Chatting with Chat-GPT4 on EVOO frauds

Yesterday, I spent 15 minutes chatting with the most well-known AI around food fraud in the extra virgin olive oil sector. The conversation will include elements around the prices of such oil, since Italian surveys show that now – at 9-10 €/liter – consumers are shifting to oil mixes and seeds oil to absorb the impact on their wallets.

In this article, you can read our entire chat, so that you can appreciate the good, the bad, and the shortcomings of large language models.

It is a simple example that shows on one side the huge potential of a technology that, at the actual stage, is possibly like the first modem connecting to the internet (do you remember how slow they were? and the creepy sound?!)

On the other hand, it demonstrates how important it is to ask the right questions, challenge the AI, and refine the results you obtain.

The key point for me is that being essentially language models, AI tools should not be used to get straight answers: they are not an encyclopedia, they hallucinate and they could source data from questionable places. They should be used to speed up and refine our writing, to make better questions, to complement our capability to generate new ideas and to analyze problems: and as more you interact with them, the more you will be able to ask the right questions and get better answers.

But I’ll leave you now to the conversation: the paragraphs will be marked by C (= Cesare), AI (=ChatGPT-4), N (= my notes and thoughts, in red font). Please leave your comments and feedback!

* * * * *

C: How would you define “food fraud”?

AI: Food fraud refers to the deliberate and intentional substitution, addition, tampering, or misrepresentation of food, food ingredients, or food packaging, or false or misleading statements made about a product, for economic gain. It encompasses various fraudulent activities, such as adulterating a product (mixing with inferior substances), mislabeling (incorrectly stating the origin, species, or composition), dilution (adding liquid to increase weight), and counterfeiting (creating unauthorized replicas of popular products). These actions can compromise food safety, undermine consumer trust, and cause economic loss to both consumers and producers.

N: so far so good

C: What are the most prevalent kinds of food frauds?

AI: The most prevalent kinds of food fraud typically involve products that are high value, frequently consumed, or easily adulterated. Some common examples include:

1. Olive Oil Fraud: Often, cheaper oils like soybean or sunflower oil are mixed with olive oil to increase profits. This is one of the most common types of food fraud due to the high price and demand for pure olive oil.
2. Honey and Maple Syrup Fraud: These products are frequently adulterated with cheaper forms of sugar or corn syrup to reduce costs while maintaining the appearance and texture of the pure product.
3. Fish and Seafood Fraud: This involves mislabeling a cheaper species of fish as a more expensive one. For example, farmed salmon may be sold as wild, or cheaper white fish varieties may be labeled as more expensive ones like halibut or cod.
4. Spices Fraud: Spices such as saffron, turmeric, and paprika are often diluted with cheaper substances or dyed to appear more potent. Ground spices are particularly susceptible to adulteration since it’s harder to detect the addition of non-spice materials.
5. Meat Fraud: This includes substituting a cheaper type of meat for a more expensive one (e.g., horsemeat sold as beef), adding water or other fillers to increase weight, or selling meat from diseased animals.
6. Milk and Dairy Fraud: Adulteration of milk with water, urea, detergents, or other substances to increase volume or improve appearance is a common fraud. Cheese might also be subject to fraud by using cheaper fats than those declared on the label.
7. Organic Food Fraud: Products may be falsely labeled as organic to fetch higher prices despite not meeting organic standards.
8. Wine and Liquor Fraud: Expensive wines and spirits can be diluted with cheaper alcohol or counterfeit bottles can be filled with lower quality products.

These types of fraud are motivated by economic gain, and they can pose significant health risks to consumers as well as undermine trust in food producers and retailers. Efforts to combat food fraud include improved regulation, better supply chain transparency, and advanced testing methods.

N: nice! Now let’s challenge a bit the smarty pants AI

C: Where do you retrieve such data? How can we establish the prevalence of the above-mentioned frauds?

