00:00:00
[Music]
00:00:17
well uh in the Telco context uh since
00:00:19
we're shifting towards more
00:00:21
software-based uh
00:00:23
telecommunications uh AI will even be
00:00:26
more uh prominent so we'll discuss a a
00:00:29
fundament Al concern in uh in
00:00:33
telecommunications in our Modern Life
00:00:35
and that is uh AI uh and uh when when we
00:00:39
talk about technology inevitably we'll
00:00:41
talk about its proper use what it's
00:00:44
really about how we uh see ourselves uh
00:00:48
the lives we want to lead and that is uh
00:00:52
falling within the perview of AI ethics
00:00:55
and uh
00:00:57
governance so uh as you can see even for
00:01:01
those School based uh we're pretty much
00:01:03
uh inundated by uh good news and bad
00:01:07
news with AI uh good use and misuse uh
00:01:11
even scientists are not spared uh I see
00:01:14
that my colleagues for instance are
00:01:15
already using uh Chad GPT to create quiz
00:01:20
uh uh if you are a student please raise
00:01:22
your hand uh if you're still a
00:01:26
student so you have a sense of uh so
00:01:29
it's not as if your only students would
00:01:32
use AI to uh to get past exams or to try
00:01:38
to uh overcome uh requirements or yeah
00:01:41
fulfill requirements also teachers so um
00:01:45
as you can see but but
00:01:48
for um the for the Philippines uh as a
00:01:52
as a country we are small country
00:01:54
relative to uh India United States we're
00:01:57
about 110 111 million yeah as you can
00:02:00
see we're pretty heavy users of AI now
00:02:04
what do you think is that good or bad as
00:02:06
you can see Japan uh small use and then
00:02:10
some others and even Indonesia which is
00:02:13
bigger than us 350 million we outrank
00:02:16
Indonesia in terms of traffic on chat
00:02:19
GPT uh but if you probe deeper uh the
00:02:22
problem here is that we're using uh AI
00:02:25
to search for information especially
00:02:28
chat gbt which is may not which may not
00:02:31
be the proper use of of
00:02:34
AI all right
00:02:38
so so what is AI it is a field to uh
00:02:42
dedicate to developing systems capable
00:02:44
of Performing tasks and solving problems
00:02:46
associated with human
00:02:47
intelligence most if not all systems
00:02:50
that make decisions normally recording
00:02:52
human expertise fall within the purview
00:02:55
of of AI there's also a conflation of
00:02:58
terms uh data science it looks like uh
00:03:02
uh be becoming less sexy a discipline
00:03:05
because of of this conflation but
00:03:07
there's an overlap actually you need a
00:03:10
robust uh understanding of data
00:03:13
scientific understanding of data to be
00:03:15
able to do Ai and uh you have machine
00:03:18
learning at more deeper uh more more
00:03:22
deeply level and you have deep learning
00:03:25
uh AI especially with the models we're
00:03:27
dealing with now which is which are
00:03:29
really is artificial neural Nets no the
00:03:31
ones popular anyway like chat PT but uh
00:03:36
it's not just one area uh there's
00:03:39
there's more uh natural language
00:03:42
processing uh knowledge representation
00:03:44
machine learning computer vision uh
00:03:46
speech recognition Robotics and the
00:03:49
challenge really is to combine all these
00:03:52
and to have a a singular uh contiguous
00:03:55
uh uh Services no uh of of AI so uh how
00:04:01
do we deal with that we will be talking
00:04:04
about the principles that govern AI
00:04:07
right today we embark on a journey for
00:04:09
the values informing the future of AI
00:04:12
before we begin let's reflect on a real
00:04:14
life story that highlights the
00:04:16
importance of ethical principles and
00:04:18
considerations in AI in 2018 Amazon
00:04:21
developed an AI powered recruiting tool
00:04:23
to assist with hiring the tool was
00:04:25
designed to scan resumes and identify
00:04:27
the most qualified candidates however
00:04:29
ever it was later discovered that the
00:04:31
tool was biased against female
00:04:33
candidates the reason it was trained on
00:04:36
resumes submitted to Amazon over the
00:04:38
past 10 years which were predominantly
00:04:40
from male applicants as a result the
00:04:43
system learned to favor male candidates
00:04:45
and downrank rums with words commonly
00:04:48
used by women with that in mind let's
00:04:51
take a look at the principles that
00:04:53
hopefully can help us be fair and
00:04:56
develop AI to serve our best goals and
00:04:59
aspirations
00:05:00
as a people a report on an AI
00:05:03
development framework available at
00:05:07
ai.org
00:05:09
theframe offers a set of value based
00:05:12
guidelines covering inclusive growth
00:05:15
human centered values transparency
00:05:18
robustness and accountability these
00:05:21
principles are the foundation of
00:05:23
responsible AI development I strongly
00:05:26
suggest that you check out this live
00:05:28
