Technology Partners

Our Team

Our Awards

Social Impact

Newsroom

Locations

Careers

About Us

Contact

Do Not Sell my Personal Information

Cookie Preferences

Culture Value Chain

iLabs Center for Innovation

Industries

Insights

Investor Relations

TELUS International is a subsidiary of: 

Privacy Shield Framework

Privacy Policy

Solutions

Subscribe to Newsletter

Website User Terms

logo-x

Icon facebook

Icon youtube

Icon linkedin

Technology

Financial Services & Fintech

Automotive

Healthcare

Communications & Media

Games

Retail & Ecommerce

Travel & Hospitality

infobrief

Guide

Resource Type Article

Article

case-studies thumb

Case Study

Preventing fraud for a global tech company in the travel and hospitality space image with fern in background

TI resource case 900x510

Video

whitepapers thumb

Research & Reports

checklists thumb

Checklist

brochures thumb

Brochure

ebooks thumb

E-book

Podcast Type Image

Podcast

The recent proliferation of generative AI (GenAI) models has sparked a lot of interest and excitement. These advanced large language models (LLMs) can do many extraordinary things — from crafting email responses to recommending a fitness routine to generating computer code; the list is seemingly endless.

However, these models aren't infallible. In fact, their propensity to sometimes hallucinate — a phenomenon where they provide responses that incorporate fabricated data that appears authentic — could significantly impact user trust. A recent TELUS International survey found that 61% of respondents were concerned about generative AI increasing the spread of inaccurate information online. 

Understanding what hallucinations are, why they happen and how to prevent them is critical to maintain users' trust in GenAI applications.

Introduction to generative AI hallucinations

A hallucination describes a model output that is either nonsensical or outright false. An example is asking a generative AI application for five examples of bicycle models that will fit in the back of your specific make of sport utility vehicle. If only three models exist, the GenAI application may still provide five — two of which are entirely fabricated. While this is a relatively low-stakes example of a hallucination, some could have far greater consequences. 

Recently a New York attorney representing a client's injury claim relied on a conversational chatbot to conduct his legal research. The federal judge overseeing the suit noted that six of the precedents quoted in his brief were bogus. It turns out that not only did the chatbot make them up, it even stipulated they were available in major legal databases.

While guardrails to prevent GenAI hallucinations need to be put in place, the first step is understanding how, or rather why, these inaccuracies happen.

Why do generative AI hallucinations occur?

While leading AI experts aren't entirely sure what causes hallucinations, there are several factors that are often cited as triggers.

First, hallucinations can occur if the training data used to develop the model is insufficient or includes large gaps leading to edge cases that are unfamiliar to the model. For example, in drafting a contract, a generative model that was fine-tuned using contracts from the financial services industry may not have sufficient exposure to terms and concepts in healthcare law to draft contracts for that industry. Most of the time, generative AI's primary goal is to produce an output in response to a prompt. Even if it doesn't "understand" the prompt, it might craft a response based on insufficient training data, thereby leading to a faulty result.

Similarly, some machine learning (ML) models suffer from overfitting, where the output is accurate for its training data, but inaccurate for new data. The goal of ML model training is to generalize from training instances so that the model can correctly operate on instances it hasn't seen before. Overfitting occurs when a model is trained to the point where it effectively memorizes the inputs and appropriate outputs for the training set, leaving it unable to generalize effectively to new data. For example, a model used to approve loan applications may be considered to be 90% accurate in predicting the likelihood of loan default. However, if it has been subject to overfitting, it likely has an accuracy rate closer to 70%. Therefore, applying the model to future loan decisions will result in a decrease in business profits and an increase in dissatisfied customers.

Another model development challenge is ensuring training texts and prompts are properly encoded. LLMs map terms to a set of numbers — a process known as a vector encoding — and these encodings have some key advantages over working with words directly. For example, words with multiple meanings can have one encoding per meaning, reducing the chance of ambiguity. So, the word "bank" would use one vector representation when referring to a financial institution and another when referring to the shore of a river. Vector representations also allow mapping semantic operations — like finding a similar word — into mathematical operations. Problems with encoding and decoding between text and representations can lead to hallucinations.

Best practices to prevent generative AI hallucinations

While generative AI hallucinations are a concerning issue, there are ways to reduce their frequency and intensity. Consider the following best practices.

