AI採用昇進ツールのバイアス防止

3 Mr Patrick Tay Teck Guan asked the Minister for Manpower whether the Ministry will consider introducing guidelines or regulations on employer's use of Automated Employment Decision Tools (AEDT), which are artificial intelligence technologies that substantially assist or replace discretionary decision-making in hiring or promotions, in order to prevent biases and including measures such as requiring companies to conduct a bias audit or to make disclosure where such tools are used to rank candidates or assess employees for promotion.

The Minister for Manpower (Dr Tan See Leng) : Mr Speaker, regardless of the technological tools used to aid employment decisions, such as hiring or promotions, employers must comply with the Tripartite Guidelines on Fair Employment Practices, which promotes fair and merit-based employment practices.

If certain artificial intelligence (AI) use results in discriminatory employment practices, workers or job applicants can approach the Tripartite Alliance for Fair and Progressive Employment Practices (TAFEP) for assistance. TAFEP will work with the employer to address the grievances of the workers or job applicants and ensure that the employer’s recruitment or performance appraisal processes adhere to the principles of fair employment. To date, TAFEP has not received complaints of discrimination arising from the use of AI tools.

I would also like to caution that AI technologies are evolving at a fast pace. In deploying AI-powered human resources (HR) tools, organisations should refer to the guidelines introduced by the Government to support the responsible development and use of AI. The Government will continue to closely monitor the trends in AI adoption and work with our tripartite partners, the Institute for Human Resource Professionals (IHRP) and the broader HR community to regularly assess if existing guidelines and regulations are adequate.

Mr Speaker : Mr Tay.

Mr Patrick Tay Teck Guan (Pioneer) : Mr Speaker, I thank the Minister for his response. Just one supplementary question on how we can better protect workers' data privacy when AI systems are used, especially in HR processes.

I am glad that the Minister has introduced the Workplace Fairness Bill as well as some of the measures and Tripartite Advisories, but can we ensure that workers' consent is obtained before their data is used for AI-driven decisions, including promotions, hiring, re-employment and layoffs?

Dr Tan See Leng : I thank the Member for his supplementary question. We certainly will take the Member's suggestion to continue to improve the processes and to tighten it.

As I have shared earlier on, the adoption of AI, the use of the multiple tools is proceeding at a fairly rapid pace. To be overly prescriptive, to be overly tight in terms of our regulations, I think it would not just be not practical, but I do not think at this particular point in time, given the stage of evolution of the entire AI industry, we can really catch up with them adequately.

I think the best, nuanced approach is to continue to maintain clear responsible use of AI, which the Government has actually introduced in the form of a model AI governance framework. And one particular example is AI Verify. I think for the purpose of the brevity of this discussion, I do not want to go into too much discussion on that part of it. This is a toolkit, which is developed by IMDA. I am happy to walk the Members of the House through at subsequent Parliamentary Sittings if there is another Parliamentary Question filed on that, to talk about how we can use that.

For the use of data in terms of the algorithms that many of these companies may want to use, it is important that the data pertaining to individuals is anonymised. And, of course, consent would really be one of those things that we are looking at as well. So, I hope that gives the Member that reassurance that we are doing everything that we can to stay on top of it.

Mr Speaker : Dr Tan Wu Meng, a short one please.

Dr Tan Wu Meng (Jurong) : Can I ask the Minister, in his assessment of the current landscape of TAFEP cases on unfair HR practices, roughly what proportion of cases did the adjudication hinge on proof of intention by the hiring manager or the company's management? Because with AI, it can be difficult to ascertain intention because the AI is not able to give testimony and be cross-examined or provide information for investigation the way a human can be questioned.

Dr Tan See Leng : I thank Dr Tan for his supplementary question. As I have said, we are at a very pivotal state of transformation and the adoption of AI. If we were having a series of discussions earlier on and Dr Tan himself also brought to our attention that, today, you can actually file a legal suit using ChatGPT.

So, what is fundamentally important for us today is to work closely with employees or with potential employees who may feel that they are aggrieved, to surface such cases to us so that we can investigate. Then, obviously, we will work with the companies to see if some of the algorithms – sometimes, it may not be an intention, it could be a function of the datasets that the company is using – have an inherent bias, for instance, in looking at certain characteristics and, therefore, favour hiring or promotion in favour of those characteristics.

So, we need that constant sense of vigilance, we need the participation of all parties coming together. We also need different agencies, the IHRP, the Labour Movement and we need our tripartite partners to come into the space alongside with us. Then we can ensure a more equitable society and workplace for everyone.

Mr Speaker : Mr Pritam Singh.

Mr Pritam Singh (Aljunied) : Mr Speaker, just a response to the Minister through a question vis-à-vis what he said about AI. Unfortunately, from the worker's perspective, one usually is not in a position of information superiority over the employer, so you do not know what back-end selections or pre-qualifications your AI system has done.

So, in that context, where there is always a power imbalance and there is always information asymmetry in favour of an employer, what new approaches would the Ministry of Manpower (MOM) consider to encourage employees to raise concerns to, say, TAFEP or to any other appropriate authority in MOM?

Dr Tan See Leng : I thank the Leader of the Opposition for his question. I think that it is important to recognise that in today's climate, depending on which aspects that you are looking at, the employee also has certain advantages vis-à-vis choosing the choice of employers that they want to work for, because we do have a very tight labour market. So, it is how we achieve that good balance, of balancing responsible employees who really form the majority of our workforce, alongside responsible employers who are also the majority of our entire job ecosystem, with a very good balance achieved through tripartism – working with the Singapore National Employers' Federation (SNEF), working with the Labour Movement and also getting other non-government organisations, the IHRP and other organisations which have been working alongside with us. We hope that we can continue to maintain this harmonious ecosystem.

Of course, we cannot take things for granted because we continue to work hard at tripartism, at working with the different chambers to achieve this. But I want to reassure the Leader that every single complaint and every single question is dealt with expeditiously in a very clear and transparent manner. And by building a database, a repository of the nature of the different types of complaints, it would sharpen our ability to protect employees better and, at the same time, maintain that balance for employers.