June 25, 2025 • 29 mins
How can AI help navigate the complex, varied and piecemeal world of ESG regulations? In this episode of the ESG Currents podcast, Irena Vodenska, professor of finance and director of finance programs at Boston University’s Metropolitan College, joins Bloomberg Intelligence’s senior ESG analysts Shaheen Contractor and Rob Du Boff. They discuss a study that uses the power of AI to help dissect sustainability regulations. Learn more about key findings, including most-regulated themes, countries with the highest number of regulations and the superiority of AI vs. keyword-based models. The episode was recorded on June 18.
See omnystudio.com/listener for privacy information.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:07):
ESG has become established as a key business theme as
companies and investors seek to navigate the climate crisis, energy transition,
social megatrons, mounting regulatory attention and pressure from other stakeholders.
The rapidly evolving landscape has become inundated with acronyms, buzzwords,
and lingo, and we aim to break these down with
(00:29):
industry experts. Welcome to ESG Currents, your guide to navigating
the evolving ESG space, one topic at a time, Brought
to you by Bloomberg Intelligence, part of Bloomberg's Research department,
with five hundred analysts and strategists working across all major
world markets. Our coverage includes over two thousand equities and
(00:49):
credit as well as outlooks on more than nineteen industries
and one hundred market indices, currencies and commodities. I'm Shin Contractor,
senior ESG analyst Bloomberg Intelligence.
Speaker 2 (01:01):
And I'm Robb Duboff, also senior ESG analyst at BI
and we are your hosts today.
Speaker 1 (01:07):
So today we're joined by Irena Vondeska, Professor and Director
of Finance Programs and Chair of the Administrative Science Department
at Metropolitan College Boston. University. Arena recently worked on a
study that helps navigate ESG regulations using AI, two topics
that I think are extremely timely. Now we all know
(01:30):
that ESG related regulations are complex to this to say
the least varied and ever changing. And also AI is
top of everyone's mind right now, so I'm really excited
to learn more about this study. Arena. Thank you for
joining us.
Speaker 3 (01:45):
Thank you so much, Shahin and Rob for inviting me
and for having me here.
Speaker 1 (01:51):
Thank you So, Arena, can you tell us more about
the study, what was the motivation behind doing it and
some of the key findings that you found.
Speaker 3 (02:01):
So, as you said in the introduction, Shahin, the ESG
field or sustainability field in the corporate arena has become
a very confusing field, quite difficult to follow with all
the regulations that are upcoming. Some of them are mandatory,
(02:21):
some of them are voluntary. Globally, it has become a
real alphabet soup of documents that professionals need to follow
continuously look after new documents, and some countries do establish
rules and regulations that impose mandatory disclosures, for example in
(02:48):
the environmental area and also in the social and governments area,
So we looked into started looking into some studies and
also documents and realize that there are way over possibly
(03:09):
more than one hundred global ESG related regulatory documents that
we could analyze. There are many more, but we did
beke a selection using a hybrid model to combine traditional
keyword based searches with advanced AI power or artificial intelligence
(03:33):
powered natural language processing to understand whether we can create
a set of global documents that we can explore and
understand what do they cover, which areas within the ESG
field are mostly covered and the study initially analyzed the
(03:59):
documents to establish a baseline before using artificial intelligence for
more scalable and comprehensive analysis of these global documents. The
artificial intelligence model identified alignments of the ESG related regulatory
(04:20):
documents with the SaaS B topics, and these SASB topics
address areas for corporations that are relevant for disclosure for
specific industries, and the main results showed that once we
(04:46):
divided the regulations in mandatory or voluntary regulations, we saw
that the European Union came up as a leader in
effective and mandatory regulation coverage, addressing twenty out of the
(05:06):
twenty six SASBY topics, followed by Nigeria and Saudi Arabia
that addressed nineteen topics each and for non mandatory but
effective regulations, the US ranked highest, covering twenty one topics
from the SASB Materiality Map, and Malaysia and Mexico were
(05:32):
close behind with twenty topics covered by each country.
Speaker 2 (05:40):
So how do you define what makes an effective regulation?
Can you maybe give some examples?
