EXPLORING

THE CHANGING
AI LANDSCAPE
From Analytical to Generative AI
April 2023

AUTHORS
Rich Timpone, Ph.D.
Michel Guidi

IPSOS
VIEWS
 As with past major transformations, there is shape a better tomorrow, and our role in coming
ample reason to be both excited and concerned. months and years will be more important than
Ipsos is taking advantage of advances possible ever; as we provide a Total Understanding of
with this new technology, as well as building on how the world is changing, how people decide
the existing tools by leveraging our knowledge and act as citizens, consumers, patients, and
and data to improve the models, and helping our professionals as well as how they embrace the
clients navigate the new landscape. Our reason new technology in these roles.
to be is to give people a voice to understand and

ChatGPT reached its first million users in only

five days. The recognition of the potential of the GENERATIVE AI AND THE AI LANDSCAPE
tool and a fear of missing out has led a huge
number of individuals and companies to begin There are two main branches in the representation and application of ‘intelligence’ by machines with AI:
exploring its potential.
• Analytical AI has been in use for some time for tasks such as predictive
While the domain of Generative Artificial it took 3.5 years for Netflix to reach one million analytics as well as image and speech recognition. For example, we at Ipsos are
Intelligence (AI) has been growing in recent users. Since then, it took Facebook 10 months, using it to generate new insights, change how we engage with respondents, and
years, ChatGPT accelerated its adoption and Spotify five months and Instagram 1.5 months for process automation. In these areas we leverage specific tools like natural
broke through as the first “mass” Generative to reach this level. In comparison, ChatGPT language processing (NLP), speech to text transcription, image analytics, etc. in
AI application. It is important to take notice and reached its first million users in only five days.1 domains ranging from audience measurement to social listening to crisis support.
put this and other emerging tools in perspective The recognition of the potential of the tool and
as the implications of the technologies will be a fear of missing out (FOMO) has led a huge • Generative AI is a more recent extension that can create new things across
far reaching and very fast. number of individuals and companies to begin all media that would have thus far been seen as unique to human intelligence or
exploring its potential. creativity: text, video, audio, pictures – every digital medium can be powered by
As with the advent of the internet, this new Generative AI.
technology will change how we work, consume, It is our intent to put in perspective the
communicate, get informed and educated, etc. opportunities while recognizing some of the
The rise of Generative AI will have similar seismic limitations and watchouts for Ipsos and our While ChatGPT has recently captured the public the breadth and competition that will further
consequences on how we live our lives. This time clients to consider in the exploration and imagination in a compelling way, the Generative accelerate advances going forward.
though, the pace of acceleration is so incredible that implementation of ChatGPT, LLMs (Large AI space is getting crowded very quickly,
these tools – which are shocking and awing us now Language Models), even newer multimodal and will improve exponentially in the coming The size of the models and breadth of the
– will seem quaint by the end of the year because models (text and image like in the newly months. OpenAI’s release of the much more training sets are creating more general tools
their capabilities are increasing exponentially, and released GPT-4), and Generative AI more powerful GPT-4, Google’s release of Bard, as from the narrow AI we are used to seeing
they are being embedded in other tools and services generally. In this regard we will see that while well as image generation tools such as Dall-E that solve specific problems. The continued
far faster than past advances. the potential, and use cases, are often new, the 2, Midjourney, and Craiyon, and even the latest advance in this space and the combination of
evaluative criteria that have been used for other breakthroughs that allow loading and training AI algorithms will further the generalizability
To put the pace of adoption in perspective, even areas of AI, Machine Learning, and analytics LLMs on personal machines (i.e. running LLaMA making them even more ubiquitous across our
in this era of rapid internet adoption, ChatGPT’s in the past will allow teams to get the greatest and Alpaca with Dalai, although with limitations lives.
embrace has been astounding. For comparison, value out of the tools. on what it can be used for), demonstrate
2

