DO’s and DON’Ts of AI and LLMs for Hotels

Introduction

The convergence of Artificial Intelligence (AI) with Large Language Models (LLMs) is spearheading transformative changes in the hotels, travel, and tourism sectors.

Imagine a world where AI-powered virtual assistants handle hotel bookings and curate personalized experiences for guests. Or where LLMs seamlessly bridge linguistic divides, offering instant translations for travelers. AI's prowess in forecasting flight patterns and streamlining itineraries dovetails with LLMs' capability to help tourism agencies craft content that resonates across cultures. Together, they promise not just enhanced experiences for travelers but also streamlined operations for businesses in the travel sector.

While the guidelines presented here aren’t exhaustive, they are instrumental in striving towards excellence. By adopting best practices, stakeholders can ensure that guests embark on memorable journeys, all made possible through the strategic implementation of technology.

DO's

Customize Guest Journeys: Utilize AI and LLMs for bespoke guest experiences, from room ambiance to local sightseeing suggestions.
Maximize Reception Efficiency: Deploy AI-chatbots for swift check-ins, bookings, and frequent guest queries.
Stay Ahead with Learning Models: Periodically update AI systems with fresh guest data for improved service offerings.
Introduce Multimodal Guest Services: Elevate hotel concierge services by integrating text, voice, and visual AI communications.
Master Demand Forecasting: Harness predictive analytics for efficient room inventory management and dynamic pricing.
Guarantee Ethical AI Use: Assure guests of unbiased AI operations prioritising their privacy.
Pioneer Smart Room Innovations: Differentiate with rooms equipped with voice-activated amenities and AI-driven personalized entertainment.
Embrace Global Hospitality: Offer instant language translations, catering seamlessly to international guests.
Showcase Hotel Virtually: Allow guests to explore rooms, facilities, and nearby attractions using VR & AR.
Stay Tech-Updated: Regularly refine your AI-based offerings in sync with technological advancements.
Prioritize Guest Insights: Use real-time feedback to enhance AI-driven hotel services.
Empower Hotel Staff: Offer AI tools that simplify operations and boost the service level.
Diversify Guest Data Analysis: Glean insights from varied data sources to craft unparalleled guest experiences.
Be Open About AI Usage: Clearly inform guests about how AI aids in elevating their stay.
Maintain Operational Excellence: Ensure thorough real-world testing of AI systems before full-scale adoption.
Engage AI Specialists: Regularly consult AI experts to stay ahead of industry benchmarks.
Promote Eco-Friendly Tech: Opt for sustainable AI practices, enhancing your hotel's green credentials.
Ensure Universal Access: Make AI-driven features user-friendly for guests of all abilities.
Facilitate Continuous AI Improvements: Set up mechanisms for AI systems to evolve based on guest interactions.
Champion Guest Safety: Especially post-pandemic, use AI to monitor health protocols, ensuring a secure stay.

DONT's

Over-Automate Guest Interactions: Retain the irreplaceable warmth of human touchpoints in hospitality.
Overlook Feedback Channels: Regularly solicit and act on guest and staff input for AI initiatives.
Risk Guest Data: Prioritize rigorous data protection measures, safeguarding guest information.
Adopt Generic AI Solutions: Every hotel is unique. Tailor AI applications to your hotel's character and guest demographics.
Skip Staff Training: Invest in comprehensive training sessions, enabling staff to maximize AI tool benefits.
Misuse Advanced Tech: Handle surveillance AI tools, like facial recognition, judiciously.
Depend Wholly on AI Pricing: Balance AI insights with market trends and on-ground realities.
Intimidate with Tech: Maintain a fine balance between innovation and simplicity in guest-facing AI systems.
Exclude Human Supervision: Essential AI-driven functions, especially security, should have human oversight.
Narrow Down AI Communication: Offer multi-channel AI guest services spanning text, voice, and visuals.
Oversell Tech Capabilities: Be transparent about the scope and limits of your AI services.
Isolate AI Systems: Ensure AI features align and integrate well with other hotel operational tools.
Sideline Traditional Hospitality: While AI is transformative, the essence of hospitality lies in human connections.
Ignore Cultural Varieties: Ensure AI respects and understands the nuances of diverse guest backgrounds.
Operate Without Contingencies: AI is a tool, not a crutch. Always have a backup plan.
Install AI Without Purpose: Every AI integration should align with your hotel's vision and guest service goals.
Violate Regional AI Laws: Stay updated on local regulations concerning AI, data protection, and guest privacy.
Compromise on Digital Design: Ensure all AI interfaces, apps, or tools are intuitively designed.
Rest on Current Achievements: The world of AI is ever-evolving. Keep refining, testing, and innovating.
Limit Guest Autonomy: Offer guests options to customize or even opt-out of AI interactions during their stay.