Speaker 3 (05:45):
Effective regulation in this study means the definition is that
it meaningfully the regulation meaningfully aligns with one or more
of the twenty six SASB ESG topics. So effectiveness was
(06:07):
measured for example, whether the regulation clearly addresses the ESG
theme in a substantive way, it's clearly related to the theme,
whether it was currently enforce and whether the particular regulation
promoted compliance or reporting. So examples include the European Union
(06:36):
Corporate Sustainability Reporting Directive or the CSRD, Nigeria's mandatory sustainability
disclosures for listed firms and sold Arabia's comprehensive ESG requirements
for institutional investors.
Speaker 1 (06:57):
And it's really interesting to see country like Nigeria saldiri
militias the ones you mentioned. I guess a few ways
to categorize these results in these regulations by team relating
to a team's climate, social, et cetera. What were some
that were dominant?
Speaker 3 (07:18):
So, first, thank you for that question. And it's important
to understand the categorization of regulations and the dominant themes
and also some regional variations in the themes. And yeah,
the regulations were categorized using the saasby five major themes,
(07:41):
which is the environmental theme, the social capital, human capital,
business model, and innovation and leadership and governance. So the
most dominant topics within these themes were greenhouse gas emissions,
(08:03):
ecological impacts, and energy management. And less commonly addressed topics
within the documents were customer welfare for example, or competitive behavior.
And when we look into the environmental topics and the
(08:28):
dominant countries that surfaced were the European Union, the US
in coverages, and Nigeria and Saudi Arabia.
Speaker 2 (08:40):
Interestingly, yeah, I think that, I mean, it definitely makes
sense from what we see that there's a lot of
regulation on the environmental side, but I'm curious on the
social side. You mentioned that probably didn't have as much
strength and regulation, But were there any countries that maybe
stood out when it comes to regulation on the social pillar?
Speaker 3 (09:00):
All right, on the social regulation side, we looked into
the social capital and human capital. So if we look
in the mandatory regulation coverages, Saudi Arabia, Hong Kong and
Indonesia scored the highest on social capital with coverage scores
(09:24):
of five and they were closely followed by Nigeria, Singapore
and European Union each scoring four, and they were also
highest in human capital. The ones that surfaced were against
Singapore that we mentioned, but also in the human capital
(09:47):
we saw Indonesia, Nigeria and the European Union and the UK.
So these indicates a notable emphasis on community engagement, customer relations,
social cohesion in these countries regulations. And this is on
(10:09):
the mandatory regulatory coverage. So when we looked at which
countries had the voluntary regulatory coverage on social capital, highest
coverage scores were achieved by Malaysia scoring six, followed by
Australia with five and US with four. Mexico and Ghana
(10:33):
also with four.
Speaker 2 (10:36):
So were there any countries that maybe were surprising something
you expected to see but didn't or vice versa.
Speaker 3 (10:43):
Yes, actually two main surprises, I would say, if I
can single them out, were the Netherlands and Switzerland. And
so they were notable outliers because despite their reputation for
(11:03):
progressive policy, they had very low coverage in global non
mandatory is gregulatory documents. So Netherlands, for example, had total
coverage in six topics in the non man regulatory space
with zeros for social and human capital, and Switzerland had
(11:28):
only two regulatory documents. It covered two topics in the
environmental category and also zeros for social and human capital.
So we were obviously asking the question why is this
what we observe? And potentially the reason could be we
(11:52):
noticed that much of their regulation and disclosures occur locally
or through other bodies, and hence it did not appear
in global regulatory trackers which we used for this study.
Speaker 1 (12:10):
And you know, one of the interesting findings or results
was that these chatchipt models were more successful and detected
a greater number of regulations versus a keyword base model.
Now why do you think that is? And were you
able to quantify how much more successful the chat ChiPT
(12:30):
model was just curious us to those two aspects.
Speaker 3 (12:35):
Yes, so in this era of by day development of
not just new technology or using generative AI for many tasks.
We can imagine that if we continue to do this study,
or if we repeat it with the new documents, that
(12:57):
we will actually be able to capture even the development
in the area. So we're able to capture the change
over time or emphasize the role of regulations using this
fast way of retrieving the documents and then the new documents.
So this is what we call the artificial intelligence enhanced
(13:21):
retrieval using models like the retrieval outvented generation. So these
models integrate external data sources to improve contextual and factual accuracy.
And this continuous application of the method could allow for
(13:45):
real time updates and trend analysis across regions and topics.
