EBOOK

Build Responsible
Generative AI
Applications:
Introducing the
RAFT Framework
Introduction: Great Power,
Great Responsibility
Recent excitement around Generative AI — in particular Large Language Models (LLMs) — means
organizations are pushing forward with the use of AI at an unprecedented pace. There has arguably
never been a more pivotal time in the history of AI.

At the same time, it’s important to stay grounded. The truth is that flaws within AI systems and the
data they are built on can present — and have presented, even before the rise of Generative AI —
real risks.

More than ever before, organizations need to think about building AI systems in a responsible and
governed manner.

In this ebook, we will deep dive into those aforementioned risks plus introduce the RAFT (Reliable,
Accountable, Fair, and Transparent) framework for Responsible AI, showing how it can be applied
to both traditional and Generative AI systems.

EBOOK 2 DATAIKU
Understanding Values for Responsible AI
Whether your business is scaling the use of AI across the organization, interested in experimenting
with the latest developments in Generative AI, or simply looking to make sense of forthcoming
regulation, it’s important to have a set of guardrails defined for the use of AI at your organization.

Various standards organizations and governments have proposed frameworks for AI values — see
Table 1. These are a good starting point, but in the next section, we’ll take it one step further with a
more specific, robust, and tested framework.

TABLE 1: MAPPING OF AI RMF TAXONOMY TO AI POLICY DOCUMENTS

Source : NIST AI RMF V2

EBOOK 3 DATAIKU
The RAFT Framework for Responsible AI
To support the scale of safe AI, Dataiku offers a number of built-in features for responsible design,
deployment, and governance. These features align with a value framework similar to those
proposed in Table 1, which any organization can use as a starting point for their own Responsible AI
programs.

Note: See page 11 for the full framework.

These values make up Dataiku’s baseline approach to Responsible AI — we call it RAFT for Reliable,
Accountable, Fair, and Transparent. The values outlined in the RAFT framework are crucial for the
development of AI and analytics, and they cover both traditional and new methods in Generative
AI.

To effectively execute on these principles requires understanding the potential risks and impacts
of technology. In the rest of this ebook, we will cover the specific risks of Generative AI and broader
approaches to Responsible AI practices.

Our understanding of the risks and necessary steps to reduce potential harm from AI is meant as a
starting point for organizations looking to build best practices in governed and Responsible AI. In
addition to building best practices, these guidelines can support organizations in their readiness
efforts toward upcoming regulations, such as the EU AI Act.

As the field of AI continuously evolves, so do our approaches to safe and scaled uses of new
technology. We encourage readers to take the suggestions and recommendations here and adapt
and expand them as needed per industry and regulatory standards.

EBOOK 4 DATAIKU
Risks to Using
Generative AI in the Enterprise
AI systems are inherently socio-technical, which means “they are influenced by societal dynamics
and human behavior.”1 Acknowledging the way AI impacts society and vice versa allows us to
better anticipate potential negative consequences and proactively reduce or prevent those harms
before they occur.

In addition to the socio-technical risks from Generative AI, there are also emerging legal
considerations around privacy and copyright infringements. Below, we list a number of risks that
may arise in the use of Generative AI in the enterprise. These risks are common across various types
of Generative AI technology but will surface in different ways across use cases:

Toxicity: Toxic, obscene, or otherwise inappropriate outputs.

Polarity: Unfair positive or negative attitudes to certain individuals or groups.

Discrimination: Model performance is less robust for certain social groups.

Human-Computer Interactions2: Over-reliance on the outputs of AI due to

perceived sentience or blind trust in an automated system.

Disinformation: Presenting factually incorrect answers or information.

Data Privacy: Input data shared back to 3rd-party model providers and possibly
shared as future outputs to non-authorized users.

Model Security: Ability for a user to circumvent security protocols intended to

prevent social-technical harms or gain access to unauthorized data.

Copyright Infringements: Redistribution of copyrighted material, presented as

original content.

1. https://www.nist.gov/news-events/news/2023/01/nist-risk-management-framework-aims-improve-
trustworthiness-artificial#

EBOOK 5 DATAIKU
The potential harms listed here are not exclusive to language models, but they are heightened by
the use of natural language processing (NLP) techniques to analyze, categorize, or generate text in
a variety of business contexts.

Understanding and addressing these risks before implementing an LLM or other Generative AI
techniques into an AI system is crucial to ensure the responsible and governed use of the latest
technology. By thinking through potential harms, designers, developers, and deployers of LLMs can
set thresholds for risk and incorporate mitigation strategies in each stage of the AI lifecycle.

Risks of Generative AI by Context

Beyond the inherent, high-level risks associated with Generative AI technology, businesses should
consider the context in which the system will be deployed.

A baseline approach is to assess the use case across two dimensions:

1. Target of analysis, which focuses on the type of data or documents that the model will make
use of to generate output. Corporate or business documents include items as invoices,
legal agreements, or system documentation. Individual or personal data can include
traditional tabular information about a person’s specific characteristics, as well as call center
transcriptions or text written by an end user. Academic or domain-specific texts are typically
used in industry research and analysis, such as medical publications, manufacturing research,
or legal codes.

