Tag: probability

Posted on

AI: benefits but catastrophic risk



Artificial intelligence (AI) has become a key technology in the 21st Century. Businesses use AI systems to analyse data, automate routine tasks, improve customer service, write software, create content and even assist with decision-making. As AI improves and becomes even more capable, governments, economists, businesses and society are debating its effects on productivity, employment and economic growth – and also its potentially extreme dangers.

Some commentators compare AI to earlier technological revolutions such as the steam engine, electricity and the Internet. They argue that AI will continue to increase productivity, create new industries and improve living standards. Others worry that AI may eliminate a large numbers of jobs, increase inequality, concentrate economic power in a few firms and, in the most extreme scenarios, pose a threat to humanity itself.

These debates raise important questions including how society should evaluate the potential risks of a technology whose benefits may be enormous but whose long-term consequences remain uncertain.

AI and the labour market

Historically, technological change has had both positive and negative effects on employment. Automation reduced the demand for many agricultural workers, while creating jobs in manufacturing. Computers automated clerical tasks but generated entirely new industries in software, telecommunications and digital services.

AI appears likely to follow a similar pattern. According to the World Economic Forum, advances in AI, robotics and information-processing technologies are expected to transform labour markets significantly during the second half of the 2020s, creating demand for new skills while reducing demand for others. The fastest-growing skills are expected to include AI and big data, technological literacy and cybersecurity.

The potential benefits

AI may benefit labour markets in several ways:

  • Higher productivity: Workers can complete tasks more quickly with AI assistance.
  • New occupations: Demand has emerged for AI engineers, prompt specialists, data scientists and AI governance professionals.
  • Better decision-making: Firms can use AI to improve forecasting, inventory management and customer service.
  • Complementing human skills: AI may perform repetitive tasks, allowing employees to focus on creativity, problem-solving and interpersonal activities.
  • Economic growth: Higher productivity can increase profits, wages and living standards over time.

Many economists argue that AI will not simply replace workers but will change the tasks they perform. Research from the OECD suggests that even highly AI-exposed occupations continue to require management, communication, collaboration and social skills that technologies struggle to replicate.

The potential costs

At the same time, AI may create significant labour-market challenges. Many white-collar occupations previously considered relatively safe from automation are becoming vulnerable. Generative AI systems can draft reports, analyse legal documents, write computer code and create marketing content. This means that some professional and administrative roles may face considerable disruption.

The World Economic Forum reports that business leaders have differing expectations about the effects of AI. In a 2026 survey (see link below), around 54 per cent expected AI to displace existing jobs, while only 24 per cent expected it to create new jobs within their organisations. Economists have identified several potential problems:

  • Structural unemployment: workers in industries that are in decline may struggle to find employment requiring their existing skills.
  • Increased income inequality: there may be a growing income disparity between highly skilled and less-skilled workers.
  • Growing market power: the largest technology firms that own the most advanced AI systems may see their market power grow further creating dominance in certain areas.
  • Regional inequalities: AI-related investment may become concentrated in particular cities and countries, exacerbating regional inequalities within and between countries.
  • Pressure on governments: to address issues of structural unemployment and increasing inequality, governments may be forced to expand retraining and social-support programmes.

AI and catastrophic risk

Most economic discussion around AI focuses on employment and productivity. However, some researchers argue that the most significant risks from AI may be much broader.

Economists distinguish between ordinary risks and catastrophic risks. Catastrophic risks involve events with a very low probability of occurring but potentially enormous consequences. Examples include nuclear accidents, pandemics and certain climate-related disasters.

AI raises similar concerns. Advanced AI systems could potentially be used to conduct cyberattacks, spread misinformation, disrupt critical infrastructure or support the development of dangerous technologies. Some researchers have even suggested that highly advanced AI systems could pose an existential risk to humanity if they become sufficiently powerful and are not properly controlled.

This became a widely discussed topic in the media in September 2026 following the resignation of an employee, Jacob Coxon, from AI firm, Anthropic. He said that people working on AI were ‘genuinely frightened’ about how quickly AI was advancing and what it might mean for the future of humanity. He said:

I believe that if we don’t slow down at the current rate of progress, there is a strong chance that we could all die in the immediate future.

