Will AI take my job? This question continues to be the focus of countless opinion columns and academic papers. With good reason. Concerns about the power of AI are growing globally. A recent survey from market research multinational, Ipsos, found that 52% of respondents feel nervous about AI products and services, a rise of 13% since 2022. The Brookings Institute’s recent high level panel on AI and the job market turned out to be a refreshingly low-key discussion on what has become a predictably dystopian topic. So let’s dig in.
It’s true, AI is exceeding human performance on several benchmarks. So says Stanford’s annual AI Index Report. Yet it trails behind humans on more complex tasks like visual common-sense reasoning and planning. Add to this, a lack of standardised benchmarking. This means that it is difficult to be sure exactly what is being measured and compared across AI models. As Russel Wald, Director of Policy for the Stanford Institute for Human-Centred AI (HAI) explains, there is no one universal standard level benchmark. Neither is it clear who should be creating that standard. Wald goes on to explain that studies at HAI have shown AI to be most effective when used to augment human output.
AI and, not or.
The idea that AI can complement rather than substitute the work of humans is a popular one. There are three basic ways in which tech giants can develop AI, explains MIT Economist, Daron Acemoglu. They can either automate the job or provide more information for the worker so that s/he can do the job better. Or, AI can be used to create whole new jobs. The current development focus is on automation and this” benefits capital at the expense of labor” as Acemoglu puts it. He sees the regulatory role of governments as pivotal here. AI should be developed to empower rather than disempower workers. Yet a current lack of broad-based discussion surrounds what he calls, “these basic truths and choices”. This really worries him.
Economist and author of A World Without Work, Daniel Susskind, agrees in principal. However, he points out the difficult reality of a more planned approach to technological development. For example, an AI tool designed to assist doctors identify skin cancer might become so good at the task, that it replaces human users. It is difficult to predict all use cases for new technology, particularly one as complex as AI. There are also political impediments to guiding AI development in this way.
Should we be worried about AI in the short term?
While agreeing that the development of AI has been “truly spellbinding”, Acemoglu expects the impact of AI on the job market to be relatively slow over the next two years. At present, it is a small minority of people who choose to use AI in order to improve productivity. However, this is not the same as “all the AI being done to us” as he puts it. Most of us are now well aware of the many algorithms that have found their way, often uninvited, into our daily lives.
Acemoglu hopes that new companies and government can enter this space, while the pace of change is still relatively slow. He believes that “augmentation is key” and would like to see more money going into developing AI tools that help make the work of those like teachers and healthcare workers easier.
Wald agrees that up until now, industry’s dominance in the field has meant that regulation of AI and appropriate policy responses have been limited. What he terms, “a consistent lack of transparency from industry” makes regulation difficult. He also suggests that policymakers have been “captivated by language models” and points out that there is “a whole different world of AI” that we’re not spending enough time thinking about. For example, the use of AI in the sciences is unlocking a lot of new knowledge, which he expects to result in the “supercharging” of scientific discovery.
Innovation needs to be more targeted.
Yet innovation does not necessarily lead to increased productivity. Acemoglu points out that the last five decades have been hugely disappointing in terms of the link between the two. Never have we seen so many new patents being registered in the US while productivity has remained fairly stable over the same period of time. The MIT economist argues that we need to see a much greater decentralization of AI development in order to unleash genuine and useful application of this technology.
For Susskind, the challenge we face in the short term is a lack of skills for existing work opportunities. So available work might be “out of reach” for those without a certain skills base. So let’s look longer term. Susskind sees a future scenario where there is simply not enough work for people to do. This has economic, political and social implications. The economic challenge involves income – how does one get an income without a job? Or, as Susskind puts it, how to distribute wealth if the labour market doesn’t do it.
Solutions
Possible solutions include options like Universal Basic Income, or governments might simply create jobs, to give citizens something to do. But work isn’t simply about income, for many it provides purpose and meaning. “The relationship between work and meaning in our lives needs to be more thoroughly and clearly explored”, argues Susskind. The growth of AI might well force us to do this. Why do we work? Would we rather spend our time pursuing other interests, if we didn’t have to hold down a job?
The role of Big Tech and the increasing concentrations of both economic and political power that reside with these platforms is also of concern. Both Acemoglu and Wald would like to see more AI development being nurtured by the public sector.
We live in an age of exponential technological development. This is both exciting and daunting. It is also marked by high levels of uncertainty. “There is a lot of uncertainty about all this, more than most experts are willing to admit” says Acemoglu. One thing however, is certain: the importance of humanity. As Wald reminds us, the key question remains: “How are we making sure that human dignity is being preserved through all of this?”