So when you ask why we're cht GPT models more successful,
everything I said is so much more difficult to imagine
being done by just a human in real time because
it does take time to sift through the documents, to
(14:07):
read the documents, to understand them, to understand the nuances.
So the CHGPT based models outperformed the keyword searches because
they understand context and nuance, not just specific words. They
can detect implicit references to ESG topics, even when terminology
(14:32):
may vary and it's differently presented. And then when combined
with the retrieval augmented generation, they leverage external information to
validate the findings, offering more comprehensive and accurate mapping. So
just to give you a bit of an idea on
(14:53):
the example of a retrieval augmented generation and how powerful
it is, this is the retrieval augmented generation or the
RUG is an advanced natural language processing technique that combines
two powerful components. One is information retrieval, finding the relevant
(15:16):
external knowledge and second is the text generation or creating
human like responses based on that knowledge. So how does
it work? For example, if we have a Q query
and we say what are the latest ESG regulations in Europe,
(15:36):
so the retriever step will instead of relying only on
pre trained knowledge, the retrieval augmented generation uses the retriever
usually a vector database or search engine, to fetch relevant
documents or passages from an external knowledge sources I like
(16:00):
database or ESG regulations, and then it passes it to
the next step or the generator step. A large language
models like the GPT which reads both the query and
the retrieve documents and it generates a coherent, grounded answer
(16:22):
using the up to date and specific information. So it's
important to note why is it useful. It's because it
is able to retrieve up to date information. Unlike static
models that can only know what they were trained on
RAG or, the retrieval augmented generation allows real time integration
(16:43):
of current facts and in regard to the domain specific accuracy,
it can incorporate specialized content like laws, financial disclosures or
ESG frameworks that general model might not fully capture. And
most importantly, it does reduce hallucinations. By grounding the answers
(17:09):
in real documents, the retrieval augmented generation helps minimize the
risk of the model making things up.
Speaker 1 (17:21):
That's all fascinating. I think you know you spoke to
the power of generative aim. I'm now curious what are
the uses of AI for some other things in the
field of sustainability. What have you found it the most
useful in terms of in terms of your research?
Speaker 3 (17:40):
Right, So, as I said, these models are quite powerful,
but I would be amiss not to mention actually their
limitations as well. So despite the strengths they are notable limitations,
so we still need to be very careful about by
(18:00):
in the data, potentially reflecting skewed global narratives for example,
or challenges in interpreting vague or unstructured content, so we
could have potential false positives where irrelevant documents are matched
(18:21):
due to semantic similarity, or we can have difficulties in
handling language specific documents unless multilingual models are applied. So,
having said they have quite a big power and they
do have limitations, the use of AI is not limited
(18:46):
to obviously just this study. It really enters many areas,
but related to ESG and going a step further, we
could actually use AI in a broader sense in sustainability research,
(19:07):
particularly in potentially detecting greenwashing by comparing corporate ISG claims
for example independent news sources. Or we can use it
to map sentiments and consistency across stakeholder communications. Because this
(19:32):
is all texts, the natural language processing is becoming more
powerful and then we can analyze various documents. In addition
to regulations, we can analyze corporate disclosures. We can analyze
press releases news articles to reveal potential discrepancies or emerging
(19:54):
trends at a corporate level.
Speaker 2 (19:58):
So it's interesting you talk about some limitations with AI.
I wonder if there's any thoughts you have for filers
or regulators how they might craft some of these reports
so that maybe they're more friendly for an AI model
to reader, or maybe conversely, is there a risk that
maybe they can gain the AI systems by writing these
reports in certain ways or narratives in certain ways.
Speaker 3 (20:21):
So that's a very interesting question, and we obviously need
to be mindful of that, because it is possible to
write certain reports in a way to pass a threshold
and look more esg like or as if the report
(20:46):
is really positive about the company. So we also need
to be mindful of not just the way the reports
are written, but the words that are used, the sentiment
that is used in the documents. And it is possible
to create documents that may just be documents and not
(21:10):
necessarily reflect the reality. But this is why I'm saying
that we should not be looking only into a certain
type of documents produced by the corporation or the company.
We really need to look for other types of documents
as well to compare and extract better knowledge or better
(21:35):
intelligence as to the real state of the ESG or
ESG dedication of companies. So and also, the AI is
here and it isn't meant to replace the critical thinking
(21:56):
or the judgment or the expertise of the human ESG analyst.
