The Artificial Intelligence (AI) Industry and regulators have to work hand-in-hand to ensure that AI technologies remain ethical and do not cause harm as more organizations undergo rapid digitalization, a leading AI expert said.
Dr. Charibeth Cheng, co-founder and research head of homegrown AI company Senti AI, said lawmakers should consult IT and AI practitioners when it comes to crafting bills to regulate the building of AI systems, especially before particularly risky technologies are deployed.
Dr. Cheng also serves as the associate sean of the College of Computer Studies at the De La Salle University.
“At some point, we should have some methodology or workflow for the approval of critical and risky AI products just like what we do with pharmaceuticals. When we have new medicine, it doesn’t go to the market immediately — it goes through several trials,” Dr. Cheng said.
“I would think that for the risky, very critical AI systems, there should also be that method or workflow prior to deployment just to make sure that we went all the checks prior to it [potentially] affecting negatively the society.”
The previous year saw organizations all over the world rely heavily on digital channels and AI technologies to adapt to the changing business landscape brought on by the COVID-19 pandemic.
The Philippines is no exception despite the lack of laws set in place to regulate the industry.
No system in place
According to Oxford’s Government AI Readiness Index 2020, the Philippines ranked 74th out of 172 countries. The country also ranked 9th out of 15 countries at the regional level. In the 2019 report, the Philippines ranked 50th out of 174 countries.
With an overall index score of 42.94 out of 100, the Philippines scored the highest in terms of data availability at 70.92 but received a low mark in governance and ethics at 52.18.
The Department of Trade and Industry (DTI) recently reiterated its goal to make the Philippines an AI center for excellence at the Second Philippines-Singapore Business and Investment Summit in late March.
This comes as the AI roadmap the DTI and six other agencies had drafted is set for rollout this month.
The said roadmap covers the implementation of AI in select industries such as agribusiness, manufacturing, and services sectors.
It also includes plans of building a national data center as well as a national center for AI research to encourage both private and public institutions to develop AI solutions.
Meanwhile, some businesses, especially those in the financial sector, are already at the forefront of using chatbots and other AI technologies to keep up with surging customer demand and reduce transactions that needed human contact.
Dr. Cheng cautioned that without any way to ensure ethical AI, the data used to train intelligent systems may be flawed and may cause widespread harm.
“If we feed them (intelligent systems) data that is biased, it will also be biased. So it will be ‘bias in, bias out.’ If the data that we give them is discriminatory, ‘discrimination in, discrimination out.’ And the problem with these systems is that it’s automated, it’s easy to deploy,” she said.
As an example, Dr. Cheng said a company could be using AI to speed up the selection of candidates. The data used to train the system, however, could lead to the AI having unfair preference over certain candidates.
Amazon encountered this dilemma years ago when it attempted to use an AI tool to recruit talents. The system ended up having a preference for male candidates after it was fed with resumes the company received over the years.
“Are the preference and non-preference justifiable? Are they correct? And can we even identify the preferences of this system? There are systems being built and people cannot even explain how it is doing its task,” Dr. Cheng added.
“If it’s working blindly if there’s some failure at some point [and] somebody sued, who’s going to be sued? You cannot sue software, right? So it’s the company that will take in the liability and how is that going to be explained?”
Moderation is key
Dr. Cheng also cautioned that too much regulation will be counterintuitive as it might stifle progress in the field, especially when it comes to research.
“There must be some regulation but I don’t want it to be stifling, I don’t want the field to be highly regulated. Too many regulations might deter development,” she added.
As an added level of protection, Dr. Cheng also said that developers should exercise self-regulation as there are already published guiding principles that can serve as a template for AI practitioners.
“Before we build, we have to consider already, the possible effect of the system on behalf of those who would be affected,” she added.
“In the end, it is us, humans, we should be the ones who should be reflective, who should be careful, and who should be cautious of the data that we use and the applications that we are building that will influence the output of these artificially intelligent systems.”