AI: The data on the prevalence of different types of food fraud is typically gathered from various sources including:

1. Government Reports and Databases: Many governments around the world have agencies dedicated to food safety and standards. These agencies often publish reports on food fraud incidents and maintain databases of food fraud cases. For example, the U.S. Food and Drug Administration (FDA) and the European Union’s Food Fraud Network (FFN) are such entities.
2. Research Studies and Academic Journals: Universities and research institutions conduct studies on food fraud, which are published in academic journals. These studies may involve the analysis of food products on the market, the review of case studies, or the assessment of risk factors associated with different types of food fraud.
3. Industry Reports: Organizations within the food industry, including trade associations and consumer advocacy groups, may conduct their own investigations and issue reports on the state of food fraud within specific sectors.
4. Media Reports: Investigative journalism can uncover instances of food fraud and highlight issues within specific industries or regions.

To establish the prevalence of the mentioned food frauds, researchers and regulators use a variety of methods, including:

• Market Surveys and Product Testing: By sampling products from retail outlets and testing their authenticity, researchers can identify the presence and extent of fraud in different food categories.
• Supply Chain Audits: Inspecting and monitoring the supply chain of food products helps identify vulnerabilities where fraud might occur.
• Data Analysis and Trend Monitoring: Analyzing data from food safety incidents, consumer complaints, and laboratory tests can reveal trends and patterns in food fraud.
• Whistleblower Reports and Tip-offs: Information from industry insiders can also be a valuable source of data on fraudulent practices.

Establishing the prevalence of food fraud requires coordinated efforts across these different approaches to paint a comprehensive picture of the challenges and risks facing the food industry.

N: as you can see the answer is pretty accurate, but not really convincing about data sourcing and analysis. Better to dig deeper

C: Did you analyze such data or did you rely on other author papers and estimates?

AI: The information I provided on the most prevalent kinds of food fraud is synthesized from various sources, including academic research, reports from food safety organizations, and analysis from industry experts. I don’t directly analyze raw data or conduct original research. Instead, I rely on existing literature, studies, and reports to provide a summary of the types of food fraud that are most commonly reported and discussed in the field.

To establish the prevalence of the mentioned frauds, researchers and organizations often conduct studies involving supply chain audits, market surveys, and laboratory testing to detect adulteration or mislabeling. Data can also come from governmental food safety agencies, industry watchdogs, and consumer advocacy groups. They use techniques like DNA barcoding, chemical analysis, and isotopic analysis to test the authenticity of food products. The results of these investigations, along with reports of food fraud incidents, are then compiled and analyzed to identify trends and prevalence rates.

For more specific, up-to-date information, one would typically look at reports from organizations such as the Food Fraud Initiative at Michigan State University, the U.S. Pharmacopeial Convention’s Food Fraud Database, or the European Union’s Food Fraud Network. These entities collect, analyze, and report data on food fraud incidents globally.

N: so, no original analysis, just a summary of existing data, but the sources consulted are indeed the right ones. Now, let’s push a bit the boundaries and see if AI can help me to predict the trends in the sector.

C: On the basis of the data you have and the current prices of commodities (e.g. olive oil), can you predict an increment in frauds?

AI: Regarding your question on analyzing data to predict an increment in frauds based on current prices of commodities like olive oil, direct predictions would require specific data analysis, considering factors such as current market prices, historical trends in food fraud related to price fluctuations, and other economic factors such as supply chain disruptions or changes in demand.

While I don’t analyze raw data or perform original research, predictions about the increase in food fraud incidents can generally be made based on historical trends: when the price of a commodity like olive oil rises significantly, the incentive for fraudulent activity also increases, making fraud more likely if the opportunity exists and the risk of detection is low.

However, predicting exact increases in food fraud without detailed, current market data, and without a sophisticated model analyzing various economic, social, and environmental factors, would be speculative. Regulatory agencies and companies often monitor such trends closely and implement measures to mitigate the risk of fraud, especially in high-risk categories. If you’re interested, I could look up current trends and prices for specific commodities like olive oil to provide a more informed viewpoint.

N: this is a lot to chew.

The AI tells that it cannot do direct predictions, but it’s identifying correctly the variables to keep into account.

I do not completely agree about the fact that historical trends are significant in predictions: they are surely useful to identify what to look for, but – as for investments – much less for predictions. If factors are changing outcomes might diverge greatly.

What is certain is that the correlation between price increase and opportunity for fraud is clear to the AI: I cannot say the same for many regulators and industry people I met and trained.