So we're not advocating, we're not advocating that the human
is replaceable. Instead, we are advocating that the AI is
(22:16):
intended here to be really powerful tool and to help
the analysts, to help ESG analysts to augment their capabilities.
So at the end of the day, we believe that
the human needs to be definitely in the loop and
bring that final conclusion to the table of what really
(22:42):
is read, what is understood by the documents, by the comparison,
by the analysis, and to make that final determination of
the result.
Speaker 2 (22:53):
So hopefully Shaheana and I will still have jobs, but.
Speaker 3 (22:57):
You will be more successful in your.
Speaker 2 (22:59):
Jobs, let's hope. I just want to circle back again
on this idea of regulation. So you know, governments are
starting to work to refine regulations. What would your advice
be to them? Is there anything your study can show
about how to make regulations more effective?
Speaker 3 (23:15):
So it's interesting to think of regulations, as I mentioned
in our study we looked into what is mandatory and
what is voluntary. So in a sense, we do argue
in this study, and we're strong advocates for mandatory regulation.
(23:38):
If regulation needs to exist, it needs to be mandatory,
as voluntary regulations are like telling a kid that they
have homework but don't have to do it. It's voluntary.
So we think eventually, whatever the regulation is in place
(23:59):
is better to be mandatory. And even if it's a
certain level that is not the level that countries would
like to see have in their regulatory system, it's still
whichever level it is, it's better to be mandatory rather
(24:21):
than voluntary. So in other words, we think voluntary compliance
les enforcement. It's leading to inconsistent adoption of what the
requirements are, and the mandatory frameworks actually ensure accountability and
(24:42):
ensure alignment across jurisdictions.
Speaker 1 (24:46):
I know this is also interesting. I think just to
wrap up, my last question is what are some other
key studies or topics you've worked on or are working on,
just anything we can look forward to in terms of reading.
Speaker 3 (25:02):
Right. So, we do have a few studies in the
arena of using artificial intelligence to help with greenwashing detection
and to understand the discrepancies between specific documents. And this
(25:24):
is also something that we look at as being on
the same team with governments, corporations, academia altogether, because it's
not enough if governments create regulations and then companies struggle
with the regulations, they're costly and they're not mandatory, and
(25:48):
then academics do their research based on data that may
or may not be complete. I think it's important to
work together and actually help corporations in becoming compliant. If
I can take as an example, the environmental disclosure that
(26:12):
was recently proposed by the SEC. And I say recently,
but it's really I think three years ago, so it
is a step forward. And another aspect of the research
is looking into also ESG ratings for the companies, and
(26:38):
sometimes it is challenging to find out what the company
is really about if we just look at one number,
or even if we look at sub sections of environmental
ESG rating, and if we look at ten twenty subsections
or subtopics, it is challenging to understand whether the ratings
(27:05):
are really making sense. Because there are many reputable companies
that offer ESG ratings, including Bloomberg, sustain Aalytics, MSCI and
so on, and correlations between or among all these different
(27:28):
providers on the same company on ESG ratings for the
same company are really low, so it's very hard for
researchers to understand whom to trust. And this is not
to say that one company is better than the other
or they are not doing a good job. I think
they are, but there is still a big need for
(27:50):
standardization and for taking one step into a convergence in
a way on understanding, offering this to investors, offering better landscape,
less confusion even for corporations on how to report, what
to report, and better beginning of doing research that will
(28:17):
be useful for policy makers, so that academics are not
going to all look for different sources provide different results
and then it's becoming more rather than less confusing at
the end. So I think to summarize looking into standardization,
(28:37):
helping corporations report better, not necessarily more or less, and
standardizing the ESG ratings across the board.
Speaker 1 (28:49):
Thank you, Ireen. I think all of this just speaks
to the power of generative AI and how they can
make one allies more efficient hopefully, and I'm really looking
forward to the research on detecting greenwashing. So with that arena,
thank you so much for joining us.
Speaker 3 (29:06):
Thank you, thank you very much, Shahan and Rob. It
was a great pleasure to talking to you and wish
you all the best.
Speaker 1 (29:14):
Thank you and for our listeners. You can find more
information on sustainability issues on bispace esc go on the
Bloomberg terminal. If you have an ESG quandary you would
like to ask BI expert analysts or learn more about
our research, send us an email at ESG Currents at
bloomberg dot net. And if you like this episode, please
(29:36):
subscribe on Apple, Spotify, or your favorite podcast platform. Thank
you everyone, and have a great day.