2 IPSOS VIEWS | THE CHANGING AI LANDSCAPE THE CHANGING AI LANDSCAPE | IPSOS VIEWS 3
 “AI WILL NOT REPLACE YOU – SOMEONE USING AI LIMITS, RISKS, AND THREATS
WILL” – OPPORTUNITIES!
While the potential is huge, adequately explicitly know the difference), highlight risks
The advent of Generative AI will change the way create hypotheses, understand the market evaluating these tools in specific use cases is for companies and public opinion of generating
we work, interact, search for information, and landscape and your place in it, do projective key to getting the most value from them while and propagating misinformation.5 The latest
create content. It is a “supercharger” that will exercises, analyze, summarize, and share avoiding significant potential risks. While some breakthroughs are democratizing LLMs and
augment the output of those that use it well. knowledge throughout your organization. At of the existential questions on how these tools allowing more to build their own models. The
Ipsos, we are exploring the use of Generative may change work and society are beyond the quality and accuracy of these will absolutely
If you are a business or marketing leader, it AI for use cases as diverse as enhancing scope of this brief point of view, we focus need to be evaluated as they will likely be less
can significantly evolve and improve the way desk research, questionnaire and discussion here on practical questions of how to evaluate transparent than the major players; OpenAI,
you determine and design your new products guide development, scripting, coding, and their utility in practice. Although there are Microsoft, Google, Meta, etc., and trusted
and services, and how you market, sell to, data processing, analysis, data summaries, some unique new aspects, the dimensions partners become even more important.
service and support your customers. It is knowledge management, and reporting and overlap with how we have evaluated Analytic
also critical that you understand how your identifying meta-analysis themes across AI, Machine Learning, analytics, and data in “Never trust anything that can think for itself
customers find your brand or company, and projects. the past. We classify these at Ipsos into the if you can’t see where it keeps its brain.” That
how they interact with you. It is already being domains of Truth, Beauty, and Justice. 3
sound advice, given by Mr. Weasley to his
leveraged to accelerate the creation of new Beyond others who are experimenting with daughter Ginny at the end of “Harry Potter
product concepts, copy and messaging, and and exploring these tools, Ipsos is additionally TRUTH and the Chamber of Secrets”, could well apply
advertisements themselves. improving and building on their core capabilities to AIs. While based on gigantic and ever-
by augmenting and training on domain and Given the importance Generative AI promises for growing data sets, Generative AIs are entirely
If you are an insights professional, Generative specific data, while also considering the learning, research, and production, the accuracy dependent on the data they were exposed
AI will help you every step of the way: evaluation of the models to address their of these tools is of primary importance. Although to. A good example is the public version of
perform desk research, write proposals, limitations as well as their great potential. the tools provide results with authority, they are ChatGPT, which was not exposed to data post
not always right. While we will not go into detail 2021. Understanding what the “training data”
on how these tools work, it is worth highlighting was, and if there is evaluative criteria, will be
that they are designed to identify the next word fundamental to determine the level of trust we
in a sequence based on their training data and can grant the output of any Generative AI.
sequences of text built to that point. This makes
4

their language incredibly natural, but the content

is not necessarily accurate. If you are a business or
marketing leader, it can
Knowing how to ask and sequence questions is
important (an element of prompt engineering),
significantly evolve and
but while the tools can be invaluable to improve the way you
accelerate work for subject matter experts who determine and design your
can evaluate the content, it is riskier for learning new products and services,
and emergence. Examples of companies and
their products harmed by being erroneously
and how you market, sell
classified based on training data with errors and to, service and support
from sources of opinion (as the models do not your customers.

