IEEE Technology and Society Magazine - March 2021 - 65

b) Scientific evidence: We select only use cases
were AI-based technologies could provide
and demonstrate reliable, repeatable, and
relatable results. In this pilot (see the " Use
case: AI for diabetic retinopathy " section), we
address an important health care problem.
c) Technology readiness levels (TRLs): We
selected AI technology at least TRL level 5.
In this case, AI for DR has TRL level 9, since
many companies have commercial products.
d) Ethics and data protection: We propose only
cases to obtain high social impact, and
where employment is guaranteed. Besides,
we are looking for a workflow without sensitive data. In the case of health applications,
it is essential to take into account the HIPAA
privacy rule and the seven main principles
proposed in [66]. We expect to include in
the final workflow a formal ethical protocol
for AI application such as IEEE P7000 [73].
In the case of the transparency and interpretability of the model, we present a preliminary
model where it is possible to highlight the
hallmarks and features of DR in the retina
image (see Figure 4 as an example).
e) Experts and stakeholders: In this case of DR,
we build an active community between retina specialists and AI experts to exchange
knowledge and complement our technical
resources. Also, we have the support of public policy experts and decision-makers.
* High-performance technical team collaboration.
To develop this project, we work with a high-performance technical team in data science and
DL. Besides, we collaborate with top-notch

Figure 4. Preliminary fully convolutional
model to detect DR features.

March 2021

researchers, experts, and international institutions such as IDB.
* Create new data sets: Although the government
generates a large amount of data, many of these
are not accessible and most of the time are not
labeled for supervised learning. To face this problem, we use public data (to apply transfer learning) and collaborate with the expert to label the
data. The collaboration and resources from IDB
are beneficial to generate reliable local data for
supervised learning.
* Infrastructure: Our team in Jalisco uses DL as the
main AI technique, and we need special computing infrastructure (GPU) to obtain the result
in a reasonable amount of time. In this project,
we spend more than 500 h of GPU-cloud computing, but we rely, for larger applications, on the
MareNostrum IV facilities at the Barcelona Supercomputing Center, in Spain.
The challenges of creating a culture of designing
a trustworthy, responsible, human-centric AI starts
with creating the appropriate and accessible common vocabulary. This challenge is even greater in
countries like Mexico where the accelerated adoption of AI technology is not associated with a complete understanding of the dangers it entails. In our
fAIr pilot, we promote the use of AI and other digital
technologies as an essential support for the protection of fair and equal treatment under the law of all
individuals. We stress the importance in the following sections.

Building a culture of AI responsible innovation
From our experience in the development of
the fAIr pilot in Jalisco, we believe that to build
and maintain a culture of responsibility during
the design, development, and deployment of an
AI project, it is crucial to follow a design methodology that includes the ethical issues from the
very beginning (see [3] and [74]). For us, it is
clear that for the development of a trustworthy
AI, we must ensure investment for good, fair, and
inclusive incubation to scale solutions that work.
In particular, one should make sure that the AI
project is the following:
* Ethically permissible: Designers have to consider
the impacts that the service(s) may have on the
well-being of affected stakeholders and communities. The actions performed by the system ought

https://www.nasa.gov/directorates/heo/scan/engineering/technology/txt_accordion1.html https://www.hhs.gov/hipaa/for-professionals/privacy/index.html

IEEE Technology and Society Magazine - March 2021

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - March 2021