Other researchers have raised similar concerns and there have since been calls from some of the biggest AI companies for regulation of AI to prevent this.

The policy debate and CBA

All of this creates a challenge for cost-benefit analysis. Suppose AI generates trillions of pounds of economic benefits. But, if there is also a very small probability of catastrophic harm, how should policymakers weigh up the two?

Traditional cost-benefit analysis values risk by multiplying the size of a potential outcome by its probability. However, this approach becomes problematic when both the probability and the consequences are highly uncertain. The risks may be extremely difficult to estimate, while the potential costs could be vast and affect future generations. For this reason, governments and firms increasingly use scenario analysis, stress testing and AI safety assessments to evaluate potential risks. These approaches attempt to prepare for extreme outcomes rather than relying solely on probability calculations.

Supporters of AI argue that technological progress has historically improved living standards and that restricting AI too heavily could reduce innovation and economic growth. Critics argue that uncertainty about potentially catastrophic outcomes justifies a more cautious approach.

The debate therefore extends beyond labour economics to the wider issue of managing catastrophic risk. As with nuclear power or climate change, policymakers must decide how much risk society is willing to accept in exchange for potentially large economic benefits.

Articles

Reports

Questions

  1. How might AI increase productivity while also causing unemployment in some sectors? Which sectors are likely to be affected the most?
  2. Why is it difficult to estimate the costs and benefits of advanced AI?
  3. Assume that a disaster is estimated to cost society £1000 billion (£1 000 000 000 000). The chances of the disaster occurring are said to be minute, however. Estimates vary from a probability of one in a million to one in a billion. What estimate of this cost would you include in a cost–benefit analysis?
  4. Why are many low-income countries apparently prepared to accept riskier projects than are high-income ones?
  5. Discuss whether the greatest economic challenge posed by AI is (a) job displacement; (b) increased inequality; (c) market concentration and the power of large technology firms; or (d) catastrophic long-term risks.
  6. Read the article by Bill Gates, The turbulent AI era is here. The choices we make now are critical. According to him, what steps should the world take to ensure that ‘AI will be a force for good and leave everyone better off’?
Posted on

Game Theory: Rafa is the King

Together with Formula 1, tennis is the other sport I love – and my favourite player by far is Rafa!

We often apply game theory to various sports and consider how players, teams and individuals can think strategically. One of the big debates in tennis is ‘who is the best ever’ and I argue that Nadal is the greatest, based on a huge range of metrics.

I saw this article in the Economist, providing analysis and comparison between some of the best tennis players. It shows how we can use economic thinking, probability, game theory and analysis to come to something of an answer about who is the greatest, considering the various players’ runs to the title in the Grand Slams. Of course the reason I’m posting this is because according to the Economist, Rafa is the best! And the reasoning is very sound. Enjoy. I certainly did.

Sorry Roger: Rafael Nadal is not just the King of Clay The Economist (13/09/17)

Questions

  1. What is game theory and why is it useful?
  2. How does the rating system aim to measure the skill of a tennis champion?
  3. In this particular scenario, why is it important to use probabilities?
  4. We can use game theory to think about penalty shoot outs and whether footballers play to the Nash equilibrium. Can we also use the Nash equilibrium when thinking about tennis? (Think about the serve!)
Posted on

A flood of insurance concerns

Next year a government agreement with insurance companies is set to end. This agreement requires insurance companies to provide cover for homes at a high risk of flooding.

However, in June 2013, this agreement will no longer be in place and this has led to mounting concerns that it will leave thousands of home-owners with the inability either to find or afford home insurance.

The key thing with insurance is that in order for it to be provided privately, certain conditions must hold. The probability of the event occurring must be less than 1 – insurance companies will not insure against certainty. The probability of the event must be known on aggregate to allow insurance companies to calculate premiums. Probabilities must be independent – if one person makes a claim, it should not increase the likelihood of others making claims.

Finally, there should be no adverse selection or moral hazard, both of which derive from asymmetric information. The former occurs where the person taking out the insurance can hide information from the company (i.e. that they are a bad risk) and the latter occurs when the person taking out insurance changes their behaviour once they are insured. Only if these conditions hold or there are easy solutions will the private market provide insurance.