The most obvious protection from hallucinations is high-quality input data. It's critical that your GenAI model is trained on a diverse and representative dataset that covers a wide range of real-world examples. Using sufficient, unbiased data is key to training an accurate, resilient model.

Another crucial part of the training process is incorporating human feedback, also known as reinforcement learning from human feedback (RLHF). Because AI models lack a nuanced understanding of language, they can produce responses that are out of context or irrelevant. RLHF provides models with regular corrections via feedback from people, so that an understanding of the world that only a human can have is incorporated into the training process.

Transparency in AI ensures that it's possible to understand a model's inner workings and how it makes decisions. The issue is coming to the forefront as LLMs evolve, particularly because more advanced models are harder to understand.

Users and machine learning professionals should be able to evaluate how a model was trained and review for potential weaknesses in the learning process. Transparency also makes it easier to identify and correct errors that can lead to hallucinations.

A malicious user of a generative AI system could experiment and probe a model looking for ways to make it hallucinate. It's necessary to incorporate counter measures to help monitor and control the types of inputs allowed in generative AI systems. These include using adversarial examples in the model's training to improve its ability to respond appropriately and to audit the model's anomaly detection capabilities. A further means is by data sanitization, which involves checking that no malicious inputs are being fed into the model. And finally, security updates and patches can help block external threats to your AI model.

While generative AI hallucinations are a frustrating hurdle that can significantly impact user trust, they aren't insurmountable. By training models using high-quality data, ensuring model transparency and enacting effective quality control, developers can improve their model's resiliency to deliver accurate predictions.

Hallucinations are an obstacle to building user trust in generative AI applications. Learn about the phenomenon, including best practices for prevention.

Single line drawing of the side profile of a head and the term "AI" imposed on a wavy background connoting hallucination

ai hallucination

Check out our solutions

Generative AI

Generative AI hallucinations: Why they occur and how to prevent them

Facial recognition is a field within AI and computer science that seeks to give machines the ability to interpret human faces. This beginner’s guide explains types of face recognition processes, how they work, various applications and how accurate they are today.

Person with facial recognition / biometric analysis

What is Facial Recognition? – Applications & How it Works

What is Facial Recognition? Applications and How it Works

In this virtual chat with TELUS International computer vision experts Ajinkya Malasane and Kaushik PS, we’ll dive deep into end-to-end sensor fusion data collection and annotation solutions for autonomous vehicle training needs.

Three cars on road with labeling boxes around them

3d data annotation

Leveraging automation for labeling large 3D pipelines

The cover of The future of content moderation: Strategies and tools for 2023 and beyond e-book with a fern in the background.

The future of content moderation: Strategies and tools for 2023 and beyond

Organizations have a responsibility to create and maintain digital experiences that are truthful, welcoming and safe. Learn how brands can keep pace with the growing levels of user-generated content (UGC) to build trust with their customers and protect their reputation.

An illustration of various people creating user generated content on their phones.

Preview image of the Forrester report "How To Set CX Metrics Goals" and a plant

TI Resource ForresterReport

Gain an understanding of the challenges faced by customer experience (CX) leaders in setting goals for their metrics — and effective techniques to persevere.

Illustration of a person and various graphs and scales symbolizing metrics, measurement and KPIs

Hero image for Forrester - How to set CX metrics goals - 1620x550

Forrester - How to set CX metrics goals - card image

How to set CX metrics goals

Uncover how travel and hospitality brands can combat fake reviews, scams and more with a robust content moderation strategy.

Airplane leaving a trail of five stars, meant to symbolize positive customer experiences in the travel and hospitality industry

travel como3

Four ways content moderation can build trust for travel and hospitality brands

Preview image of the case study "Creating annotation work opportunities through impact sourcing" with plant in background. 

resource annotation impact sourcing

Discover how TELUS International and the SETU Foundation spearheaded an initiative to provide occupational and learning opportunities to young women in India while delivering high-quality results for the client.

Two women sit in front of computer monitors doing annotation work.

annotation impact sourcing

Creating annotation work opportunities through impact sourcing

Preview image of the report "CX outsourcing in 2023" accompanied by a plant

TI Resource WhitePaper 2023

Delve into new insights about customer experience outsourcing from a survey of 712 enterprise executives with strategic decision-making authority.