Glossary

AGD detection: Is the process of identifying and mitigating adversarial examples in AI and large language models (LLMs). Adversarial examples are inputs that are designed to fool AI models into making incorrect predictions. They can be created by intentionally adding noise or perturbations to the input data, or by carefully crafting the input data to exploit weaknesses in the AI model.
AI Crawlers: A type of software program that is used to collect and process information from the internet automatically. AI crawlers are typically used by large language models (LLMs) to train themselves on a massive dataset of text and code.
AI Ethics: This includes considering the potential risks of AI, such as bias, discrimination, and privacy violations, as well as the potential benefits, such as improved healthcare, education, and transportation. It also involves ensuring that AI systems are transparent and accountable and that they are used in a way that respects human rights and values.
Anomalous behavior: Are behaviors that are unexpected or out of the ordinary. Various factors, such as Bugs or errors in the AI model, can cause them. Insufficient training data. Adversarial examples. Biases in the AI model.
Analyze the Output: Understand where the response met expectations and where it deviated.
Anthropomorphizing: Refers to ascribing human-like emotions, intentions, or consciousness to computational algorithms that generate human-like text.
Artifact: Refers to a tangible or visible product resulting from the application of AI methodologies, such as trained models, software tools, or datasets.
Artificial Intelligence (AI): The capability of a machine to imitate intelligent human behavior. In the hotel, travel, and tourism sectors, AI can be used for personalizing guest experiences, optimizing pricing strategies, forecasting demand, and improving operational efficiencies.
Biased outcome: Refers to a decision or result produced by a machine learning model that unfairly favors or discriminates against certain groups or categories, often due to underlying biases in the training data or algorithms.
BYOK: Bring Your Own Key - as in OpenAI Key - a unique identifier that allows authenticated access to the API, ensuring that only authorized users can make requests to the model.
Chatbots: AI-driven programs designed to interact with users in a natural, conversational manner. Hotels and travel companies can use chatbots for bookings, answering frequent queries, and providing 24/7 customer support.
Classifiers: Refers to algorithms or neural network architectures designed to categorize or label input data based on learned patterns from training data; for large language models, this might mean determining the topic, sentiment, or intent of a given text input.
Compute on the edge: Refers to performing AI processing and continuous learning directly on local devices, like smartphones or IoT devices, rather than in centralized data centers.
Conditional Prompts: By setting conditions, you can guide the AI’s response more effectively. For instance, “If X, then provide information on Y.”
Cost Implications: More complex prompts might require more tokens, increasing computational costs.
Customer Insights Engine: AI-driven tools that analyze customer data to glean insights into preferences, behaviors, and patterns. This helps businesses tailor their offerings more effectively.
Data Analytics: The process of examining data sets to draw conclusions. In tourism, this can be used for understanding traveler behavior, optimizing marketing strategies, or assessing the performance of various services.
Data crawl: Refers to the automated process of systematically searching and collecting data from various sources, often the web, to feed and refine machine learning models.
Data point: Refers to a single observation or entry in a dataset. Data points are the basic units that models use to learn patterns, make predictions, or derive insights.
Datasets: A structured collection of data, often comprising text, images, audio, or video, used to train, validate, or test machine learning models.
Data poisoning: refers to the deliberate introduction of inaccurate or malicious data into a model's training set to compromise its integrity or functionality.
