" and "Why has the repair cost increased?"
Limitations of ChatGPT
ChatGPT is a chatbot developed to deliver information and respond to inquiries. This artificial intelligence application may be used for customer service, answering queries, software development, and other domains. While it does appear to be effective in specific contexts, there are certain limitations:
- Incorrect or Non-sensical Answers – ChatGPT may produce plausible yet non-sensical or false answers, owing to the limitations of how the model was trained. Training the ChatGPT model to be more cautious with its responses may lead to overcorrection, and the model may decline to answer even the questions it can answer correctly. [Source: OpenAI] Receiving non-sensical or false answers while conducting a vehicle inspection will create unnecessary confusion.
- Does Not Seek Clarifications – ChatGPT currently tends to guess the user's intent instead of seeking clarifications from the user. [Source: OpenAI] This limitation can be quite problematic. ChatGPT can guide the owner through a car photo inspection for insurance. However, if it does not seek clarifications regarding the specific details of the accident, it will generate vague or inapplicable responses.
- Does Not Offer Unique Insights – ChatGPT is an excellent tool for enhancing customer interactions but cannot replace human operators. In reality, the AI model lacks the knowledge and emotions required to tackle complex issues. ChatGPT can respond to simple customer care inquiries, but if the question needs more in-depth expertise, it will be unable to assist. [Source: ResearchGate]
- Excessively Verbose – ChatGPT can be overly verbose and generates comprehensive and lengthy responses even when it is not required. Due to this limitation, ChatGPT does not provide the exact answer in certain situations. [Source: OpenAI] The user will likely expect concise and accurate information after an accident. Burdening the user with lengthy responses can lead to them missing key details or instructions.
- It Can Be Too Inhumane and Detached – A significant critique of ChatGPT is that its responses are far too logical. After an accident, the car owner may want or expect a compassionate tone rather than a highly logical one. It may be off-putting in such sensitive contexts.
- Not Connected To The Internet – Since ChatGPT is not linked to the internet, it may sometimes provide wrong responses. It has limited awareness of the world and events that will occur beyond 2021 and may occasionally offer damaging instructions or prejudiced material. [Source: OpenAI] So, the part prices or estimations that ChatGPT shows may be obsolete and no longer relevant.
- Legal Issues – ChatGPT suffers legal and regulatory difficulties. Several educational institutions have prohibited its usage, for instance. Others have cautioned against cybercriminal behavior. Some scholars have even been concerned about copyright violations. This limitation can be highly problematic for auto insurers and body shops because ChatGPT may guide the user based on a competitor's information.
To illustrate this particular limitation, we ran the results of our experiment through a plagiarism checker CopyScape. And, the responses have been plagiarised by ChatGPT.
The plagiarised text arose from the results of Experiment 1. This type of blatant plagiarism may lead to legal disputes.
Way Forward
While ChatGPT offers dependable replies, it is not flawless. It may provide inaccurate responses, sometimes with information that seems convincing but is false. The answers will rely on the model's current knowledge of the subject. In addition, it often generates meaningless texts. Even if it can do very well, it cannot replace a human agent. It has been regarded as a valuable tool for a variety of application scenarios despite these limitations.
ChatGPT has the reach to help improve vehicle inspections. ChatGPT can guide the owner through a car photo inspection for insurance. ChatGPT can also help users understand the meaning of specific terms in vehicle condition reports. Digital vehicle inspections assisted by ChatGPT can help enable a comprehensive framework for vehicle inspections for accident management.
Conclusion
ChatGPT is still in training and requires human oversight, according to Open AI. The creators of ChatGPT are seemingly aware of these limitations and are actively working on them to improve their capabilities. Humans will still be around even if ChatGPT is integrated into most aspects of the vehicle inspection workflow. ChatGPT or Inspektlabs would not fully automate the process, but AI will improve workflow efficiency over time.