online doent for a detailed discussion
00:05:32
of today's topic principle one inclusive
00:05:35
growth sustainable development and
00:05:37
well-being artificial intelligence plays
00:05:39
a crucial role in sustainable
00:05:41
development intertwined with our
00:05:42
national goals for inclusive growth and
00:05:45
well-being as countries Embrace AI it is
00:05:47
essential to consider both its
00:05:49
advantages and risk mitigating potential
00:05:51
negative effects is vital ensuring AI
00:05:54
benefits are shared equitably across
00:05:56
Society principle two human centered
00:05:59
values and fairness fairness is a
00:06:01
Cornerstone of AI bias in AI systems can
00:06:04
lead to discriminatory outcomes
00:06:06
affecting various sectors in society
00:06:08
defining and evaluating fairness in AI
00:06:10
is a challenge but we must ensure AI
00:06:13
respects human rights and data privacy
00:06:15
rights instead of relying solely on AI
00:06:18
robots or automation it's essential to
00:06:20
involve humans directly especially for
00:06:22
high-risk systems while AI can offer
00:06:24
innovative solutions human participation
00:06:27
remains crucial to ensure these systems
00:06:29
enhance human capabilities rather than
00:06:31
causing harm ai's potential for
00:06:34
Innovation is Limitless but it also
00:06:36
opens doors to potential misuse ensuring
00:06:39
fairness in the development of AI is
00:06:41
challenging but our stakeholders argue
00:06:43
that end users should have transparency
00:06:46
into ai's decision-making process and
00:06:48
the ability to influence results in some
00:06:50
cases human involvement is necessary to
00:06:53
avoid purely algorithmic decision making
00:06:56
ensuring clear human accountability and
00:06:58
system audit ability however it's
00:07:00
essential to recognize that autonomous
00:07:03
systems may not always be under human
00:07:05
control to some degree therefore we must
00:07:08
qualify human involvement in AI systems
00:07:11
particularly in high-risk applications
00:07:13
in such cases having humans in the loop
00:07:16
HL is crucial for high-risk AI systems
00:07:20
the EU AI act mandates human oversite to
00:07:24
ensure safe and responsible use human
00:07:26
involvement in AI goes beyond hit a
00:07:30
successful approach involves leveraging
00:07:32
both human and machine competences in a
00:07:34
virtuous cycle to produce valuable and
00:07:37
positive outcomes at its core AI must
00:07:39
prioritize the protection of Human
00:07:42
Rights principle three robustness
00:07:45
security and safety building trust in AI
00:07:48
requires us to prioritize robust secure
00:07:51
and Safe Systems whether it's
00:07:53
self-driving cars or medical
00:07:54
applications reliability is of utmost
00:07:57
importance to ensure safety standards
00:07:59
and protect human rights adequate
00:08:02
regulations and oversight play a vital
00:08:04
role while it's essential to acknowledge
00:08:07
that the majority of AI systems deployed
00:08:09
so far are largely safe it's
00:08:12
understandable that people might get
00:08:14
fixated on the more dramatic incidents
00:08:17
for instance earlier this year there was
00:08:19
a tragic incident involving a Belgian
00:08:21
man who reportedly engaged in a six week
00:08:24
long conversation with an AI chatbot
00:08:27
called Eliza about the ecological future
00:08:29
of the planet the chatbot supported his
00:08:32
Echo anxiety and tragically encouraged
00:08:34
him to take his own life to save the
00:08:37
planet instances like this remind us of
00:08:40
the responsibility we hold as AI
00:08:42
developers to prioritize safety and
00:08:44
well-being recently the launch of open
00:08:47
AI chat GPT language model stirred mixed
00:08:51
reactions this model showcased its
00:08:54
ability to mimic human conversations and
00:08:56
generate unique text based on users
00:08:58
prompts however this has also raised
00:09:00
concerns about potential misuse or
00:09:03
unintended consequences moving forward
00:09:06
it is crucial for AI developers to
00:09:08
strive for continuous Improvement in
00:09:10
making their products and services safer
00:09:13
to use by emphasizing robustness
00:09:15
security and safety we can Foster Public
00:09:18
trust and ensure that AI technology is a
00:09:21
Force for good in our
00:09:23
lives principle four
00:09:26
accountability a actors must be
00:09:28
accountable for their actions and
00:09:29
decisions responsible AI involves
00:09:32
transparency and ability to explain the
00:09:34
reasoning behind AI system choices
00:09:37
auditability helps ensure compliance
00:09:39
with regulations and mitigates potential
00:09:41
risk associated with AI the risk of
00:09:44
thisinformation has gained prominence
00:09:46
recently with the Advent of chat GPT and
00:09:49
generative AI consider the case of Brian
00:09:52