Deep fakes: A type of artificial intelligence (AI) that uses machine learning to create or alter images or videos in a way that makes it look like someone is saying or doing something they never did. This can be done by using a technique called “d eep learning,” which allows computers to learn from data and identify patterns.
Deep learning: A subset of machine learning that uses multi-layered neural networks to model and process complex data representations, enabling AI systems to learn and adapt continuously from vast amounts of data.
Dynamic Pricing: An AI-driven pricing strategy where prices are adjusted in real-time based on various factors like demand, competitor prices, and other market dynamics. Widely used in airline ticketing and hotel room pricing.
Empathy: Refers to the simulation or mimicking of human-like understanding and responsiveness to emotions and sentiments in user input, although the AI itself does not genuinely experience or possess emotions.
Facial Recognition: AI technology that identifies people based on their facial features. Some hotels are exploring its use for seamless check-ins or to enhance security.
Generative AI [GenAI]: Refers to algorithms and models that can create new content, such as images, music, or text, that was not previously in their training data.
GPUs: (Graphics Processing Units) are specialized hardware components that accelerate the computations required for training and deploying large-scale AI and neural network models.
GPT-4: A state-of-the-art language model developed by OpenAI, designed to understand and generate human-like text based on its vast training data and deep learning techniques.
Hallucinate: Refers to situations where the model produces outputs that aren't based on the data it was trained on or are not consistent with the real-world observations. This is especially common in tasks involving generative models or when models make predictions about unfamiliar or rarely seen data.
Information extraction: Is the process of automatically identifying and extracting structured information, such as entities, relationships, and events, from unstructured data sources, like text documents.
Language predictor: Refers to an algorithm or model that forecasts the next word or sequence of words in a given text based on learned patterns from vast amounts of training data.
Large Language Models (LLMs): A subset of AI that revolves around processing and generating human-like text based on massive datasets. In the tourism industry, LLMs can be employed for customer support, personalized travel recommendations, and content generation.
Matrix multiplication: Is a fundamental mathematical operation used in neural networks to transform and propagate data through layers, facilitating learning and predictions.
Max Tokens: Setting a limit ensures the response doesn’t exceed a specific length, crucial for applications with space constraints.
Modalities: Refers to the different types or channels of data input and output, such as text, images, sound, or video, which an AI system can process or generate.
Multimodal conversational AI: Integrates multiple input and output modalities, such as text, voice, and visuals, to facilitate richer and more dynamic interactions between users and AI systems.
Natural Language Processing (NLP): A branch of AI that focuses on the interaction between computers and humans. In the travel sector, NLP can be used to understand customer reviews, feedback, and to improve voice-assisted search.
Neural networks: Are computational models inspired by the human brain's structure, used in AI to process and learn from large amounts of data.
Open source: Refers to AI algorithms, models, or tools that are publicly available, allowing anyone to use, modify, and distribute them without proprietary restrictions.
Operational Automation: Using AI to streamline and automate daily operations. This can range from automated check-ins in hotels to AI-driven systems that manage inventory in travel agencies.
Overfitting: Overly specific prompts might yield perfect responses in one scenario but fail in others. Striking a balance is essential.
Parameter Tuning: Beyond the prompt text, adjusting model parameters like temperature and token limits can significantly affect outputs.