EBOOK 6 DATAIKU
2. Delivery method, which looks at how the output of a model is delivered to end users. In the first
instance, the output is shared as a report, recommendation, or suggested actions in response to
a single query. This is different from the next category of virtual assistants, chatbots, etc., that
respond in a human-like way to end users’ queries. The final category is automated processes,
such as sending mass marketing emails or robotic process automation (RPA).

Each category within these two dimensions will carry different risk tradeoffs and strategies to
prevent harm to the business, clients, and broader society.

Target of Analysis Delivery Method

Corporate or Business Documents Output Shared as a Report,

Recommendation, or Suggestion Actions
While social risks from the use of these documents
are lower, it is important to ensure the documents are Generally this type of delivery ensures that end
up to date and relevant for the question at hand. This users have control over how the output is shared or
should include designing the input parsing strategy deployed down the line. It provides an opportunity for
to leverage the correct documents and content to review of outputs before action is taken, allowing the
answer a given query. Prevent leakage of sensitive end user a chance to review explanations and provide
or copyright material into the model output — users feedback.
should not be able to circumvent model parameters to

Individual or Personal Data Virtual Assistant

Personal data on customers, users, or patients should Virtual assistant delivery includes chatbots, text to
be protected from unauthorized access. Models that voice responses, and question answering. These
make use of personal information to draw conclusions interactions should be clearly marked as a computer/
about individuals should be tested for fairness across model and not a human agent. Extra precautions
subpopulations of interest, language output should should be taken to make it clear the model is not a
not be toxic or reinforce stereotypes about groups of sentient agent and could potentially provide incorrect
people, and end users should know that their data is responses. Additionally, generative text models
being used to train and deploy models (with the ability should be developed with guardrails around the type
to opt out from this usage). Take care that personal of language and discussion permissible by the model
information or details that are not relevant to the and end users to prevent toxic or harmful conversation
specific use case are not used in the model or shared from occurring.

Academic or Domain-Specific Texts Automated Process

Texts should be carefully cultivated to ensure fit Results of a Generative AI model may be passed
with the intended use case. Model outputs should be directly to an end user without review. Automated
able to provide citations to real texts in the corpus to processes are typically used to scale AI use, meaning
support any generated answers or recommendations. real-time human intervention is not possible. Regular
review of model outputs, quality, and the ability to
pause automated processes when necessary is critical
to catch any potential harms. Ensure you clearly
document accountability over stopping an automated

EBOOK 7 DATAIKU
Concerns Based on Expected Audience
One important dimension of Generative AI use cases we have not yet covered is the type of
audience or expected reach for model outputs. The audience for model outputs are usually
business users, consumers, individuals or some combination of both groups.

However, no matter the expected audience for model outputs, there are core criteria that should
be met in the deployment of any AI system — these four criteria further the goals of reliability and
transparency as detailed in the RAFT principles and support broader trust in AI systems:

EBOOK 8 DATAIKU
Assessing Potential Impacts of AI
Before deploying an AI system —generative or not — it’s important to assess the potential
impact on individuals or groups once a solution is in use. While specific impacts or unintended
consequences will vary from use case to use case, Dataiku suggests two broad dimensions that can
be used to understand impact from a deployed AI system. These impacts and potential risks are
based on standards such as the NIST Risk Framework or the EU AI Act and are meant to guide our
customers as they implement AI systems.

Depending on the use case, an AI pipeline may have more than one type of model providing output.
For example, the next best offer project3 uses both a traditional recommendation model and an
LLM to help write the text of an email to a customer.

In such an instance, it is important to assess impact and apply Responsible AI principles to both
sets of models in the project. Potential bias or poor performance from a recommendation model
will have different impacts than bias or toxicity from a language model.

The risk scoring for unintended consequences is based on two variables:

1. Whether the risk could materialize as a harm to individuals and groups directly because of
the solution’s implementation or indirectly because of some constellation of factors that are
difficult to qualify at the time of deployment.

EBOOK 9 DATAIKU
2. Whether the risk could materialize as a harm immediately or over a longer period of time.

This results in two larger guiding questions about the nature of an AI system :

Q1 : Does the output of this project lead to direct impact on individuals or groups?
Q2 : Is the impact felt immediately or observed over time?

Putting these two variables together, we qualify the solution’s Responsible AI considerations in one
of the following categories :

Putting Principles into Practice

How do we move from defining principles, potential risks, and impacts of AI to implementing best
practices and guardrails in the development of AI systems?

Let’s return to the RAFT framework, which provides a baseline set of values for safe AI and that can
serve as a starting point for your organization’s own indicators for Responsible AI. We encourage
the governance, ethics and compliance teams at your organization to adapt the framework to
accommodate specific industry requirements, local regulations or additional business values.
As with assessing impact, it is necessary to apply the principles from the RAFT framework to all
models (both traditional and generative) used in an AI system.