ESG has become established as a key business theme as
companies and investors seek to navigate the climate crisis, energy transition,
social megatrons, mounting regulatory attention and pressure from other stakeholders.
The rapidly evolving landscape has become inundated with acronyms, buzzwords,
and lingo, and we aim to break these down with
(00:29):
industry experts. Welcome to ESG Currents, your guide to navigating
the evolving ESG space, one topic at a time, Brought
to you by Bloomberg Intelligence, part of Bloomberg's Research department,
with five hundred analysts and strategists working across all major
world markets. Our coverage includes over two thousand equities and
(00:49):
credit as well as outlooks on more than nineteen industries
and one hundred market indices, currencies and commodities. I'm Shin Contractor,
senior ESG analyst Bloomberg Intelligence.
Speaker 2 (01:01):
And I'm Robb Duboff, also senior ESG analyst at BI
and we are your hosts today.
Speaker 1 (01:07):
So today we're joined by Irena Vondeska, Professor and Director
of Finance Programs and Chair of the Administrative Science Department
at Metropolitan College Boston. University. Arena recently worked on a
study that helps navigate ESG regulations using AI, two topics
that I think are extremely timely. Now we all know
(01:30):
that ESG related regulations are complex to this to say
the least varied and ever changing. And also AI is
top of everyone's mind right now, so I'm really excited
to learn more about this study. Arena. Thank you for
joining us.
Speaker 3 (01:45):
Thank you so much, Shahin and Rob for inviting me
and for having me here.
Speaker 1 (01:51):
Thank you So, Arena, can you tell us more about
the study, what was the motivation behind doing it and
some of the key findings that you found.
Speaker 3 (02:01):
So, as you said in the introduction, Shahin, the ESG
field or sustainability field in the corporate arena has become
a very confusing field, quite difficult to follow with all
the regulations that are upcoming. Some of them are mandatory,
(02:21):
some of them are voluntary. Globally, it has become a
real alphabet soup of documents that professionals need to follow
continuously look after new documents, and some countries do establish
rules and regulations that impose mandatory disclosures, for example in
(02:48):
the environmental area and also in the social and governments area,
So we looked into started looking into some studies and
also documents and realize that there are way over possibly
(03:09):
more than one hundred global ESG related regulatory documents that
we could analyze. There are many more, but we did
beke a selection using a hybrid model to combine traditional
keyword based searches with advanced AI power or artificial intelligence
(03:33):
powered natural language processing to understand whether we can create
a set of global documents that we can explore and
understand what do they cover, which areas within the ESG
field are mostly covered and the study initially analyzed the
(03:59):
documents to establish a baseline before using artificial intelligence for
more scalable and comprehensive analysis of these global documents. The
artificial intelligence model identified alignments of the ESG related regulatory
(04:20):
documents with the SaaS B topics, and these SASB topics
address areas for corporations that are relevant for disclosure for
specific industries, and the main results showed that once we
(04:46):
divided the regulations in mandatory or voluntary regulations, we saw
that the European Union came up as a leader in
effective and mandatory regulation coverage, addressing twenty out of the
(05:06):
twenty six SASBY topics, followed by Nigeria and Saudi Arabia
that addressed nineteen topics each and for non mandatory but
effective regulations, the US ranked highest, covering twenty one topics
from the SASB Materiality Map, and Malaysia and Mexico were
(05:32):
close behind with twenty topics covered by each country.
Speaker 2 (05:40):
So how do you define what makes an effective regulation?
Can you maybe give some examples?
Speaker 3 (05:45):
Effective regulation in this study means the definition is that
it meaningfully the regulation meaningfully aligns with one or more
of the twenty six SASB ESG topics. So effectiveness was
(06:07):
measured for example, whether the regulation clearly addresses the ESG
theme in a substantive way, it's clearly related to the theme,
whether it was currently enforce and whether the particular regulation
promoted compliance or reporting. So examples include the European Union
(06:36):
Corporate Sustainability Reporting Directive or the CSRD, Nigeria's mandatory sustainability
disclosures for listed firms and sold Arabia's comprehensive ESG requirements
for institutional investors.