Parsers: Refer to tools or components that analyze and interpret the structure of sentences, breaking down input text into grammatical components to better understand its meaning and structure.
Personalization Engines: AI systems that analyze user data to provide customized recommendations and experiences. For instance, suggesting travel itineraries based on past trips or tailoring room preferences in a hotel.
Predictive Analytics: Using AI to analyze historical data to forecast future events. This can help hotels and tourism companies in predicting peak demand periods, setting dynamic pricing, or identifying potential maintenance issues.
Prompt Chaining: This involves feeding the model’s output back as a new prompt, allowing for extended and more nuanced responses.
Prompt engineering: Involves crafting and refining input prompts to guide AI language models, like GPT-4, towards producing desired and accurate outputs.
Prompt injection attacks: These occur when a user creates an input aimed at causing the AI model to behave in an unintended way — such as generating offensive content or disclosing confidential information.
RAG (Retrieval-Augmented Generation): Refers to a methodology that combines information retrieval from large databases with sequence generation, enabling the model to pull and incorporate external knowledge to produce more informed and detailed outputs.
Reasoning engine: A system that simulates human-like logic and inference capabilities to derive conclusions or make decisions based on given information or knowledge.
Recommendation Systems: Algorithms that provide suggestions to users based on their preferences and past behaviors. For example, suggesting destinations, hotels, or attractions based on a traveler's interests.
Repeat: Feed the adjusted prompt and analyze the new output. Continue this process until the desired result is achieved.
Retrieval augmentation: Refers to enhancing machine-generated responses by accessing and integrating information from external knowledge sources or databases to provide more accurate and contextually relevant answers.
Root cause analysis: Refers to the systematic approach of identifying the fundamental reasons or underlying issues causing unexpected behaviors, errors, or biases in the model's outputs or decisions.
Semantic search: Refers to search processes that understand and consider the intent and contextual meaning of user queries, rather than just keyword matching, to provide more relevant results.
Sentiment analysis: Is the process of determining and extracting the emotional tone or attitude conveyed in a piece of text.
Smart Rooms: Hotel rooms equipped with AI-powered devices and systems to enhance guest experiences, such as voice-controlled lights, personalized entertainment options, and automated room service.
Temperature Settings: This determines the randomness of the AI’s response. A higher value makes outputs more random, while a lower value makes it more deterministic.
Tokens: Every word or character the model processes is considered a token. Managing tokens is essential for optimal performance and cost-efficiency.
Trustworthy: Refers to the reliability, safety, and ethical considerations of artificial intelligence systems, ensuring that they operate as intended without bias or harm.
Tweak the Prompt: Adjust wording, add clarifications, or set conditions.
Unpredictability: Despite advancements, AI responses can sometimes be unpredictable. Continuous testing is crucial.
Vectorization: Refers to the process of converting data, especially text or images, into numerical vector representations for efficient processing and analysis by machine learning models.
Virtual Assistants: AI-driven platforms that can assist users in various tasks. In the travel industry, these can help in planning trips, providing real-time travel updates, or answering common travel-related queries.
Virtual Reality (VR) & Augmented Reality (AR): Technologies that either create a simulated environment or enhance real-world settings with digital elements. Hotels and tourist spots can use VR and AR for virtual tours, enhancing guest experiences, or providing interactive information.
Word vectors: Are numerical representations of words in a high-dimensional space, capturing semantic meaning and relationships, enabling machine learning models to process and understand text.