On the demand-side, consumers must be willing to pay for insurance, which provides them with protection against certain contingencies: in this case against the cost of flood damage. Given the choice, rational consumers will only take out an insurance policy if they believe that the value they get from the certainty of knowing they are covered exceeds the cost of paying the insurance premium. However, if the private market fails to offer insurance, because of failures on the supply-side, there will be major gaps in coverage.

Furthermore, even if insurance policies are offered to those at most risk of flooding, the premiums charged by the insurance companies must be high enough to cover the cost of flood damage. For some homeowners, these premiums may be unaffordable, again leading to gaps in coverage.

In light of the agreement coming to an end next year, there is pressure on the government firstly to ensure that insurance cover is available to everyone at affordable prices and secondly to continue to build up flood defences in the most affected areas. Not an easy task given the budget cuts. The following articles provide some of the coverage of the problems of insuring against flood damage.

Articles

200,000 homes ‘at flooding risk’ BBC News (3/1/12)
MPs slam government flood defences Post Online, Chris Wheal (31/1/12)
Flooding: 200,000 houses at risk of being uninsurable The Telegraph (31/1/12)
Flood defences hit by government cuts ‘mismatch’, says MP Guardian, Damian Carrington (31/1/12)
Fears over cash for flood defences The Press Association (31/1/12)
ABI refuses to renew statement of principles for flood insurance Insurance Age, Emmanuel Kenning (31/1/12)

Questions

  1. Consider the market for insurance against flood damage. Are risks less than one? Explain your answer
  2. Explain whether or not the risk of flooding is independent.
  3. Are the problems of moral hazard and adverse selection relevant in the case of home insurance against flood damage?
  4. If ABI doesn’t put in place another agreement to provide insurance to homeowners at most risk of flooding, what could be the adverse economic consequences?
  5. Is there an argument for the government stepping in to provide insurance itself?
  6. Explain why insurance premiums are so much higher for those at most risk of flooding. Is it equitable?
Posted on

Should cyclists have insurance?

Most people are risk-averse: we like certainty and are generally prepared to pay a premium for it. The reason is that certainty gives us positive marginal utility and so as long as the price of insurance (which gives us certainty) is less than the price we place on certainty, we will be willing to pay a positive premium. By having insurance, we know that should the unexpected happen, someone else will cover the risk. As long as there are some risk-averse people, there will always be a demand for insurance.

However, will private companies will be willing to supply it? For private market insurance to be efficient, 5 conditions must hold:

1. Probabilities must be independent
2. Probabilities must be less than one
3. Probabilities must be known or estimable
4. There must be no adverse selection
5. There must be no moral hazard

If these conditions hold or if there are simple solutions, then insurance companies will be willing and able to provide insurance at a price consumers are willing to pay.

There are many markets where we take out insurance – some of them where insurance is compulsory, including home and car insurance. However, one type of insurance that is not compulsory is that for cyclists. No insurance is needed to cycle on the road, but with cycle use increasing and with that the number of accidents involving cyclists also increasing, the calls for cyclists to have some type of insurance is growing. If they are hit by someone without insurance and perhaps suffer from a loss of income; or if they cause vehicle damage, they will receive no compensation. However, whilst the risk of accident is increasing for cyclists, they are still statistically less likely to cause an accident than motorists. Perhaps a mere £30 or £40 per year for a policy is a price worth paying to give cyclists certainty. At least, this is what the Association of British Insurers (ABI) is claiming – hardly surprising when their members made a combined loss of £1.2 billion!

Articles

Cyclists ‘urged to get insurance’ BBC News, Maleen Saeed (26/11/11)
Cyclists urged to get more insurance by … insurance companies Road.CC, Tony Farrelly (26/11/11)
The future of cycle insurance Environmental Transport Assocaition (24/11/11)

Questions

  1. With each of the above conditions required for private insurance to be possible, explain why each must hold.
  2. What do we mean by no moral hazard and no adverse selection? Why would their existence prevent a private company from providing insurance?
  3. Using the concept of marginal utility theory, explain why there is a positive demand insurance.
  4. What might explain why cyclists are less likely to take out insurance given your answer to the above question?
  5. Do you think cyclist insurance should be compulsory? If governments are trying to encourage more sustainable transport policy, do you think this is a viable policy?