AI’s Hidden Costs: The Trade-Offs Behind the Tech

 Sacked by the Spreadsheet

Imagine waking up to find your job replaced by an algorithm. For millions of people this is now becoming reality; AI has redefined customer service and manufacturing in countless industries across the world (Sharp, 2024). While some marvel at the gains in productivity, there emerges a burning economic question: what are the hidden costs to progress?

The Price of Progress

In a free market, all participants in the economy freely exchange information as to their needs, wants, skills and abilities. This process creates an ever-updating network of prices that accurately reflect the relative costs and benefits of undergoing different economic activities. However, the efficacy of this mechanism often breaks down. Sometimes, prices fail to capture all the effects of producing a good or service. Economists call these hidden costs ‘negative externalities’ - when someone else ends up paying the price for a decision they weren’t part of (Kenton, 2024). If a firm replaces workers with AI, it may reduce its private costs by increasing productivity. However, the social costs of retraining workers and mental health crises caused by unemployment are externalised onto society. Thus, these costs are a classic case of a market failure, unaccounted for in corporate balance sheets.

Autonomous vehicles will destroy 500,000 jobs in California’s trucking industry, but Tech firms won’t pay the social cost (Garrifo, 2022). Displaced workers, increased demand for public support services, and the psychological toll on affected families are all effects that were not priced in by markets. Sound familiar? The process is much like factories polluting rivers without covering the bill for cleanup.

 

Convenience Over Competence

There are clearly social costs to AI that have been ignored - but are there also benefits that have been overestimated? Here’s where two classic economic ideas come in: moral hazard and the principal-agent problem. Moral hazard occurs when a party takes risks they ordinarily wouldn’t, because they don’t bear the full consequences of those risks. The principal-agent problem, on the other hand, occurs when one party (the agent) makes decisions on behalf of another (the principal), but their unaligned interests affect economic efficiency.

In many ways, AI is the time-saving tool people claim it is. There’s no doubt it stands out as a powerful method to generate refined signals that reduce information asymmetry in economic transactions. By embedding these signals into principal-agent contracts, moral hazard can be mitigated, boosting efficiency across various market structures. Take Uber as an example: AI tools help reduce asymmetric information by producing precise ‘effort signals’. For instance, cross-referencing a driver’s GPS data with assigned trips to see whether they are genuinely working or idling. If agents know their behaviour is being tracked closely, they have less leeway to shirk or misreport, reducing the moral hazard of not being monitored (Zhang, 2025).

The twist? AI doesn’t just fix moral hazard, it can create it too. As workers have increasing access to AI tools, they become increasingly reliant on them. Reliance on AI can affect workers’ own skills and pose potential problems if the AI software is not always readily available (Klingbeil, 2024). Let’s use an example that may cut close to the bone for fellow students. How many times have you asked ChatGPT for an explanation, copied it into your notes, and moved on?  This often means you’ve not really understood the topic. You miss figuring things out for yourself - making connections, thinking critically, or even just sitting with a tricky concept until it clicks. It might feel like you're learning, but you’re really just going through the motions.   

In economic terms, society doesn’t gain as much from AI as firms think they do, because the private benefits to firms outweigh the social benefit to everyone else. That’s where things get out of balance and may warrant action.

                                  

Policies and Protection

We’ve established that the unchecked adoption of AI is creating economic inefficiencies, because firms and individuals focus primarily on their private costs and benefits (Lane, 2021). Moral of the story? Sometimes the ‘invisible hand’ of the market needs a little nudge.

So how can governments and institutions intervene? Some possibilities include:

1.      Pigouvian Taxes: Financial penalties for firms fully replacing jobs with AI, negating the social costs of unemployment. This may realign the private and social cost of replacing workers with AI, ensuring a more socially optimal level of automation.

2.      Transparency Mandates: Require firms to disclose automation plans, empowering workers to adapt to AI developments. This could increase the productivity of workers, lowering the private cost of employing humans compared to automating the process.

3.      Mandatory Evaluation of Employee Skill Retention: Require firms to periodically evaluate potential erosion of employee skills due to overreliance on AI, mitigating overestimations of automation’s benefits.

Some governments have already acted. The EU’s AI Act (2024) set a precedent by classifying high-risk AI systems. Meanwhile, former US President Joe Biden’s AI Executive Order (White House, 2023) emphasises ‘worker-centric’ innovation. Nevertheless, more may be needed to ensure externalities are internalised - government action is not always enough. In theory, the market could sort all this out on its own, that’s what the Coase Theorem suggests. But in reality? Workers don’t have the power or information to negotiate with big tech firms, so it rarely works out that way.

Human-AI collaboration may offer a middle ground without intrusive or misguided government regulation. Germany’s 'AutoUnion 4.0' pays manufacturers to use AI to assist, not replace, workers. Without layoffs, BMW’s AI-assisted lines saw productivity rise by 25%. Evidence from California suggests autonomous vehicles might grow the economy while creating jobs and raising wages without mass layoffs (Wilkinson, 2023).