Hood an Australian mayor H was a
00:09:57
whistleblower praised for showing trend
00:09:59
discourage by exposing a worldwide
00:10:01
bribery Scandal linked to Australia's
00:10:04
national Reserve Bank however his voters
00:10:07
told him that chat GPT named him as a
00:10:10
guilty party and was jailed for it in
00:10:13
such a bribery scandal in the early
00:10:15
2000s should open AI the company behind
00:10:18
chat GPT be held responsible for such
00:10:21
apparent disinformation and reputational
00:10:23
harm even it could not possibly know in
00:10:26
advance what their generative AI would
00:10:28
say the question of whether open AI
00:10:31
should be responsible for this is a
00:10:33
complex one on the one hand open AI
00:10:35
could argue that it is not responsible
00:10:38
for the content that its AI system
00:10:40
generates on the other hand open AI
00:10:43
could also be seen as having a
00:10:45
responsibility to ensure that its AI
00:10:47
system is not used to spread this
00:10:51
information principle five transparency
00:10:53
explainability and traceability
00:10:55
transparency in AI policies and
00:10:57
decisions is vital for a democratic
00:10:58
societ
00:11:00
understanding AI systems even for
00:11:01
non-technical stakeholders Foster truth
00:11:04
and informed decision making
00:11:05
explainability allows us to identify
00:11:07
potential biases and ensure Fair AI
00:11:10
outcomes in Singapore it's required that
00:11:12
AI decisions and Associated data can be
00:11:15
explained in non-technical terms to end
00:11:17
users and other stakeholders this
00:11:19
openness promotes informed public debate
00:11:21
and Democratic legitimacy for AI however
00:11:24
the concern of AI systems being
00:11:26
perceived as black boxes lacking
00:11:28
transparency and explainability has been
00:11:29
raised during our stakeholder
00:11:31
consultations AI systems navigate
00:11:33
through billions trillions of variables
00:11:36
that influence outcomes in complex ways
00:11:38
making it challenging to comprehend even
00:11:40
with human attention large language
00:11:41
models like chat GPT with trillions of
00:11:44
parameters have made explainability
00:11:46
elusive even to their own developers
00:11:49
nonlinear models further complicate
00:11:51
understanding the connection between
00:11:53
inputs and outputs despite these
00:11:55
challenges developers are working on
00:11:57
Solutions more interpret able models
00:12:00
like decision trees and rule-based
00:12:01
systems are being explored techniques
00:12:04
such as human readable rule extraction
00:12:06
sensitivity analysis and localized
00:12:07
explanations are also enhancing
00:12:09
explainability additionally detailed
00:12:11
documentation of model architecture
00:12:13
training data and evaluation metrics and
00:12:15
provide valuable insights into AI system
00:12:18
Behavior regarding transparency some
00:12:20
stakeholders propose focusing on
00:12:22
policies and processes rather than
00:12:24
revealing AI algorithms entirely this
00:12:27
approach acknowledges potential risk as
00:12:29
excessive transparency might hinder
00:12:31
Innovation by diverting resources from
00:12:33
improving safety and performance as the
00:12:36
European Union moves towards adopting
00:12:38
the AI act there's another important
00:12:40
principle linked to transparency called
00:12:42
traceability traceability is distinct
00:12:44
from explainability but equally
00:12:46
significant while explainability focuses
00:12:48
on understanding how an AI system works
00:12:50
traceability involves actively tracking
00:12:52
its use to identify potential issues
00:12:55
this empowers AI system operators to
00:12:57
spot and address risk like data bias and
00:13:00
coding errors achieving traceability
00:13:02
means keeping records of the data used
00:13:04
the decisions made and the reasons
00:13:06
behind them explainability on the other
00:13:08
hand plays a critical role in building
00:13:10
user trust and aiding informed decision
00:13:12
making it provides a human readable
00:13:14
explanation of how an AI system makes
00:13:16
decisions both traceability and
00:13:18
explainability contribute to the broader
00:13:20
principle of transparency however it's
00:13:22
important to recognize that transparency
00:13:24
alone may not automatically build public
00:13:27
trust Professor Anor O'Neil highlighted
00:13:29
this concern in her BBC right lectures
00:13:31
two decades ago noting that while
00:13:33
transparency and openness have advanced
00:13:36
they have not done much to build public
00:13:38
trust in fact trust may have even
00:13:40
diminished as transparency increased
00:13:42
this Insight remains relevant in today's
00:13:44
discussions about Ai and H regulations
00:13:47
some stakeholders propose focusing on
00:13:50
policies and processes rather than
00:13:51
revealing