Illustration of customer support agents surrounded by related iconography like charts, star ratings and text boxes.

CX Outsourcing2023 Hero Image

Cover image for the dynamic report "CX outsourcing in 2023" featuring the title and illustrations of customer support agents and related iconography 

CX Outsourcing in 2023

Executive survey results: CX outsourcing in 2023

TI resource eBook 800x450 The essential guide to training data 

Discover best practices for the sourcing, labeling and analyzing of training data from TELUS International, a leading provider of AI data solutions.

guide to AI training data - website hero image 

The essential guide to AI training data

Join experts from TELUS International and Automation Anywhere, a top cloud automation platform, to learn how banks and financial services firms can leverage automation to enhance the customer experience, optimize operations and foster customer loyalty.

An illustration showing various digital automation technologies for banking including, artificial intelligence, mobile payments, chatbots and data analytics.

The future of banking and financial services webinar

The future of banking and financial services: Unlocking customer experience potential with intelligent automation

Model drift degrades the accuracy of your ML model over time. Discover some best practices for detecting and mitigating it quickly and efficiently.

Illustration showing lines separating into strands and becoming increasingly separated, meant to symbolize machine learning model drift

ml drift

How to detect and mitigate machine learning model drift

industry-icon-ecommerce 2x

Depiction of data and icons meant to symbolize financial services and fintechs

bfsi solutions hero

Fintech & Financial Services

industry-icon-tech-games 2x

Industries - Games page - Hero

icon-ecommerce 2x

Retail and ecommerce hero banner displaying a smartphone with a buy button, mouse cursor and a bunch of shopping bags

ecommerce hero

industries-icon-automotive

Autonomous vehicles driving on a purple background with mobile phone indicating location.  

Automotive Industry Page - Hero Banner

industry-icon-travel

The view of a sunset from an airplane window.

Travel and Hospitality hero

icon-tech 2x

Industries - Technology - hero - landscape

icon-communications 2x

Abstract image of interconnected data points meant to convey networks and connection

Communications and media hero

icon-healthcare 2x

An illustration of a collection of icons that represent the healthcare industry, such as, a needle, DNA, a pill, a blood drop and a clipboard.

Healthcare - Hero

hero risk solutions

Trust, Safety & Security

Hero - Solutions category - digital experience

Digital Experience (DX)

Hero - AI Data Solutions

AI Data Solutions

hero advisory

Advisory Services

hero back office

Back Office & Automation

IT Lifecycle - cloud, application development, system operations, service desk, QA and Testing, IoT

it lifecycle2 hero

IT Lifecycle

Hero - Solutions - Customer Experience

Customer Experience (CX)

Invest in and empower your support teams to truly excel at serving customers.

Learning Excellence Solutions

Strive for the perfect connected ecosystem of engagement between sensors, devices, data and more. 

Internet of Things (IoT)

Build and manage multi-cloud platforms to accelerate digital transformation.

Cloud Platform Solutions

Uncover the right processes for enhanced business efficiency and CX outcomes.

CX Consulting 

View every touchpoint as a brand-building opportunity.

Contact Center Outsourcing

Enhance your sales pipeline with both new and current customers.

Customer Acquisition

Ensure a great user interface with an unforgettable user experience. 

UX / UI Design

Harness the power of GenAI to achieve amplified productivity and heightened customer satisfaction.

Fuel iX

Strategically source and manage your talent using our global full-service or on-demand recruiting expertise. 

Talent Acquisition - RPO

Make inefficient workflows a thing of the past. RPA can eliminate repetitive tasks and free up your valuable human capital. 

Robotic Process Automation (RPA)

AI Training Data

Engage your customers while protecting your online reputation.

Content Moderation Solutions

Our Platform

Address the problems inherent in legacy systems and digital adoption.   

Platform Transformation

Optimize the patient experience for improved health outcomes.

Healthcare / Patient Experience

Ensure optimal performance with quality products that deliver amazing user experiences.

Software Testing Services

Build your digital roadmap using customer-centric approaches and human-centered design.

Digital Consulting

Data Collection

Enhance CX across all channels with our holistic approach. 

Computer Vision

What is Facial Recognition? Applications and How it Works

What is Facial Recognition? Applications and How it Works

Be the first to know