Will Artificial Intelligence (AI) liberate or enslave humanity? As an economist with zero technical expertise, I will not venture a guess. But some economic fundamentals offer some perspective on the AI economy.
AI will significantly transform the labor market. The potential jobs apocalypse is not my topic today. Many cognitive jobs will surely be automated, as physical tasks have been for decades. Until AI, only humans trained in the field could answer law or accounting questions. We will need fewer lawyers and accountants, but we will still need some.
But AI will not abolish scarcity. Indeed, data centers need enormous amounts of electricity, both of which will need materials. AI will not make everything free.
Tech gurus often overlook the subjectivity of economic value. Goods and services are valuable because people exchange valuable things for them. Value arises from our personal and varying preferences. Something can be worth millions to one person and nothing to another.
No one must justify their expenditure to others in a market economy. And firms seek to deliver the goods and services customers want, whatever they might be.
Subjective preferences will likely slow or limit the dissemination of AI-enhanced products. To take one example, consider novels, poems, and travel stories. Do you wish to read AI generated novels or travelogs? Perhaps. Yet many readers value the encapsulated human perspective on life as opposed to the string of words.
An important source of value creation in the AI economy consequently should be guaranteeing human experiences. A publisher catering to readers wanting books written by humans must exclude AI-generated content.
Subjective value will also impact AI-enhanced services. AI will be able to provide legal advice and medical diagnoses comparable to humans. Tech experts may view AI services as equivalent to human services, but value arises from the human users. Some people will want a human doctor or lawyer, perhaps assisted by AI.
I will not speculate on the size and extent of the human-focused market. Subjective preferences need not exhibit logical consistency.
Another challenge to AI adoption involves alignment of products with users’ values. AI learns to write poems or solve problems through random trials which are scored using metrics. Through billions of repetitions, AI learns how to form words into a high scoring “poem.”
The scores embody values. Sometimes everyone will agree on scores, like the outcome of a game. But differing preferences or values lead to diverging evaluations of outputs.
Google’s AI image generator illustrates the issue. The training clearly prioritized diversity over historical accuracy. Some conservatives recognize the potential for embedding woke, progressive values in AI. President Trump’s AI executive orders have voiced concern about this.
A related issue involves the consumer of AI expertise. We might presume that the user is the consumer. But critics observe how users and their information are the product marketed by Facebook, Instagram, and Google to their paying customers.
The users of free AI may well be the product, but users pay for job replacing AI tools. This suggests greater alignment between user preferences and the company’s product.
But the alignment may not be perfect. Both Open AI and Anthropic products reportedly try to keep users from turning them off. If true, this cannot align with customers’ interests. No trucking company wants trucks that they cannot turn off when parked.
Beyond these trust issues are fears of malign, powerful AI controlling everyone. Whether justified or not, such fears will lead some to avoid AI. Earning and keeping users’ trust should be crucial for AI companies.
I do not fear people or businesses delegating their autonomy and thinking to AI. This is already possible (to outside consultants, astrologers, etc.). Provided that we maintain a free human society, people and businesses will remain free to decide for themselves how to use AI.
Colleges could have automated lectures forty years ago through video recording. Yet student and parent preferences kept this from happening. If AI serves us, human values will still shape our market economy.
Daniel Sutter is the Charles G. Koch Professor of Economics at Troy University. The opinions expressed in this column are the author’s and do not necessarily reflect the views of Troy University.