Speaker 1 (06:57):
And it's really interesting to see country like Nigeria saldiri
militias the ones you mentioned. I guess a few ways
to categorize these results in these regulations by team relating
to a team's climate, social, et cetera. What were some
that were dominant?
Speaker 3 (07:18):
So, first, thank you for that question. And it's important
to understand the categorization of regulations and the dominant themes
and also some regional variations in the themes. And yeah,
the regulations were categorized using the saasby five major themes,
(07:41):
which is the environmental theme, the social capital, human capital,
business model, and innovation and leadership and governance. So the
most dominant topics within these themes were greenhouse gas emissions,
(08:03):
ecological impacts, and energy management. And less commonly addressed topics
within the documents were customer welfare for example, or competitive behavior.
And when we look into the environmental topics and the
(08:28):
dominant countries that surfaced were the European Union, the US
in coverages, and Nigeria and Saudi Arabia.
Speaker 2 (08:40):
Interestingly, yeah, I think that, I mean, it definitely makes
sense from what we see that there's a lot of
regulation on the environmental side, but I'm curious on the
social side. You mentioned that probably didn't have as much
strength and regulation, But were there any countries that maybe
stood out when it comes to regulation on the social pillar?
Speaker 3 (09:00):
All right, on the social regulation side, we looked into
the social capital and human capital. So if we look
in the mandatory regulation coverages, Saudi Arabia, Hong Kong and
Indonesia scored the highest on social capital with coverage scores
(09:24):
of five and they were closely followed by Nigeria, Singapore
and European Union each scoring four, and they were also
highest in human capital. The ones that surfaced were against
Singapore that we mentioned, but also in the human capital
(09:47):
we saw Indonesia, Nigeria and the European Union and the UK.
So these indicates a notable emphasis on community engagement, customer relations,
social cohesion in these countries regulations. And this is on
(10:09):
the mandatory regulatory coverage. So when we looked at which
countries had the voluntary regulatory coverage on social capital, highest
coverage scores were achieved by Malaysia scoring six, followed by
Australia with five and US with four. Mexico and Ghana
(10:33):
also with four.
Speaker 2 (10:36):
So were there any countries that maybe were surprising something
you expected to see but didn't or vice versa.
Speaker 3 (10:43):
Yes, actually two main surprises, I would say, if I
can single them out, were the Netherlands and Switzerland. And
so they were notable outliers because despite their reputation for
(11:03):
progressive policy, they had very low coverage in global non
mandatory is gregulatory documents. So Netherlands, for example, had total
coverage in six topics in the non man regulatory space
with zeros for social and human capital, and Switzerland had
(11:28):
only two regulatory documents. It covered two topics in the
environmental category and also zeros for social and human capital.
So we were obviously asking the question why is this
what we observe? And potentially the reason could be we
(11:52):
noticed that much of their regulation and disclosures occur locally
or through other bodies, and hence it did not appear
in global regulatory trackers which we used for this study.
Speaker 1 (12:10):
And you know, one of the interesting findings or results
was that these chatchipt models were more successful and detected
a greater number of regulations versus a keyword base model.
Now why do you think that is? And were you
able to quantify how much more successful the chat ChiPT
(12:30):
model was just curious us to those two aspects.
Speaker 3 (12:35):
Yes, so in this era of by day development of
not just new technology or using generative AI for many tasks.
We can imagine that if we continue to do this study,
or if we repeat it with the new documents, that
(12:57):
we will actually be able to capture even the development
in the area. So we're able to capture the change
over time or emphasize the role of regulations using this
fast way of retrieving the documents and then the new documents.
So this is what we call the artificial intelligence enhanced
(13:21):
retrieval using models like the retrieval outvented generation. So these
models integrate external data sources to improve contextual and factual accuracy.
And this continuous application of the method could allow for
(13:45):
real time updates and trend analysis across regions and topics.
So when you ask why we're cht GPT models more successful,
everything I said is so much more difficult to imagine
being done by just a human in real time because
it does take time to sift through the documents, to
(14:07):
read the documents, to understand them, to understand the nuances.
