The “Death” of Grades as a Signal: What Happens Next?

 What’s the point in slaving away for hours a day studying for an exam when AI can complete it in twenty minutes? A question many university students are contemplating.

 

Clearly students are increasingly picking the easier option displayed by a Guardian study, which found there were 7,000 cases of AI cheating across UK universities in 2023-2024 (The Guardian, 2025). A deeper economic issue arises from this; how do employers distinguish between high-ability students and lower-ability students if AI is allowing everyone to achieve the same grades? As AI develops achieving higher marks for students, the signal for employers grows weaker, stretching the issue outside of the classroom and into the real world.

 

Why Grades Matter, the ‘Label’ Problem:

Employers cannot understand our capability just by looking at us; in economics this hurdle is called asymmetric information, where ability is masked leading employers to rely on exam results as a ‘signal’ of ability. The signalling theory suggests that higher ability students can achieve higher grades at a lower cost or effort. This difference results in a separating equilibrium, working like a filter, where individuals of different abilities choose different performance levels (Spence, 1973). Under this equilibrium, exam results are more than just numbers on a paper, instead they work as signals enabling employers to select suitable candidates.

 

From Separation to Pooling: How AI Reshapes Academic Signals

As COVID-19 swept across the world, it accelerated the shift from in-person to online exams. Technological advancements reduce the costs of online exams whilst enhancing efficiency, flexibility, and standardization, making them a growing trend (Susnjak and McIntosh, 2024). At the same time, AI tools such as ChatGPT have emerged and provided new opportunities for students to boost their grades.

Image1: More students are using AI in online exams

Source: The Guardian (2024)

 

So what will happen when students, particularly those with lower academic performance, start using ChatGPT to boost grades? Evidence suggests that AI-generated responses can achieve similar or even better scores than those produced by students themselves (The Guardian, 2024). Assessments that once required hours of preparation and research can now be completed within mere minutes. This implies that AI significantly reduces the cost of achieving higher scores, especially for students who would otherwise need to exert greater effort to succeed.

Figure1: AI lowers the cost of high grades for low-ability students, shifting the outcome from separating to pooling

 

As shown in Figure 1, AI lowers the cost gap between high-ability and low-ability students, enabling lower-ability students to produce similar outcomes. Consequently, exam results can no longer clearly distinguish between different ability levels. Meaning the separating equilibrium that once made exam performance informative, has begun to collapse. We are left with a pooling equilibrium where grades can no longer reliably reflect underlying ability.

 

What does this shift mean for an employer? As the cost of achieving high scores declines, more students obtain similar high scores, raising the overall grade distribution and leading to grade inflation. While first-class grades become easier to obtain, the informational value of grades diminishes, weakening their role as a signal of ability. When employers can no longer effectively distinguish between candidates based on exam performance, they are more likely to rely on average expectations, thereby deepening information asymmetry. Under such conditions, this creates a “market for lemons” (Akerlof, 1970), where poor information makes it harder to separate from high quality from low quality and adverse selection may arise, which means that low-ability candidates are more likely to enter the market, while higher-ability individuals may be undervalued and crowded out, reducing the overall quality of the labor market.

 

Alternative Signals: A Solution or a New Problem?

When grades lose credibility, firms will pivot to alternative signals to filter ability, rather than relying on this single indicator (Stiglitz, 1975). For instance, indicators such as internships and interview performance are typically harder to ‘hack’ and therefore become significantly influential in evaluating candidates.

 

Are these firms actually better off after this shift? First, it may be too absolute to conclude the ‘death’ of the degree, academic signals still carry weight, as the use of AI does not necessarily eliminate all cost differences in achieving high scores when higher-ability students may still use these tools more effectively. This suggests that some degree of separation may continue to exist. Similarly, alternative signals are not perfect either - interviews can be biased, and internships may reflect access to opportunities rather than ability (Arrow, 1973; Rivera, 2012), thereby exacerbating inequality. Moreover, hiring costs may increase as firms adopt alternative signals and evaluate candidates with a broader range of criteria.

 

Can We Fix the Signal?

This brings us back to the first question: can we restore the academic signal? Leading to a better outcome for firms and students. One way is to regulate AI use by requiring students to declare AI use or by introducing detection tools (Russell Group, 2023). Although this does not fully solve the problem, it could artificially raise the cost of over-relying on AI and reduce the information gap in the evaluating process. Another approach is to include more in-person exams and varied assessment formats, such as presentations or in-class tasks (Attewell, 2024). These are harder to “game” using AI, forcing more direct engagement and ensuring the maintenance of the separating equilibrium once again.