Balancing Benefits and Burdens

AI’s promise is genuine, but so are its distortions. When markets ignore hidden costs and inflate hidden benefits, progress has a serious price. It’s time to treat automation not just as a tool, but as a trade-off. It’s microeconomic tools like taxes and regulation that can help bring the balance society desperately needs.

References

1.      Brookings Institution (2023). Automation and Artificial Intelligence: How Machines Affect People and Places. Retrieved from https://www.brookings.edu/articles/automation-and-artificial-intelligence-how-machines-affect-people-and-places/

2.      California Legislature (2024). Senate Bill 1047 (Automation Transparency). Retrieved from https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202320240SB1047

3.      European Commission (2024). EU Artificial Intelligence Act. Retrieved from https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai

4.      Gariffo, M. (2022) Automated trucks could cost 500,000 US jobs, researchers say, ZDNET. Available at: https://www.zdnet.com/article/university-of-michigan-study-claims-automated-trucks-could-cost-500k-us-jobs/ (Accessed: 31 March 2025).

5.      Image 1, AI generated using ChatGPT 4o

6.       Image 2, Available at:(9) Five ways AI is transforming the trucking industry | LinkedIn

7.      Image 3. Available at:ChatGPT in education: How students and teachers can use AI to transform learning | YourStory

8.      Kenton, W. (2024) Externality: What it means in economics, with positive and negative examples, Investopedia. Available at: https://www.investopedia.com/terms/e/externality.asp

9.      Klingbeil, A., Grützner, C. and Schreck, P. (2024) ‘Trust and reliance on AI — an experimental study on the extent and costs of overreliance on ai’, Computers in Human Behavior, 160, p. 108352. doi:10.1016/j.chb.2024.108352.

10.   Lane, M. (2021) ‘The impact of artificial intelligence on the labour market’, OECD Social, Employment and Migration Working Papers [Preprint]. doi:10.1787/7c895724-en.

11.   Lane, M. and A. Saint-Martin (2021), “The impact of Artificial Intelligence on the labour market: What do we know so far?”, OECD Social, Employment and Migration Working Papers, No. 256, OECD Publishing, Paris, https://doi.org/10.1787/7c895724-en.

12.   Sharps, S. (2024) The impact of AI on the labour market, Tony Blair Institute for Global Change (TBI). Available at: https://institute.global/insights/economic-prosperity/the-impact-of-ai-on-the-labour-market (Accessed: 31 March 2025).

13.   White House (2023) What they are saying: President Biden issues executive order on safe, secure, and trustworthy artificial intelligence | The White House, National Archives and Records Administration. Available at: https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2023/10/31/what-they-are-saying-president-biden-issues-executive-order-on-safe-secure-and-trustworthy-artificial-intelligence/ (Accessed: 02 April 2025).

14.   Wilkinson, L. (2022) Autonomous long-haul trucking stands to grow the Golden State’s economy while creating jobs and raising wages without mass driver layoffs, Silicon Valley Leadership Group. Available at: https://www.svlg.org/study-shows-autonomous-trucking-will-grow-californias-economy/ (Accessed: 31 March 2025).

15.   World Economic Forum (2023). The Future of Jobs Report 2023. Retrieved from https://www.weforum.org/reports/the-future-of-jobs-report-2023/

16.   Zhang, T. and Zhang, Y. (2025) Generative AI and information asymmetry: Impacts on adverse selection and moral hazard. Available at: https://arxiv.org/html/2502.12969v1 (Accessed: 31 March 2025).

Why Online Thrifting Feels Like a Gamble (and How Economics Explains It!)

 Am I the only one that is fed up with getting scammed on secondhand clothing platforms? The other day I bought a jacket off Vinted, only to receive a crisp packet in the mail. As a student that was my last £30 (not my wisest decision, I admit) and now I'm left questioning whether I can ever trust these platforms again. These days authentic finds are becoming rarer, buried beneath an overwhelming flood of low-quality listings and outright scams! That jacket was truly a diamond in the rough. It's a shame didn't actually receive it.

 

Fortunately (or unfortunately), I'm not the only one who has fallen victim to these scams.

The online second-hand clothing market suffers from what economists call an adverse selection problem, where sellers have incentive to list low quality products, such as stained, ripped clothing, or even straight up scams (note the crisp packet) rather than high quality goods, such as the vintage Carhartt jeans you’ve had in your shopping basket for the last 3 months. For every online secondhand purchase, buyers feel as if they are in a game of Russian roulette - will you receive your desired item, or will be left disappointed?