00:13:53
AI systems navigate through billions
00:13:55
trillions of variable this approach
00:13:58
acknowledges potential r as excessive
00:14:00
transparency might hinder Innovation by
00:14:02
diverting resources from improving
00:14:04
safety and performance as the European
00:14:07
Union moves towards adopting the AI act
00:14:09
there's another important principle
00:14:11
linked to transparency called
00:14:12
traceability traceability is distinct
00:14:15
from explainability but equally
00:14:16
significant while explainability focuses
00:14:18
on understanding how an AI system works
00:14:20
traceability involves actively tracking
00:14:23
its use to identify potential issues
00:14:25
this empowers AI system operators to
00:14:27
spot and address risk like data bias and
00:14:30
coding errors achieving traceability
00:14:32
means keeping records of the data used
00:14:34
the decisions made and the reasons
00:14:36
behind them explainability on the other
00:14:38
hand plays a critical role in building
00:14:40
user trust and aiding informed decision
00:14:42
making it provides a human readable
00:14:44
explanation of how an AI system makes
00:14:46
decisions both traceability and
00:14:48
explainability contribute to the broader
00:14:50
principle of transparency however it's
00:14:52
important to recognize that transparency
00:14:55
alone may not automatically build public
00:14:57
trust professor O'Neil highlighted this
00:15:00
concern in her BBC right lectures two
00:15:02
decades ago noting that while
00:15:04
transparency and openness have advanced
00:15:06
they have not done much to build public
00:15:08
trust in fact trust may have even
00:15:10
diminished as transparency increased
00:15:13
this Insight remains relevant in today's
00:15:15
discussions about Ai and H regulations
00:15:18
principle six trust trust is a crucial
00:15:21
element in AI adoption AI systems must
00:15:24
prove themselves to be reliable and safe
00:15:26
especially in applications impacting
00:15:28
lives livelihoods earning trust requires
00:15:31
adherence to high standards and
00:15:33
inclusive AI governance we now know that
00:15:36
transparency does not automatically
00:15:38
translate to trust we need trust to
00:15:40
provide space for our Filipino AI
00:15:42
developers to pursue Innovation that
00:15:45
benefit Society in turn they have to act
00:15:47
responsibly and be trustworthy AI
00:15:50
research is a public good that needs to
00:15:53
be supported by all
00:15:55
stakeholders this is where my
00:15:57
presentation ends even as we all
00:15:59
continue with our journey through AI
00:16:02
principles for more details check out
00:16:04
our report on AI governance framework
00:16:06
for the Philippines available at ai.org
00:16:12
theframe the values and principles we
00:16:14
discuss today are the compass guiding AI
00:16:17
featured let's continue to develop AI
00:16:19
responsibly ensuring it benefits
00:16:21
everyone while respecting human rights
00:16:24
and promoting a fair and Equitable
00:16:26
Society
00:16:30
all right
00:16:33
uh so uh just uh run through some of the
00:16:37
points made there uh inclusive growth
00:16:40
sustainable
00:16:41
development uh and uh well-being we see
00:16:44
that there is uh this is something that
00:16:46
is embedded in our Philippine Innovation
00:16:49
act uh a burdens and benefits have to be
00:16:52
shared uh equitably we also see how AI
00:16:55
could potentially bring in uh trill ions
00:16:59
of uh of economic activity uh trillions
00:17:02
of uh of benefits uh valued at trillions
00:17:07
of of dollars uh we're also seeing uh
00:17:10
70% of companies would have uh adopted
00:17:14
at least one type of AI technology uh
00:17:16
right now for the boo industry for
00:17:18
instance 60% the last survey are already
00:17:22
a uh using AI so if you're headed to Bo
00:17:24
most likely you'll be using Ai and uh
00:17:27
some other companies in the Philippines
00:17:28
as well uh there is increased uh
00:17:31
productivity that's why in my workplace
00:17:34
um it's default that uh my my staff
00:17:37
would be using AI so that uh the burden
00:17:41
I mean the justification would be on
00:17:42
people the Honus on them if they don't
00:17:46
uh use AI um however you see that um
00:17:51
it's not uh it's not something that is
00:17:53
straightforward it's easier said than
00:17:55
done um AI as a matter of fact will
00:17:59
potentially also bring in added
00:18:01
dimension of inequity um as opposed to
00:18:05
just simply a providing access to um say
00:18:09
internet so if you're in tawi tawi you
00:18:11
probably would experience uh internet
00:18:15
via uh by uh uh starlink and that's fine
00:18:18
and dandy however there's an other
00:18:21
dimension there that if you are uh going
00:18:23
to be using AI um there are going to be
00:18:27
additional skills that are expected of
00:18:29