So the CHGPT based models outperformed the keyword searches because
they understand context and nuance, not just specific words. They
can detect implicit references to ESG topics, even when terminology
(14:32):
may vary and it's differently presented. And then when combined
with the retrieval augmented generation, they leverage external information to
validate the findings, offering more comprehensive and accurate mapping. So
just to give you a bit of an idea on
(14:53):
the example of a retrieval augmented generation and how powerful
it is, this is the retrieval augmented generation or the
RUG is an advanced natural language processing technique that combines
two powerful components. One is information retrieval, finding the relevant
(15:16):
external knowledge and second is the text generation or creating
human like responses based on that knowledge. So how does
it work? For example, if we have a Q query
and we say what are the latest ESG regulations in Europe,
(15:36):
so the retriever step will instead of relying only on
pre trained knowledge, the retrieval augmented generation uses the retriever
usually a vector database or search engine, to fetch relevant
documents or passages from an external knowledge sources I like
(16:00):
database or ESG regulations, and then it passes it to
the next step or the generator step. A large language
models like the GPT which reads both the query and
the retrieve documents and it generates a coherent, grounded answer
(16:22):
using the up to date and specific information. So it's
important to note why is it useful. It's because it
is able to retrieve up to date information. Unlike static
models that can only know what they were trained on
RAG or, the retrieval augmented generation allows real time integration
(16:43):
of current facts and in regard to the domain specific accuracy,
it can incorporate specialized content like laws, financial disclosures or
ESG frameworks that general model might not fully capture. And
most importantly, it does reduce hallucinations. By grounding the answers
(17:09):
in real documents, the retrieval augmented generation helps minimize the
risk of the model making things up.
Speaker 1 (17:21):
That's all fascinating. I think you know you spoke to
the power of generative aim. I'm now curious what are
the uses of AI for some other things in the
field of sustainability. What have you found it the most
useful in terms of in terms of your research?
Speaker 3 (17:40):
Right, So, as I said, these models are quite powerful,
but I would be amiss not to mention actually their
limitations as well. So despite the strengths they are notable limitations,
so we still need to be very careful about by
(18:00):
in the data, potentially reflecting skewed global narratives for example,
or challenges in interpreting vague or unstructured content, so we
could have potential false positives where irrelevant documents are matched
(18:21):
due to semantic similarity, or we can have difficulties in
handling language specific documents unless multilingual models are applied. So,
having said they have quite a big power and they
do have limitations, the use of AI is not limited
(18:46):
to obviously just this study. It really enters many areas,
but related to ESG and going a step further, we
could actually use AI in a broader sense in sustainability research,
(19:07):
particularly in potentially detecting greenwashing by comparing corporate ISG claims
for example independent news sources. Or we can use it
to map sentiments and consistency across stakeholder communications. Because this
(19:32):
is all texts, the natural language processing is becoming more
powerful and then we can analyze various documents. In addition
to regulations, we can analyze corporate disclosures. We can analyze
press releases news articles to reveal potential discrepancies or emerging
(19:54):
trends at a corporate level.
Speaker 2 (19:58):
So it's interesting you talk about some limitations with AI.
I wonder if there's any thoughts you have for filers
or regulators how they might craft some of these reports
so that maybe they're more friendly for an AI model
to reader, or maybe conversely, is there a risk that
maybe they can gain the AI systems by writing these
reports in certain ways or narratives in certain ways.
Speaker 3 (20:21):
So that's a very interesting question, and we obviously need
to be mindful of that, because it is possible to
write certain reports in a way to pass a threshold
and look more esg like or as if the report
(20:46):
is really positive about the company. So we also need
to be mindful of not just the way the reports
are written, but the words that are used, the sentiment
that is used in the documents. And it is possible
to create documents that may just be documents and not
(21:10):
necessarily reflect the reality. But this is why I'm saying
that we should not be looking only into a certain
type of documents produced by the corporation or the company.
We really need to look for other types of documents
as well to compare and extract better knowledge or better
(21:35):
intelligence as to the real state of the ESG or
ESG dedication of companies. So and also, the AI is
here and it isn't meant to replace the critical thinking
(21:56):
or the judgment or the expertise of the human ESG analyst.
So we're not advocating, we're not advocating that the human
is replaceable. Instead, we are advocating that the AI is
(22:16):
intended here to be really powerful tool and to help
the analysts, to help ESG analysts to augment their capabilities.
So at the end of the day, we believe that
the human needs to be definitely in the loop and
bring that final conclusion to the table of what really
(22:42):
is read, what is understood by the documents, by the comparison,
by the analysis, and to make that final determination of
the result.