Top 40+ Generative AI Tools (as of September 2023 - Arham Islam)

Adobe Enhance - This AI tool removes background noise from audio recordings.
AlphaCode - AlphaCode has been developed by DeepMind and is capable of writing computer programs at a competitive level.
Adobe Firefly - Firefly is an image generation and editing tool known for its prompt-to-image output accuracy. It encompasses a wide range of image modification features, including content type, color, tone, lighting, and composition tools.
Bard - Bard is a chatbot developed by Google, which is seen as Google’s counterpart to OpenAI’s ChatGPT.
Bardeen - Bardeen is an automation platform that replaces the repetitive tasks of users.
Bing AI - Bing AI is powered by the GPT-4 model of OpenAI and can traverse the web to provide accurate answers. It also has the ability to generate images from user prompts.
Boomy - Boomy allows users to create their own original songs within seconds.
Character.ai - Character AI allows users to create characters and talk to them.
ChatGPT – GPT-4 -GPT-4 is the latest LLM of OpenAI, which is more inventive, accurate, and safer than its predecessors. It also has multimodal capabilities, i.e., it is also able to process images, PDFs, CSVs, etc. With the introduction of the Code Interpreter, GPT-4 can now run its own code to avoid hallucinations and provide accurate answers.
ChatPDF - This tool allows users to upload any PDF file and chat with it.
Chatflash - Chatflash is a tool that allows users to create content through a chat option.
CLIP Interrogator - This tool analyses an image and identifies the prompts one might need to input to generate it.
Cohere Generate - Cohere Generate leverages the potential of AI to enhance business operations. It offers personalized content for emails, landing pages, product descriptions, and various other requirements.
Copy.ai - Copy.ai allows users to generate high-quality marketing copies in seconds, whether for a blog post, an email, or a social media update.
Descript - Descript is a versatile video editing tool that allows users to create, record, transcribe, edit, collaborate on, and easily share their videos and podcasts.
DALL-E 2 - DALL-E 2 is a text-to-image generation tool developed by OpenAI that creates original images based on the user’s prompt. It has been designed to reject inappropriate user requests.
Designs.ai - This tool allows users to generate logos, videos, and banners within a few minutes.
ElevenLabs - ElevenLabs have developed an advanced text-to-speech and voice cloning tool.
Engage AI - Engage AI makes prospect engagement better by improving and adding context to comments. It helps break the ice and builds stronger relationships with potential clients.
Ernie - Baidu has been embedding Ernie, which resembles OpenAI's ChatGPT, into its search engine and other products, allowing many of them to gain market share while waiting for Chinese regulators' approval.
Extrapolate -Extrapolate takes in user images as input and shows how they will age.
Gamma - Gamma generates decks based on the user prompts.
GitHub Copilot - AI code completion tool that analyzes code and provides instant feedback and relevant code suggestions.
GPTZero - GPTZero is an AI plagiarism checker tool.
Kickresume - AI resume builder that simplifies the process of writing and creating effective resumes.
Memecam - This tool gives the user’s image meme-like captions.
Microsoft Designer - This tool creates posters, illustrations, and artwork based on the user’s input.
Midjourney - An mage generation tool that creates images according to the user’s prompts.
Murf.ai - Murf is a text-to-speech tool that allows users to generate studio-quality voiceovers in minutes.
Palette - Palette colorizes the black-and-white images within seconds.
Perplexity AI - Perplexity AI is an all-in-one search engine utilizing GPT-4, enabling intelligent exploration across diverse databases with ease.
Pi - Pi is a chatbot that talks almost like a therapist.
Picsart AI Writer - This tool has features like an Ad copy generator, LinkedIn headlines generator, rephraser, summarizer, and more.
Poe - Poe is a platform that provides access to the major chatbots like GPT-4, Claude, Llama, etc., in one place.
Prompt Vine - Prompt Vine is like a virtual library for ChatGPT prompts.
Quizify -This tool allows users to create quizzes on different topics.
Remove.bg - This tool removes the background of any image.
Replika -Replika is an AI-driven chatbot designed to be a virtual companion. It has the capability to form deep connections with users and even allows them to consider it as a significant other. The platform offers features like video calls to enhance the interactive experience, and Replika keeps a journal where it records its emotions and thoughts.
RoomGPT - This tool allows users to leverage AI to redesign their rooms within seconds.
Scribble Diffusion - This tool converts hand sketches into professional images.
Socratic - Socratic allows users to upload a photo of their homework, and it solves the problem almost immediately.
Soundraw - Soundraw is a music generator that allows users to create their own unique and royalty-free music.
Synthesia - Synthesia is a video generation tool that converts texts into high-quality videos using AI avatars and voiceovers.
Text to Pokemon - This tool converts the user’s input into a Pokemon.
Tome - Tome is a storytelling tool that drafts text and generates images.
Tutor AI - Tutor creates full-scale courses along with modules based on the user’s topic.
Type Studio -Type Studio is a video editing tool that enables users to edit their videos by making changes directly to the transcribed text.
Voicemod - Voicemod allows users to create full-fledged songs just by entering text.

This paper is part of our unique DO’s and DON’Ts series, which we are making in collaboration with SUBJECT MATTER EXPERTS - GROWTH ADVISORS INTERNATIONAL NETWORK - GAIN

Terence Ronson
Managing Director
Pertlink Limited