you that are required of you algorithmic
00:18:32
skills uh the your ability to access
00:18:36
Fair datab basis and uh the um the
00:18:40
capacity to be treated fairly or to um
00:18:43
the the right to be treated fairly in
00:18:45
those databases and it's quite a leap
00:18:47
it's no longer just access to AI as as
00:18:50
you can see in the previous slides I had
00:18:52
I had a slide on Philippines being on
00:18:55
top of uh countries using uh chat GB
00:18:59
the problem with our use according to
00:19:02
data is that our we use chat GPT to look
00:19:05
for facts to look for uh certain
00:19:08
information uh and those um uh the
00:19:12
information the bits and pieces of
00:19:14
information could have been
00:19:15
hallucination so in other words our
00:19:18
usage of AI so far is uh shallow so so
00:19:22
that it's a problem when you have to uh
00:19:24
think in terms of inclusive growth
00:19:26
because even as we have access to AI
00:19:29
it's not just access we're talking about
00:19:31
it's about being able to access properly
00:19:34
and that requires more than just uh
00:19:37
access no um you also see that uh right
00:19:40
now ai is getting to be uh stale in some
00:19:44
areas uh it's uh it's people are um not
00:19:48
seeing beyond the hype um it it appears
00:19:51
that we have a peak in uh uh of there's
00:19:55
a peak already of inflated expectations
00:19:58
so it's a let down for others for
00:19:59
instance if they're expecting to AI to
00:20:02
do more uh so we could be seeing uh this
00:20:05
illusionment already and some are
00:20:08
enlightened hopefully that uh when we
00:20:11
truly understand AI we are experiencing
00:20:14
a plateau of productivity and this is
00:20:16
really where it matters most we see
00:20:18
beyond the hype and we go straight to uh
00:20:22
productivity uh in our workplace I see
00:20:25
this happening uh I'm not so sure in in
00:20:29
in other areas of of the country no so
00:20:33
uh as this has been emphasized earlier
00:20:35
as well uh human- centered values um
00:20:37
treating people fairly um avoiding
00:20:40
algorithmic decisions and their
00:20:42
discriminatory consequences so if you
00:20:44
look at uh algorithms they have the
00:20:46
tendency to perpetuate or if not
00:20:49
amplify uh existing social economic uh
00:20:53
and cultural
00:20:54
inequalities uh so that the idea really
00:20:57
is to really have
00:20:59
fairness uh and being respectful of
00:21:02
Human Rights and and data privacy um
00:21:06
all practically all disciplines all
00:21:09
professions are already affected you
00:21:10
might think that if you're a hairdresser
00:21:13
you would not be affected or a makeup
00:21:15
artist you would not be affected by AI
00:21:17
but as you can see in this uh in this
00:21:20
headline um and a a a makeup artist lost
00:21:23
his job uh assessing by with with AI
00:21:27
assessing his body her body language so
00:21:30
um it looks like there's no job anymore
00:21:33
that is safe from Ai No at least
00:21:35
directly or indirectly uh you see this
00:21:38
also uh some some countries um being
00:21:42
defensive about about AI uh um but that
00:21:45
is already reversed in Italy um they now
00:21:48
have access um uh to chat GPT there are
00:21:52
also certain areas of concern especially
00:21:54
when uh um open AI is uh I mean
00:21:57
introduces new uh new version of of chat
00:22:01
GPT uh you have uh relatively increased
00:22:04
risk as well no and in some companies
00:22:07
they are uh worry about intellectual
00:22:09
property being exposed to um some Trade
00:22:13
Secrets being exposed to to Ai and
00:22:17
meaning to the rest of the world as well
00:22:19
no so uh we have discussed this uh um uh
00:22:23
well enough in in the video but but just
00:22:26
to point out that this is an ongoing
00:22:28
concern uh every time you have a new
00:22:30
model of AI um there is increased
00:22:33
security as I said there is increased
00:22:34
safety as well even as you learn from
00:22:36
previous models because the more you
00:22:38
push the boundaries of AI the more you
00:22:41
are exposing yourselves actually to risk
00:22:45
accountability is something that is a
00:22:47
moving Target as well uh as as uh AI
00:22:51
progresses uh that is and as as new
00:22:54
domains of applications are being
00:22:56
considered new areas of
00:22:58
of expertise are being generated in AI
00:23:02
that is a continuing uh problem as uh
00:23:05
discussed earlier on transparency uh is
00:23:10
um is something that is almost
00:23:12
intractable to to some Regulators
00:23:15
because for the simple reason that
00:23:19
um systems uh tend to be blackboxes uh
00:23:23
and by transparency we mean um you know
00:23:26
operations as well of of AI
00:23:28
uh that may tend to be inexplainable so
00:23:32
uh neural nets for instance um there is
00:23:35
no um straightforward explanation why