4 IPSOS VIEWS | THE CHANGING AI LANDSCAPE THE CHANGING AI LANDSCAPE | IPSOS VIEWS 5
 “The greatest enemy of knowledge is not word and not necessarily true citations. When readily available. It will be absolutely necessary feedback to re-train the models. This treats the
ignorance, it is the illusion of knowledge”, Henry Kissinger asked ChatGPT for six sources to see the sources of the data used by an AI to data as public and could go so far as violating
Stephen Hawking warned us. Generative AI on ‘Kissinger’s thoughts on technology’, one form a response or recommendation. GDPR and customer contracts or give away
is not always accurate, consistent across was an actual title (with the date wrong), but competitive advantage, and Italy became the
languages or trained for all language tasks.6 the other five were pure fabrications that were “I who wrote this message, am a Certified first western country to ban and investigate
If you have worked with ChatGPT already, you generated to sound like legitimate citations.7 Human”. Generative AI is going to allow you ChatGPT for privacy concerns. Second, as
may have noticed how “confidently wrong” it This is why using LLMs where you have the to speak to ‘people’ on the internet, via text with all AI tools, the accuracy, fairness, and
can be. Because of the conversational nature of understanding to confirm the quality of the and video, who will look like human beings but implications of the models can create their own
such applications, and because some often offer content yourself and check the references can will in fact be avatars.9 While that has many ethical issues and the more autonomous the
one clear, well written, structured response, be invaluable, but for learning new things it positive implications, it also means that it will be tools, the greater the risk. Finally, Generative
with little nuance, we would be inclined to trust should be used more cautiously. increasingly hard to know if the person you are AI adds a new dimension of ethics around the
them. But they are sometimes wrong, for all talking to is a human being. And while avatars rights of the creators of original content that
confidence they convey. Given the magnitude of the errors found, even will become more and more “intelligent”, they feeds into the training. Thus, the rights of artists
the impressive improvement of GPT-4, being still will not exist in real life, they won’t vote, and authors whose styles can be emulated,
BEAUTY 60% less likely to make stuff up, still leaves a they won’t need healthcare, or consume your or examples where text snippets have already
lot of room for concern if one does not actively goods and services. There will be cases where been found that are challenged as plagiarism
When considering beauty in AI models, check accuracy and the sources provided.8 using “synthetic data” will be valuable; in general (not just by students, but corporate messaging,
one focus is on their explainability. While While major LLMs are working to reduce and though, we will need validation to make sure we speeches etc.) raise questions on the rights and
explainability is the most important aspect of minimize ‘hallucinations’, where the tools are speaking to ‘real people’ when we want to. responsibilities of original creators and the users
this dimension, it is again even more critical generate responses that go beyond the training of these tools.
for Generative AI. The description of how LLMs data and are factually inaccurate, this is still an JUSTICE
build their text is again relevant to understand area of significant current concern. “When it’s free, you are the product.” Whatever
issues here. If you ask ChatGPT for citations for While this dimension used to be viewed as less someone inputs into an open Generative
a ‘fact’ or position it provides, it will readily do “In God we trust, all others must bring data.” important than accuracy and explainability, AI application becomes publicly available.
so. However, these are also generated word by Generative AI will have to make its sources given the broad use and practical implications, Therefore, one must make sure we keep away
the fairness and ethical issues of AI and from it any IP, Personally Identifiable Information
Generative AI are critically important today. 10
(PII), anything proprietary or of high value, as it
There are three critical aspects to this domain. would then become available to anyone on the
First is the use of data. One should never web. That is in the process of being resolved by
put client data, confidential, or Intellectual creating “walled gardens” within which specific
Property (IP) in the current versions that use Generative AIs can be trained on proprietary

Generative AI is going to allow you to speak

to ‘people’ on the internet, via text and
video, who will look like human beings
but will in fact be avatars.

6 IPSOS VIEWS | THE CHANGING AI LANDSCAPE THE CHANGING AI LANDSCAPE | IPSOS VIEWS 7
 information that will only be accessible to the unveiled plan to integrate ChatGPT into its
relevant people. Except in those cases though, suite of products). A few such AIs will likely
beware of what you share and never simply dominate the landscape at that time, and for
assume feedback is off. the vast majority of users that will not only level
the playing field, but it could also lead to less
“Average is not enough”. Fast forward a creativity with people becoming too lazy to think
couple of years, when Generative AI is likely by themselves. We could observe a new era of
everywhere (including embedded in Word, regression to the mean, of flat dullness induced
Excel, Teams, etc. as per Microsoft’s recently by a few AIs that dominate the conversation.

HOW IPSOS HELPS NAVIGATE THIS

BRAVE NEW WORLD
While the last section raises issues that need and leveraging quality data to build on base
to be considered, they do not temper the huge models to improve them further for practical use
impact that LLMs, multimodal models, and cases. We will fine tune and extend the existing
Generative AI will have. These tools are already models with Ipsos data and IP, leveraging our
being embedded in tools used by millions decades of expertise as well as our 20,000 staff
everyday from search engines to productivity across 90 markets to train our selected AIs
suites and many more. Given their potential for on not just what everyone knows, but also on
learning, research, insights, and production, where Ipsos adds decisive value, to your benefit.
building on the explicit evaluation of use cases
is key. In this way, we see the greatest potential If you are a business leader, you may already
by leveraging these tools with humans in the be experiencing FOMO: you may feel that
loop at Ipsos. something is happening, that it’s big, and
you will want to understand what that means
The idea of having humans in the loop has been for your customers, providers, technologists,
key in AI for years and is relevant for all aspects and employees. At Ipsos, we can help
of the conceptualization, selection of training you understand what is changing for your
data, training, modeling, and use of AI models to organization, your constituents or stakeholders,
maximize their utility and quality as we strive for your industry, and the world. That’s what we
We will continue to lead the way by truth, beauty, and justice. have always done, and we will continue to lead
understanding how Generative AI the way by understanding how Generative AI

