Building AI skills, vs AI building skills

Generative AI can 'democratise skills' – that's a frequently heard argument in favour of its implementation and use. It takes automation one step further, not just replacing activities that are time and labour-intensive, but replacing the capabilities that make people good at higher-level activities including creative work.

In practice, what does this mean for work and the workplace? The history of automation has been a constant replacement of human labour and human skills by machines of increasing complexity, with each iteration of technological advancement directly affecting who works, where and how they work, the skills they need, and even the kind of culture that develops within an organisation.

Some experts believe, though, that generative AI is not as revolutionary as it is made out to be.

"Actually, there are not a lot of new things happening here," commented Terence Chia, Director of the Human Capital Cluster at Singapore's Infocomm Media Development Authority. Speaking at a panel hosted by Salesforce during the company's World Tour Essentials Asia last week, he pointed out that AI is already everywhere: in people's phone apps, supporting their financial transactions, helping them shop, powering their entertainment recommendations.

“The more important question is, do you know how to use AI in your day to day life?” Chia asked. Comparing generative AI to the evolution of mathematical calculations – from printed tables to slide rules to the abacus to electronic calculators and then Excel spreadsheets – he said, “We don't need people to know every intricacy of using neural networks, transformation layers or other kinds of algorithms. We just need them to know how they are going to use this to make their lives better and increase the opportunities for themselves and the companies that they will automate.”

The skills that will be important for AI, vs the skills that will be eroded

Despite the hype, the skills required for implementing and operating generative AI are not that different from the skills required to use any other technology. There are three categories, said Chia:

General skills – basic usage, which in the case of generative AI involves asking the AI the right questions or giving it the right instructions. Increasingly, these also include awareness of legal and ethical issues, and the ability to evaluate for accuracy, given that generative AI is exceptionally bad at referencing its sources. These are necessary for general workers and easy to pick up, and will rapidly become widespread – suggesting that 'prompt engineering' may not be a long-lived specialty.

Sectoral skills – use-case applications, for developing industry or organisation-specific solutions. These skills are closely tied to the industry in question and include specialised or general skills used  outside of operating generative AI. They are likely to be more technical or 'hard' skills, required for the application of knowledge.

Complementary skills – the ability to leverage generative AI such that it can complement or improve what someone is doing, and ultimately create new opportunities or new applications that are not available today. These may be less tangible or 'soft' skills including creativity, flexible thinking, and judgement, among others.

On the other hand, the skills that the AI directly replaces will be eroded, according to Professor Damien Joseph of NTU's Nanyang Business School. Comparing the increasing reliance on generative AI to the way students have come to rely on the common calculator, he said:

“Skills like writing a report from scratch will soon disappear, just as we have lost our ability to do mental calculations from using calculators. But we shouldn't worry about it. Because that's the evolution of skills. What's important is to keep up with what is currently required.”

Can AI take the place of institutional knowledge?

Generative AI's ability to replace skills and knowledge isn't just about machines taking the place of humans, though. An ideal model for the use of AI involves using the technology to enhance existing processes, including onboarding and training. For example, generative AI trained on the specific skills, behaviours, and expectations of an organisation can speed up the integration of a new hire, whether into the organisation as a whole or just into a particular team.

“As you grow the team and bring in new people, there's some element in there that's hard to replicate,” observed Chia. “Even with the most advanced knowledge management systems, you cannot just take someone on the team and plug them in elsewhere or vice versa, and expect them to fit immediately. But perhaps with generative AI, that problem will be solved. And then the question is no longer about bringing expertise in. It becomes about how much of that knowledge and expertise can be stored in the AI, and how much of it you need to hire actual human beings for.”

Professor Joseph, who researches AI extensively, says that the ability of generative AI to learn from user responses and feedback can increase institutional knowledge simply by training the AI on what output is desired for the organisation's particular context. Whether that will change companies' skill needs over time is still uncertain.

He suspects that productivity expectations will change and affect hiring needs, but in quite a different way from what people tend to imagine.

“The other theory is that throughput will increase,” he said.

“These technologies do things cheaper, faster, better, which means that at higher levels, there will be pressure to increase the amount of work done, and you would need then to hire more people.”

Ultimately though, the effectiveness of generative AI is still very contextual, according to Chia, and it comes down to human receptiveness.

“When we think about digital transformation, there is an interplay of two elements: the technology and the people. You may have the most advanced technology, but if the people themselves are not ready to receive it or to make use of the benefits, I think then the technology will actually be underutilised or potentially fail.”