00:23:38
input gets to have certain outputs for
00:23:41
instance no so uh the interaction of
00:23:44
with humans especially when you have uh
00:23:46
in learning context in the in in in in
00:23:49
relation to for instance CH gbt and
00:23:51
other large language models uh the more
00:23:54
you put in human elements the more
00:23:57
mysterious serous the outcomes become no
00:24:00
so that is a a problem no and then when
00:24:04
it comes to producing context uh you see
00:24:07
perhaps uh more recently in the
00:24:08
Philippines uh you see your messenger
00:24:12
having a a uh an AI uh uh tab already an
00:24:17
AI button uh where you can interact with
00:24:21
uh uh with an AI agent you see this uh
00:24:25
generating images uh that could
00:24:27
potentially be misused no I was uh uh
00:24:30
checking out for instance uh certain
00:24:33
images of Jose Rizal and uh um you know
00:24:38
uh combining him with uh with certain
00:24:40
scenarios and I could see potential for
00:24:43
uh for misuse as well no so let me just
00:24:47
uh um Breeze through these points
00:24:49
because uh we're running out of time uh
00:24:51
in a way I'll be sharing the the slides
00:24:54
with
00:24:55
you and just to point out that uh when
00:24:57
you talk about AI governance there are
00:25:00
many elements of there as well
00:25:02
leadership is one if you are in the
00:25:05
context of a company or School uh your
00:25:09
um overlords your bosses the board of
00:25:12
regions or trustees need to be really
00:25:14
engaged uh AI ethics to be front and
00:25:17
center looking at the core technical
00:25:19
elements of of AI this is not something
00:25:22
that you will just have to be left just
00:25:24
have to be left to the technical people
00:25:27
uh you see how this involves in an
00:25:29
organization more importantly you have
00:25:31
to consider the people of your uh in
00:25:33
your organization and the culture that
00:25:36
is uh um that is dominant in that
00:25:40
area you have to look at um risk in
00:25:44
terms of uh deciding go or no go for
00:25:48
certain AI operations uh looking at
00:25:50
operational structures and processes and
00:25:52
mechanisms as well uh especially how how
00:25:56
um how AI performs in in your
00:25:59
organizational context no so I will just
00:26:02
uh skip of uh this these elements and uh
00:26:05
leave this with uh with you um of the
00:26:08
link later on uh this has been alluded
00:26:11
to uh earlier on in in the in the
00:26:13
discussion so um some of one of the last
00:26:17
points I have to um to discuss with you
00:26:19
would be the human involvement uh
00:26:22
scenario or or consideration in AI
00:26:25
because if you come to think about it AI
00:26:28
is really about autonomy so um this is
00:26:32
an an area as well that is distinctive
00:26:35
of uh say simple data science um AI is
00:26:39
always about developing systems that are
00:26:42
aimed uh at becoming autonomous and if
00:26:46
you consider the notion of autonomy
00:26:49
actually uh by definition it's out of
00:26:52
control uh from humans or out of human
00:26:55
reach no even if you say oh uh I want
00:26:58
want to just insert myself there and uh
00:27:00
take over um over time you increasingly
00:27:04
lose control because your aim is to
00:27:07
develop autonomy in a way in in machines
00:27:11
no so there are um potentially
00:27:13
conflicting Tendencies there with human
00:27:15
control and autonomy so you have to be
00:27:18
qualifying what you really mean by by AI
00:27:22
autonomy because as you um as you
00:27:25
progress as technology progresses um
00:27:28
there is greater autonomy and therefore
00:27:30
less human control so the idea is um
00:27:33
especially for highrisk applications you
00:27:36
would need humans in the loop and that
00:27:37
is a concept that is uh hard to
00:27:40
operationalize actually because you have
00:27:42
a long continuous process um and some of
00:27:46
these are pretty boring and humans are
00:27:49
are terrible at dealing with boredom as
00:27:51
a matter of fact we uh um we try
00:27:56
everything we have everything to just uh
00:27:59
Escape boredom uh possibly including
00:28:02
from uh boring lectures no so prohibited
00:28:05
use of AI um when we talk about human
00:28:08
involvement we don't want AI to be
00:28:11
applied of as weapon systems uh New
00:28:14
Zealand is leading the way uh in in
00:28:19
advancing The View that uh we shouldn't
00:28:22
be um using Killer Robots uh AIS Killer
00:28:25
Robots we shouldn't be looking looking
00:28:28
at uh manipulation and exploitation uh
00:28:31
with the use of AI unfortunately some
00:28:33
countries this is more of a norm rather
00:28:35
than an exception uh in discriminate
00:28:38
surveillance there are societies that
00:28:39
are um
00:28:41
basically uh dominated by surveillance
00:28:44
Technologies um um cameras for instance