Speaker 2 (22:53):
So hopefully Shaheana and I will still have jobs, but.
Speaker 3 (22:57):
You will be more successful in your.
Speaker 2 (22:59):
Jobs, let's hope. I just want to circle back again
on this idea of regulation. So you know, governments are
starting to work to refine regulations. What would your advice
be to them? Is there anything your study can show
about how to make regulations more effective?
Speaker 3 (23:15):
So it's interesting to think of regulations, as I mentioned
in our study we looked into what is mandatory and
what is voluntary. So in a sense, we do argue
in this study, and we're strong advocates for mandatory regulation.
(23:38):
If regulation needs to exist, it needs to be mandatory,
as voluntary regulations are like telling a kid that they
have homework but don't have to do it. It's voluntary.
So we think eventually, whatever the regulation is in place
(23:59):
is better to be mandatory. And even if it's a
certain level that is not the level that countries would
like to see have in their regulatory system, it's still
whichever level it is, it's better to be mandatory rather
(24:21):
than voluntary. So in other words, we think voluntary compliance
les enforcement. It's leading to inconsistent adoption of what the
requirements are, and the mandatory frameworks actually ensure accountability and
(24:42):
ensure alignment across jurisdictions.
Speaker 1 (24:46):
I know this is also interesting. I think just to
wrap up, my last question is what are some other
key studies or topics you've worked on or are working on,
just anything we can look forward to in terms of reading.
Speaker 3 (25:02):
Right. So, we do have a few studies in the
arena of using artificial intelligence to help with greenwashing detection
and to understand the discrepancies between specific documents. And this
(25:24):
is also something that we look at as being on
the same team with governments, corporations, academia altogether, because it's
not enough if governments create regulations and then companies struggle
with the regulations, they're costly and they're not mandatory, and
(25:48):
then academics do their research based on data that may
or may not be complete. I think it's important to
work together and actually help corporations in becoming compliant. If
I can take as an example, the environmental disclosure that
(26:12):
was recently proposed by the SEC. And I say recently,
but it's really I think three years ago, so it
is a step forward. And another aspect of the research
is looking into also ESG ratings for the companies, and
(26:38):
sometimes it is challenging to find out what the company
is really about if we just look at one number,
or even if we look at sub sections of environmental
ESG rating, and if we look at ten twenty subsections
or subtopics, it is challenging to understand whether the ratings
(27:05):
are really making sense. Because there are many reputable companies
that offer ESG ratings, including Bloomberg, sustain Aalytics, MSCI and
so on, and correlations between or among all these different
(27:28):
providers on the same company on ESG ratings for the
same company are really low, so it's very hard for
researchers to understand whom to trust. And this is not
to say that one company is better than the other
or they are not doing a good job. I think
they are, but there is still a big need for
(27:50):
standardization and for taking one step into a convergence in
a way on understanding, offering this to investors, offering better landscape,
less confusion even for corporations on how to report, what
to report, and better beginning of doing research that will
(28:17):
be useful for policy makers, so that academics are not
going to all look for different sources provide different results
and then it's becoming more rather than less confusing at
the end. So I think to summarize looking into standardization,
(28:37):
helping corporations report better, not necessarily more or less, and
standardizing the ESG ratings across the board.
Speaker 1 (28:49):
Thank you, Ireen. I think all of this just speaks
to the power of generative AI and how they can
make one allies more efficient hopefully, and I'm really looking
forward to the research on detecting greenwashing. So with that arena,
thank you so much for joining us.
Speaker 3 (29:06):
Thank you, thank you very much, Shahan and Rob. It
was a great pleasure to talking to you and wish
you all the best.
Speaker 1 (29:14):
Thank you and for our listeners. You can find more
information on sustainability issues on bispace esc go on the
Bloomberg terminal. If you have an ESG quandary you would
like to ask BI expert analysts or learn more about
our research, send us an email at ESG Currents at
bloomberg dot net. And if you like this episode, please
(29:36):
subscribe on Apple, Spotify, or your favorite podcast platform. Thank
you everyone, and have a great day.
Host
Eric Kane
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Special Summer Offer: Exclusively on Apple Podcasts, try our Dateline Premium subscription completely free for one month! With Dateline Premium, you get every episode ad-free plus exclusive bonus content.
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!