The Social Benefits of AI Should Match Its Costs and Risks

In this post I would like to set aside for a moment our worries about the existential risks of runaway AI and take up another issue that should concern us in the very costly pursuit of research and development of AI. 

Investing in any project entails opportunity costs in the form of other projects and endeavors we must forego plus all possible collateral effects that result from the pursuit of that project.  The opportunity costs and side effects are justified if they generate outcomes of equal or greater value.

Both hype and substance have made AI look like the next frontier of fabulous profits.  Not surprisingly, the market has responded by pouring enormous sums of money into its development with the expectation that the profits will accrue to the private investors.  The thing is, however, that not all costs are contained within the ranks of investors.  A big part of the costs falls on all of us.  The question we need to ask, therefore, is what benefits ought to accrue to society to justify its contribution to the research and development of AI.

Let’s start with the financial investment in AI.  PIMCO (an asset management firm) estimates that building AI infrastructure (data centers, etc.) will consume $5 trillion dollars by 2030.  Just five companies, Alphabet, Amazon, Meta, Microsoft, and Oracle are projected to invest $1 trillion.  These huge investments are on top of the hundreds of billions of dollars already invested in AI.  All this is money that could potentially fund other valuable projects.  Right now, it is not clear how much of this extraordinary commitment of financial resources is driven by hype and how much by substance.  Even more disconcerting is that we do not know how much of this financing is done in the fanciful pursuit of superintelligent AI with all its dangerous consequences and how much will go toward the development of really useful applications. 

One of the most important functions of capital markets is the allocation of capital to uses that promise to generate value to consumers and society.  The enormous faith the private markets are currently placing on AI will truly test the allocational efficiency of the market.  Just for a cautious perspective, I will remind us of the faith the stock market placed on firms with dot.com in their names which was responsible for the collapse of stock prices in 2001 and the faith the market placed on the housing market debacle almost brought down the global economy in 2008.

Next, let’s consider the side or collateral effects of AI investments.  I don’t have an exhaustive list, but these are some of the effects I have come across.  First, AI infrastructure requires huge amounts of energy and water resources.  The unavoidable rise in energy and water prices falls on the public.  Studies by the Federal Reserve and others suggest that the overall demand for everything that goes into building the AI infrastructure is contributing to the persistently high inflation we experience.  Through its contribution to inflation and its voracious demand for capital, AI is also contributing to the elevated interest rates for public as well as private debt (consumer loans and home mortgages).

Therefore, huge amounts of capital as well as energy and water resources, and higher inflation and interest rates comprise the private and public direct and indirect costs going into the development of AI.  Society’s contribution to the research and development of AI does not end here.  The heart of AI development is the training of its agents.  This is possible because AI firms appropriate our private data and information and the intellectual stock humanity has produced since historical times.  Without training on this input AI development as it is done now would not be possible. 

Given the indirect costs society must bear as well as its direct contribution to the research and development of AI, I think that we have more than a credible and valid claim to demand that AI produces tangible social benefits.  It is interesting that this proposition has been voiced by liberal and conservative figures.  For example, Bernie Sanders, Eric Schmidt (former chief executive of Google, and Oren Cass (chief economist of a conservative economic think tank) have suggested that AI-generated revenues be used to establish a sovereign wealth fund for the purpose of sharing profits and mitigating negative effects on society.

We need, however, to go beyond a wealth sharing plan.  If AI has the potential to be as transformative as its creators and enthusiasts believe, then AI should contribute more directly toward achieving a better society and a better world more generally.  To give AI a free hand to profit while it generates adverse effects it’s not the smartest way to harness its potential.  Here are three areas where we can use AI to produce socially beneficial outcomes.

The first is health care.  Yes, AI can give us superior medical solutions, drugs, and health care management, but unless these benefits improve public health and touch all people, they will remain an incomplete promise.  In life expectancy, number of comorbidities, mental health, and health insurance the US lags other advanced countries.  To correct this gap, we do not only need the exploits of AI; more importantly, we need to rectify institutional inequities so that all can share the AI-based improvements in health care.

The second is education.  By all evidence, we are still searching for ways to make the use of AI a positive contributor to studying and learning.  Thus far, we have heard alarming stories of how AI stifles the development of critical thinking, creativity, and writing skills.  The early adoption of AI tools in schools and colleges has been followed by a retreat.  AI firms, educators, and educational institutions must work together to harness the potential of AI to produced better educated Americans, an area where we also lag other countries.

The third area we need AI to produce socially beneficial results encompasses economic equality, productivity, and workers’ incomes.  Economic equality will start to improve when the productivity gains are more fairly shared with workers.  As of now, the evidence points toward significant productivity gains in some professional sectors but modest economy-wide gains.  In their book Power and Progress, the economists Daron Acemoglu and Simon Johnson argue that at the present time AI tools are geared more towards automating, replacing, and surveilling workers than augmenting the productivity of workers or upgrading the work of low-skill workers.  If this direction persists, workers will lose bargaining power and will be pushed farther from the decision-making process as far as their future is concerned.  This possibility can be averted only if we reform the institutional arrangements pertaining to corporate governance and managing work conditions and remuneration.

The bottom line is that investing so much in AI is not a bet that affects only the private sector.  It is a huge bet for society as well.  Therefore, we are entitled to demand that AI generates significant socially beneficial outcomes for all.  But for this to happen just regulating the AI industry is not enough.  The equitable distribution of the AI-generated benefits will be possible only if we have the political will to set up the right institutional framework.