00:28:48
surveillance cameras and so on but even
00:28:50
if we might think that we are free there
00:28:53
is actually surveillance go going on
00:28:55
there is a book on surveillance
00:28:57
capitalism which is essentially
00:28:59
monetizing about monetizing our
00:29:02
activities online so if you use Facebook
00:29:04
if you use um um other social media
00:29:09
you're pretty much being monitored even
00:29:10
as you surf the Internet even as you
00:29:13
browse uh sites you are still being
00:29:16
surveilled at and the cookies will uh be
00:29:20
gathered and uh uh and and certain
00:29:23
patterns will be um uh will be
00:29:26
determined so so that when you and
00:29:28
probably even listening to you no
00:29:31
sometimes when you have conversation
00:29:32
with your friends about certain dress or
00:29:35
certain products you'll be surprised
00:29:37
sometimes that when you open your
00:29:39
internet uh when you open your browser
00:29:41
you see a a a an advertisement of the
00:29:45
product similar product that you are
00:29:47
interested in Social scoring is another
00:29:51
area where supposed to be prohibited but
00:29:53
this is happening in one country at
00:29:55
least uh when you are not doing well on
00:29:58
online if you are misbehaving online you
00:30:00
will not have your passport and you
00:30:02
cannot travel because uh you have very
00:30:05
low social score so so these
00:30:08
considerations um will have to be um put
00:30:12
front and center when we talk when we
00:30:14
talk about uh AI uh governance now so uh
00:30:19
there's there are risk profile in in
00:30:21
different areas of the society Criminal
00:30:23
Justice System Financial Services Health
00:30:25
and Social care social and digital media
00:30:29
uh energy and utilities there is a u an
00:30:32
accounting of of the risk that are
00:30:34
involved here uh although I think there
00:30:35
are variations uh when we come to the
00:30:38
Philippines or when we apply to the
00:30:40
Philippines for instance we have higher
00:30:41
risk of social media manipulations
00:30:43
during elections for instance in the US
00:30:46
now uh there is a uh there are
00:30:49
controversies around the use of certain
00:30:52
images uh use of centered synthetic data
00:30:56
uh and you can pretty much uh see how
00:30:59
how how they stuck up against other
00:31:01
other risk of of AI so we can knowing
00:31:05
this risk would be um would be a
00:31:08
prerequisite to being able to deal with
00:31:10
them no
00:31:12
so more uh bad news so to speak but we
00:31:16
already alluded to this earlier on
00:31:19
no uh in Southeast Asia uh there is U
00:31:22
it's changing now this is this was last
00:31:25
year but uh the recent initiatives they
00:31:28
want to come up with with AI regulation
00:31:30
AI but uh it's not happening anytime
00:31:33
soon uh my colleagues are participating
00:31:36
I think right now in uh Lao where this
00:31:39
is being discussed but uh uh I don't see
00:31:42
this happening the regulation of AI in
00:31:44
Southeast Asia in in two years not even
00:31:48
in two years because uh while there is
00:31:50
uh clamour there is stock uh it's a long
00:31:53
shot to get this uh um to get into some
00:31:57
kind of regul framework that is
00:31:59
applicable to all of Southeast Asia so
00:32:01
right now we're still pretty much a wild
00:32:03
wild west uh Philippines Pogo
00:32:07
situation uh Thailand Cambodia where
00:32:10
Filipinos are human traffic to serve uh
00:32:13
in the underbellies of AI in Thailand
00:32:17
and Cambodia we see that happening so
00:32:20
that is still a a problem now so um top
00:32:24
down approach may be a problem um we see
00:32:27
our Regulators who are so uh gangho
00:32:30
about regulating AI but my my discomfort
00:32:33
really is that uh they may have been
00:32:36
misinformed there is one uh uh one one
00:32:40
uh law lawmaker saying that uh AI
00:32:44
research needs to be regulated you need
00:32:46
to register your research in AI uh I
00:32:50
don't think that is a good idea uh so so
00:32:54
um I'm we're trying to reach out to that
00:32:56
regulator that uh at least provide him
00:32:59
with
00:33:00
proper um expertise no when it comes to
00:33:03
to AI there are many unintended anti
00:33:06
unanticipated consequences especially
00:33:08
for us Philippines we are very good at
00:33:11
crafting law without thinking about
00:33:14
their uh unintended unanticipated
00:33:17
consequences uh so we shoot ourselves in
00:33:19
the foot when we do regulation uh for
00:33:23
the for the reason that we lack
00:33:25
understanding of this technology
00:33:28
so um we have to look at various
00:33:31
Technologies to be able to see in
00:33:33
comparative terms of how this may pan
00:33:36
out how they're really properly
00:33:38
regulated when when uh when people think
00:33:41
about regulation they immediately think