EBOOK 10 DATAIKU
 EBOOK

Note: Specific methods to assess and manage the bias of language models or other Generative AI are still in development. When building
or fine tuning a model, developers should use diverse and representative datasets and check model performance against risks like polarity,
Conclusion: An Evolving World & Ongoing
Considerations
It’s only the beginning for Generative AI, which means we’re only at the beginning of our
understanding of the extent of the opportunity — as well as the risks — it presents.

In addition to the RAFT guidelines for Responsible AI, Dataiku is proud to offer a comprehensive
training on how to implement Responsible AI in practice. These courses are available for anyone
who wishes to gain hands-on experience with measuring bias, building fairer models, and creating
compelling reporting on data science pipelines with Dataiku. The training covers core concepts in
Responsible AI through the lens of classification models, but much of the teachings around bias,
explainability and reliability can be applied to Generative AI models as well.

Access the Responsible AI training on Dataiku Academy

Here are some additional considerations to keep top of mind moving forward when it comes to
leveraging Generative AI in the enterprise.

EBOOK 12 DATAIKU
Third-Party Models
In many cases today, businesses will rely entirely on pre-trained, third-party models to support a
Generative AI use case. In this situation, risks arise from the inability to know how the model was
trained, not having confirmation that it is not biased or unreliable, and having no control over how
input data is shared back to the third-party provider for retraining the model.

The use of third-party models should be closely monitored across the organization, and prompts
should be regularly reviewed to ensure private data is not supplied as inputs to the model.
Additionally, education on how to work with the third-party model (plus test the reliability of
outputs) should be available for users.

Environmental Impacts
An additional area of concern is that of the financial and environmental cost of training and
deploying LLMs. Given the numerous pre-trained models available for commercial and open-source
use, many organizations do not need to build an LLM from scratch. Instead, many businesses opt
to fine tune existing models on specific data, which is far less costly and requires less compute
resources.

Fine tuning an existing model is therefore a benefit to businesses who wish to avoid the pitfalls
from using third-party models and reduce the cost of training a new model. However, it is
important to note that using a generative model in a production environment requires the use of
GPUs, which will incur higher computational and environmental costs.

EBOOK 13 DATAIKU
Everyday AI,
Extraordinary People

Elastic Architecture Built for the Cloud

Machine Learning Visualization Data Preparation

Name Sex Age

Natural lang. Gender Integer

Braund, Mr. Owen Harris male 22

Moran, Mr. James male 38
Heikkinen, Miss. Laina
Remove rows containing Mr. female 26
Futrelle, Mrs. Jacques Heath female 35
Keep only rows containing Mr.
Allen, Mr. William Henry male 35
Split column on Mr.
McCarthy, Mr. Robert male
Replace
Hewlett, Mrs (Mary Mr. by ...
D Kingcome) 29

Remove rows equal to Moran, Mr. James

Keep only rows equal to Moran, Mr. James

Clear cells equal to Moran, Mr. James

Filter on Moran, Mr. James

Filter on Mr.

Toggle row highlight

Show complete value

DataOps Governance & MLOps Applications

Dataiku is the platform for Everyday AI, enabling data experts and domain experts to work
together to build AI into their daily operations. Together, they design, develop and deploy
new AI capabilities, at all scales and in all industries.

©2023 dataiku | dataiku.com

Authors
JACOB BESWICK
Jacob Beswick is Dataiku’s Director for AI Governance
Solutions and Responsible AI. Jacob leads a team that
works with organizations globally to develop their
approaches to AI and analytics governance, ensuring
that their objectives — for example, compliance, risk
management, or responsible development and use —
can be embedded in Dataiku’s platform. Prior to joining
Dataiku, Jacob served in the UK Civil Service, where he
led a portfolio covering AI adoption, governance and
regulation. As part of this, he represented the UK at the
European Commission, drafted what has become the UK’s
proposals on AI regulation, and created BridgeAI, which is
now being delivered by Innovate UK.

TRIVENI GANDHI
Triveni is a Jill-of-all-trades data scientist, thought leader,
and advocate for the responsible use of AI who likes to find
simple solutions to complicated problems. As Responsible
AI Lead at Dataiku, she builds and implements custom
solutions to support the responsible and safe scaling of
AI. This includes hands-on training to put RAI principles
into practice as well as change management programs
that help business leaders translate ethical values into
actionable indicators. Triveni also supports field product
teams with innovative methods to govern and oversee
complex AI problems — often serving as a technical bridge
between subject matter experts and engineers. Triveni
is the winner of VentureBeats Women in AI Awards in the
category of Responsibility and Ethics in AI and was listed as
100 Brilliant Women in AI Ethics in 2022. She holds a Ph.D in
Political Science from Cornell University.

EBOOK 15 DATAIKU
©2023 DATAIKU | DATAIKU.COM