influences and impacts how we live, At Ipsos, our understanding of data is not
less relevant, but more so in the era of AI and
influences and impacts how we live, work,
learn, get informed, vote, consume, etc. as
work, learn, get informed, vote, Generative AI. This is key for both understanding well as how to leverage it for best effect while
consume, etc. as well as how to how best to leverage the models, from desk mitigating the challenges that exist.
leverage it for best effect while research to scenario-gaming, as well as creating

mitigating the challenges that exist. 

8 IPSOS VIEWS | THE CHANGING AI LANDSCAPE THE CHANGING AI LANDSCAPE | IPSOS VIEWS 9
 REFERENCES
Cover: This image was generated by Ipsos with the assistance of Midjourney

1. Buchholz, K. (2023). “ChatGPT Sprints to One Million Users”, statista.com. https://www.statista.

com/chart/29174/time-to-one-million-users/

2. Palmer, S. (2023). “The Dalai LLaMA is Wreaking Havoc on Tech Timelines.” 21 March, 2023. Shelly
Palmer blog. https://www.shellypalmer.com/2023/03/the-dalai-llama-is-wreaking-havoc-on-tech-
timelines/

3. Taber, C.S., and Timpone, R.J. (1996). Computational Modeling, Quantitative Applications in the
Social Sciences #113, Sage Publications, Thousand Oaks, London and New Delhi.

4. Wolfram, S. (2023). “What Is ChatGPT Doing … and Why Does It Work?” 14 February, 2023.
Stephen Wolfram Writings. https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-
In an era where past data will be more readily When working with Ipsos, you will have the and-why-does-it-work/
available than ever, we can help you ask the guarantee to know the sources of the data
right questions to detect the signal from the used, and also that you are speaking to “real 5. Freyfogle, E. (2023) “Don’t believe ChatGPT - we do NOT offer a ‘phone lookup’ service” 23
noise. How to converse with an AI, how to people” when you want to. Our efforts to ensure February, 2023. The OpenCage Blog. https://blog.opencagedata.com/post/dont-believe-chatgpt
craft the most relevant and useful prompts, is quality, accurate data and insights, and counter
going to become more important than ever. At misinformation carry over into our use and 6. Joseph, Y. (2018). “Stephen Hawking, in His Own Words” 14 March, 2018. The New York Times.
Ipsos, we are experts in the art of the question, support of AI as well.
and we will work with you and your teams to 7. Kissinger, H., Schmidt, E., and Huttenlocher, D. (2023) “ChatGPT Heralds an Intellectual
make the most of Generative AI and help you Ipsos’ motto “You act better when you are Revolution.” 25-26 February, 2023. The Wall Street Journal.
understand the ways to evaluate them for your sure”, is more relevant than ever. Please reach
own use. out to your Ipsos team for assistance. 8. Heaven, W.D. (2023). “GPT-4 is bigger and better than ChatGPT—but OpenAI won’t say why” 14
March, 2023. MIT Technology Review. https://www.technologyreview.com/2023/03/14/1069823/
gpt-4-is-bigger-and-better-chatgpt-openai/

9. Appleton, M. (2023). “The Expanding Dark Forest and Generative AI.” Essay:
https://maggieappleton.com/ai-dark-forest
How to converse with an AI, how to
craft the most relevant and useful 10. Timpone, R. and Yang, Y. (2018). “Justice Rising: The Growing Ethical Importance of Big Data,

prompts, is going to become more Survey Data, Models and AI.” Paper presented at the BigSurv 2018 Conference of the European
Survey Research Association; Barcelona, Spain.
important than ever. 

10 IPSOS VIEWS | THE CHANGING AI LANDSCAPE THE CHANGING AI LANDSCAPE | IPSOS VIEWS 11
EXPLORING THE
CHANGING AI
LANDSCAPE
From Analytical to Generative AI

AUTHORS
Rich Timpone, Ph.D. Head of Global Science Organization, Ipsos
Michel Guidi Chief Operating Officer, Ipsos

The Ipsos Views white

papers are produced by the
Ipsos Knowledge Centre.

www.ipsos.com
@Ipsos