00:33:43
of boss
00:33:45
so Congressman uh that may not be a uh a
00:33:50
good practice so you have to look at AI
00:33:52
as a range of uh of interventions when
00:33:56
you deal with AI so governance uh
00:33:58
regulation and legislation so they are
00:34:01
not the same so um there are
00:34:05
discriminatory biases uh if you until
00:34:08
now uh if you look at pronouns being
00:34:10
used by chat GPT there are stereotypes
00:34:12
that are being perpetuated so driver for
00:34:15
instance or scientist it's almost always
00:34:18
uh a guy so it's going to be a he but uh
00:34:22
the reality is that there are more women
00:34:25
uh scientists in some areas already uh
00:34:28
drivers are no longer just men and so on
00:34:31
and so forth no so biases are being
00:34:33
Amplified by by AI so we have to take a
00:34:36
look at our training data which is a
00:34:39
potential source of bias algorithmic uh
00:34:42
bias is also another possibility that
00:34:44
the way we parse data is biased already
00:34:47
no so um there's also the general aspect
00:34:51
of uh data Pat patrony so to speak uh do
00:34:55
we allow our national data to be fed to
00:34:58
the large language models of uh of open
00:35:03
open AI Microsoft Amazon because if that
00:35:06
is just what is going on uh then we're
00:35:09
pretty much really at the raw end of
00:35:11
what we call data colonialism uh as a
00:35:14
large contrast here is the effort for
00:35:16
instance of France they are trying to
00:35:18
come up with their own National large
00:35:20
language model based on Lama 3 and it's
00:35:23
an ongoing project uh for the government
00:35:26
of France uh precisely to combat what we
00:35:29
call uh data colonialism where where
00:35:32
French data or Filipino data uh would
00:35:35
just be um a training data for would
00:35:38
just be training data for large language
00:35:41
models that are owned by uh uh big Tech
00:35:44
no and there is no uh no conscious
00:35:48
effort to uplift uh the the um the
00:35:52
interest of the country so um very
00:35:56
quickly
00:35:57
uh you see that uh this is the
00:36:00
technology is really progressing by lips
00:36:02
and Bounds uh although I'm not going to
00:36:04
say that there's going to be General uh
00:36:07
intelligence
00:36:09
uh superhuman intelligence but we
00:36:12
already see that uh there is um greater
00:36:15
uh progress in this area we're now
00:36:18
approaching Lama 3 I think the one
00:36:19
you're using in your Facebook is already
00:36:21
Lama 3 point something but as you can
00:36:24
see the way it's by weeks uh it's been
00:36:26
estimated um that uh the compute
00:36:30
requirement for this because there's an
00:36:31
energy requirement for for compute and
00:36:34
that is the doubling time is 100 days so
00:36:36
if you're using 100 Watts now in 100
00:36:39
days uh just to power your AI you will
00:36:42
need uh uh 200 Watts uh for for for for
00:36:46
that Baseline now so um there are
00:36:50
benefits to this uh there are um UPS
00:36:53
there are um limitations we just have to
00:36:56
take a look at it but we we as as
00:36:58
Filipino researchers we have to be
00:37:00
trying this out we have to apply this in
00:37:02
our uh in our context that's why I'm
00:37:05
inviting you to the October 2425
00:37:09
conference uh I I put it in the chat and
00:37:13
finally uh so that we can talk about uh
00:37:16
application areas in in the Philippines
00:37:18
agriculture health and so on so how do
00:37:22
we deal with the respons uh with AI we
00:37:24
have to deal with AI responsibly looking
00:37:27
at legal and Regulatory Frameworks focus
00:37:29
on privacy fairness and Equity uh we
00:37:32
have to build local capacity in AI we
00:37:35
have to look at multi-stakeholder that's
00:37:38
why um it's a whole of society approach
00:37:41
uh we we look to Finland for instance
00:37:44
where they have a conscious effort to
00:37:46
educate their citizens at least 10% of
00:37:49
Vish citizens have undergone training in
00:37:53
AI at least uh familiarization with the
00:37:56
technology and Finland is now presenting
00:37:59
itself as the educator of the entire
00:38:02
Europe advocacy for greater
00:38:04
representation in global AI governance
00:38:06
we understand that we don't have the
00:38:07
compute uh right now I'm looking for 450
00:38:11
million so we can do an 8 node uh uh uh
00:38:17
compute for AI uh that is uh that is
00:38:21
really quite low but uh that's what what
00:38:24
what it really amounts to so if you have
00:38:26
400 50 million pesos uh that can help AI
00:38:30
research at least in my University uh
00:38:33
investment in AI enabled social research
00:38:35
to prioritize well-being and Equity um
00:38:38
this is not something that uh that is
00:38:40
just an afterthought right from the
00:38:42
get-go we have to design our systems to
00:38:44
produce well-being and uh equity