 

Overall, restoring the signaling power of the degree matters. We still hope that grades can continue to play an important role in the labour market as a relatively low-cost, standardized and broadly fair signal. However, we cannot ignore the limitations of using grades as a signal. As higher education becomes increasingly digital, online exams are unlikely to disappear given their flexibility and cost-efficiency and imperfect AI detection means universities cannot fully guarantee the reliability of this signal alone.

 

Even so, grades can still work as an initial filter, giving employers basic information about applicants and preserving some fairness. When combined with interviews and internships, we can form a broader and more balanced way of assessing candidates, as many firms already do today.

 

What Comes Next?

As AI advances, it has an increasingly direct effect on the evaluation of human capital. For universities, this means regulating AI use to preserve grades as credible signals of ability. For students, it means adapting to a more complex selection process and building strengths in interviews and internships. While current solutions are far from perfect, continued development offers a chance to redefine new signals, leading to a fairer and more effective system for assessing job applicants.

References

Akerlof, G.A. (1970) ‘The Market for “Lemons”: Quality Uncertainty and the Market Mechanism’, The Quarterly Journal of Economics, 84 (3): 488-500.

 

Arrow, K.J. (1973) ‘Higher Education as a Filter’. Journal of Public Economics, 2(3), pp. 193-216.

 

Attewell, S. (2024) ‘Exploring AI and assessment - avoid, outrun or embrace’, Jisc, 22 April. Available at:

https://www.jisc.ac.uk/blog/exploring-ai-and-assessment-avoid-outrun-or-embrace (Accessed: 22 April 2026).

 

The Guardian (2024) ‘Researchers fool university markers with AI-generated exam papers’, 26 June. Available at: https://www.theguardian.com/education/article/2024/jun/26/researchers-fool-university-markers-with-ai-generated-exam-papers (Accessed: 24 March 2026).

 

The Guardian (2025) ‘Revealed: Thousands of UK university students caught cheating using AI’, 15 June. Available at: https://www.theguardian.com/education/2025/jun/15/thousands-of-uk-university-students-caught-cheating-using-ai-artificial-intelligence-survey

 (Accessed: 23 April 2026).

 

Rivera, LA. (2012) ‘Hiring as Cultural Matching: The Case of Elite Professional Service Firms’, American Sociological Review, 77(6), pp. 999-1022.

 

Russell Group (2023) Principles on the use of generative AI tools in education. London: Russell Group. Available at: https://www.russellgroup.ac.uk/policy/policy-briefings/principles-use-generative-ai-tools-education (Accessed: 22 April 2026).

 

Spence, M. (1973) ‘Job market signaling’, The Quarterly Journal of Economics, 87(3), pp. 355–374.

 

Stiglitz, J.E. (1975) ‘The Theory of "Screening," Education, and the Distribution of Income’, American Economic Review, 65(3), pp. 283-300.

 

Susnjak, T. and McIntosh, T.R. (2024) ‘ChatGPT: The End of Online Exam Integrity?’ Education Sciences, 14(6), 656. Available at: https://doi.org/10.3390/educsci14060656

When Love Becomes a Signaling Game The Problem of This Age.

 Do you feel lonely? If so, you are not alone, because 61% say they do! We are deep into a mental health crisis. We are raising the loneliest generation on record (Gen Z) (Cigna, 2020), closely followed by Millennials. With the digitalization of the world and the alienation of people, it has become harder to find a partner offline. Thus, for many young people, dating and partner selection have moved increasingly online (Rosenfeld, Thomas and Hausen, 2019).

Source: Opinion and Lifestyle survey from the office for National Statistics

There are multiple things we look for in a partner in our lives, and online “dating markets,” such as Hinge, help people to find perfect fits. But then, you face a very common problem that economists call asymmetric information (Akerlof, 1970). This occurs when you do not have full information in an interaction, while the person you interact with does. In the online dating context, you don’t really know the person ‘from Hinge’ as well as they know themselves, do you? While in the past this problem was solved or at least minimised by knowing a person from school, church or a relative, now many interact with strangers on dating platforms. Now, how do you ensure that this person is a good fit for you or isn't lying? Well, you look for a signal!

 

Signalling the Solution

 

Watch closely now, because this is the gist of it. In economics, a signal is a hard-to-fake clue about a hidden quality (Spence, 1973). But a signal only works when it has differential cost: it must be easier for the “right” person to get than for the “wrong” one. If I want a curious partner, a two-hour Coursera certificate tells me very little, because almost anyone can get it. A PhD tells me more, not because it proves curiosity perfectly, but because it is much more costly for a non-curious person to complete.