This problem occurs due to asymmetric information, where one party, the sellers, has greater knowledge of the product than the other party, the buyers. Whilst you may think you have found the perfect pair of baggy jeans from websites such as Vinted or Depop, do you really know what you are buying when you can’t physically hold it and try it on? Even if you’re lucky enough to see a selection of photos, you can only see it through your iPhone screen. Sellers are incentivised to simply flood the market with low quality products or even scams as it wouldn't be differentiable from genuine worthwhile products, leading to a wine-stained jumper priced 3 times its actual worth.

When Buyers Fear the Unknown: Threadflip’s Trust Issue

Threadflip is a prime example of how asymmetric information can doom a marketplace. The online secondhand fashion site had over 1.5 million users at its peak and differentiated themselves by handling photography, pricing and shipping in order for a cut of the sale. However, despite the promise of providing a seamless concierge service, by 2016, the marketplace shut down.

So, what went wrong?

The reason for the downfall of Threadflip can be explained by theories discussed by George Akerlof in his 1970 paper, “The Market For Lemons”.

This paper introduced the concept of adverse selection which concerned information asymmetry. The key example used to explain this was the second-hand car market. He referred to low- and high-quality used cars as lemons and peaches respectively, narrowing down on how buyers are unable to distinguish between the two.

 

The graphs (doodled expertly by our team) model how Akerlof’s theory works in real life. We won’t bore you with too much detail, but a simple explanation goes as follows:

When sellers have more information about their car quality, “a lemon’s problem” arises in which low quality cars drive the high-quality ones out of the market.

·         In Diagram A, buyers originally demand high-quality cars (DH), but as they realise they can't always tell good cars from bad, they lower their expectations. This reduces their willingness to pay, shifting demand down to DM.

·         In Diagram B, the opposite happens for low-quality cars. Since buyers can't always tell they're getting a bad deal, demand for these cars increases in the first instance, from DL to DM.

Over time, buyers become even more skeptical, further reducing demand for good cars (DL), eventually driving them out of the market altogether as owners become reluctant to sell their vehicles at £5000, therefore leading to 0 high quality cars sold.

This aspect of driving sellers out the market was a key factor that led to the failure of Threadflip as high quality clothes mean that buyers hesitate to pay premiums due to fear of undisclosed flaws due to asymmetric information. Let's be honest, who wants to spend money on a “barely worn” designer jacket only to discover several tears? The lack of trust creates adverse selection where honest, good quality sellers struggle to compete. Meanwhile low-quality goods flood the market and buyers either lower their standards or leave altogether. The Threadflip market essentially unraveled.

How are Online Marketplaces Fighting Scams and Building Trust?

Fast forward to more recent times, post Threadflip, the big players in the secondhand clothing market with Vinted and Depop boast over 150 million combined registered users worldwide. Why? The reason for their “successful” business model is down to their emphasis on trust and transparency by learning from past failures. These platforms have implemented features that encourage buyer confidence and seller accountability despite the issue of asymmetrical information. These include detailed listing requirements, seller ratings and buyer protection policies. Ebay, a pioneer in the sector, even includes an authentication process that ships sellers’ items to a verification facility prior to delivery.

 

An Inevitable Gamble?

Second-hand fashion marketplaces live and die by trust. However, even billion-dollar companies cannot fully eliminate asymmetric information.

A recent study by Which? reported by The Guardian (2024) revealed that scams remain a huge issue:

·         57% of users reported being scammed on Depop, the highest among second-hand marketplaces.

·         eBay (29%), despite its authentication services, still struggles with fraud.

·         Even Vinted (22%), which prides itself on safety, isn’t immune.

The hard truth is that no matter how many safeguards exist, buyers will always be at a disadvantage when they can’t physically inspect goods before an online purchase. Platforms can reduce risk, but they’ll never kill adverse selection entirely. Our advice on the best defense? If a deal looks too good to be true, it probably is.

Having said this, it will be interesting to see how technology, particularly AI and advanced verification processes shape the future trajectory for online marketplaces.

References:

Austan Goolsbee, Steven Levitt & Chad Syverson (2020) Microeconomics. 3rd edition. Worth Publishers. https://read.kortext.com.

ITV News article:
ITV News (2024) Third of buyers experienced scam on second-hand marketplaces in last two years. Available at: https://www.itv.com/news/2024-05-08/third-of-buyers-experienced-scam-on-second-hand-marketplaces-in-last-two-years.

TechCrunch article:
Constine, J. (2016) RIP Threadflip. TechCrunch. Available at: https://techcrunch.com/2016/01/12/rip-threadflip/.

Business of Apps (2024) Vinted Statistics. Available at: https://www.businessofapps.com/data/vinted-statistics/

Akerlof, G.A. (1970) ‘The market for “lemons”: Quality uncertainty and the market mechanism’, The Quarterly Journal of Economics, 84(3), pp. 488–500. Available at: https://doi.org/10.2307/1879431