When this happens, we get a separating equilibrium: different types sort themselves out because the better signal is too costly for the wrong type to copy. But if both the genuinely curious and the merely strategic can show the same signal, then we get a pooling equilibrium: everyone looks similar, and the signal loses value. The same logic applies to wealth. A Justin Bieber mug tells me nothing. A Ferrari tells me more, because it is much harder for the wrong person to obtain. In short, signals help only when they separate; when they pool, they turn into noise.

 

Problems with Signalling.

 

We know very well that signals are not 100% representative of a trait, yet they serve as a selection criterion in our ‘dating game’. As useful as they are, they can also be harmful if stretched to the extreme. This fact creates a second dynamic: an extreme overreliance on signals. A lack of the right signals leaves a potentially perfect partner on mute. In other words, a highly curious person without a degree may not ‘register on my radar’. Even though they have all the hidden qualities, there would be a lack of an easily observable clue. Conversely, people in the non-target group can still fake or obtain a signal: posting a photo with a Ferrari that isn't actually theirs or renting it. Especially if the signal’s value is high, there is a higher incentive to cheat in the dating game by faking it. People might buy a fake watch to signal financial stability or post a fake diploma on their profile (yes, some people even get employed this way!). Economists refer to this as dishonest signalling. When signals can be easily mimicked, people with various qualities may seem indistinguishable. This makes it hard to recognise who is genuine, thus users are usually misled. As trust fails, people base their decisions on expected quality rather than actual quality. High-quality individuals are undervalued, while lower-quality ones seem more appealing. As a result some genuine high-quality users may leave the market. This is what economists call adverse selection, where the fakers push the genuine people entirely out of the market.

The incentive to cheat is high as we live in an age where the signal is so important online. This leads to many people creating fake profiles to scam naive users or generating fake photos of a trip to Bali. If not mitigated, this accelerates adverse selection: people who are not faking it will seem less attractive, while people who do fake it look like they live the most incredibly fun lives. Moreover, this drastically lowers trust in platforms, and the value of any signal plummets. This creates a massive incentive for online dating apps to invest in anti-AI tools. Many dating platforms, like Bumble, have already created fake-image detectors that can flag 95% of fake profiles (Bumble Inc., 2024). Because online dating is largely based on signalling, if the value of all signals goes down, the value of the app dissipates as well. The platforms that do will take over. Here, a larger question arises: was an overreliance on signals by our society, even if tech giants solve the fake signal problem in dating, a good choice to begin with, regardless of what dating apps and tech giants attempt to convince us of? That is not only economic but also a moral question for you.

 

Conclusion.

 

In conclusion, the 21st century offers us increased choice in finding a partner. It offers a wide variety of choices, but it also overemphasises the importance of observable signals that you can capture in a photo. At the same time, with all of this choice now available to us, we are the loneliest we have ever been. So, is the method we decided to use to find our partners actually working? Honestly, we don’t have a definitive answer to this question, but it is something to take home and mull over. We end this blog with more questions than answers, as each person must first answer individually before a societal problem can be solved.

Reference list

Akerlof, G.A. (1970) 'The market for "lemons": Quality uncertainty and the market mechanism', The Quarterly Journal of Economics, 84(3), pp. 488–500.

Bumble Inc. (2024) Bumble Inc. launches Deception Detector™️: An AI-powered shield against spam, scam and fake profiles. Available at: https://www.nasdaq.com/press-release/bumble-inc.-launches-deception-detectortm:-an-ai-powered-shield-against-spam-scam-and (Accessed: 17 April 2026).

Cigna. (2020) Loneliness and the workplace: 2020 U.S. report. Available at: https://www.cigna.com/static/www-cigna-com/docs/about-us/newsroom/studies-and-reports/combatting-loneliness/cigna-2020-loneliness-report.pdf (Accessed: 17 April 2026).

Office for National Statistics (2025) Public opinions and social trends, Great Britain: January 2025. Available at: https://www.ons.gov.uk/peoplepopulationandcommunity/wellbeing/bulletins/publicopinionsandsocialtrendsgreatbritain/january2025

Primack, B.A., Shensa, A., Sidani, J.E., Whaite, E.O., Lin, L.Y., Rosen, D., Colditz, J.B., Radovic, A. and Miller, E. (2017) 'Social media use and perceived social isolation among young adults in the U.S.', American Journal of Preventive Medicine, 53(1), pp. 1–8.

Rosenfeld, M.J., Thomas, R.J. and Hausen, S. (2019) 'Disintermediating your friends: How online dating in the United States displaces other ways of meeting', Proceedings of the National Academy of Sciences, 116(36), pp. 17753–17758.

Spence, M. (1973) 'Job market signaling', The Quarterly Journal of Economics, 87(3), pp. 355–374.