Most importantly, the AI is offering help to look at trends and prices to provide a more informed viewpoint (spontaneously!) I’ll take it!

C: Yes, please look up current trends and prices for specific commodities like olive oil to provide a more informed viewpoint.

AI: The current trends and prices for olive oil indicate a significant increase, primarily due to challenges such as a prolonged drought in the Mediterranean region, particularly in Spain, and below-average harvests. These factors have kept the prices high due to limited inventory reserves and continued demand from export markets despite a minor decline in local consumption.

The price for Extra Virgin Olive Oil in the Mediterranean region has seen a dramatic increase compared to the previous year. For example, as of early 2024, the price per liter for Italian Extra Virgin Olive Oil is approximately €9.56, reflecting a 58.3% increase from the same period in 2023. Similarly, Spanish Extra Virgin Olive Oil is priced around €9.09 per liter, marking an 83% increase from 2023. Greek olive oil also experienced a significant rise, priced at €9.25 per liter, an 84.7% increase. In Tunisia, the price reached €7.88 per liter, up by 53.7% from the previous year​​​​.

This steep price increase has been largely attributed to adverse weather conditions, including drought and heatwaves, which have negatively impacted the olive oil production cycle in several key producing countries. For example, Spain and Portugal faced extremely dry and hot springs, adversely affecting their olive oil yield. Greece, on the other hand, has been in an off-year of the alternate bearing cycle, leading to a significant production drop.

The rising costs and scarcity have led to a downturn in consumption in traditional olive oil-consuming countries like Spain, where a notable decline in retail purchases for all grades of olive oil has been observed. To address the rising cost of living and the impact of increasing olive oil prices, Spain has made a temporary decision to eliminate VAT on olive oil sales​​​​.

Based on these current trends and prices, it is plausible to predict that if these conditions persist, there could be a further increment in frauds as the incentive for economic gain increases with the rising value of olive oil. High prices combined with scarcity can lead to a rise in adulteration, mislabeling, or other fraudulent activities as unscrupulous individuals and entities may attempt to exploit the situation for profit.

Therefore, consumers and businesses alike should exercise caution and perhaps seek to verify the authenticity and quality of olive oil products, especially during this period of heightened prices and potential scarcity. It’s also crucial for regulatory bodies and the industry to enhance monitoring and enforcement to combat potential fraud and ensure consumer trust and safety.

N: “Fuck” (cit.)! AI got the prices, did a decent economic evaluation, identified the major players in the market, captured the main factors influencing prices and the consumer response, and it is making (not requested) recommendations to all stakeholders and regulators about how to counteract fraud.

Let’s give some purpose to this chat!

C: Can you create a visual for the price trends you identified and the factors that are increasing the risk of fraud in the extra virgin olive oil sector? A graphic, an infographic, a table?

AI: Here is a visual representation of the price trends for Extra Virgin Olive Oil from 2023 to 2024 for Italy, Spain, Greece, and Tunisia, along with the percentage increase in prices for each country. This graph illustrates the significant rise in olive oil prices, which is one of the factors increasing the risk of fraud in the sector. As you can see, there has been a substantial price increase in all four countries, with Spain and Greece showing the highest percentage increases.

N: here comes the trouble. The figures in the graph are not the same reported before by the same AI. The honeymoon ends.

C: The % of increase does not look correct. According to your analysis for Italy was 58.3%

AI: You’re right; there was an inconsistency with the data presented. Let me correct the percentage increases for each country according to the data previously provided:

• Italy had an increase of 58.3%
• Spain had an increase of 83%
• Greece had an increase of 84.7%
• Tunisia had an increase of 53.7%

I’ll correct this and update the visual accordingly.

N: REALLY?! you did all of that and you fuck up to put four numbers in a column?! And you don’t apologize either!

That’s the main challenge with AI today from my perspective. I do not want to enter much into the accuracy of the data (which is another story and should be checked thoroughly as well, but overall in this case the prices identified are quite ok), but the main challenge that I face, even trying to set up my own GPTs is that is lazy! Sometimes miss things that it already has. And this is why everything should be thoroughly checked and – at least for now – I see AI more as like a video game power-up for my avatar capability, than a real agent or substitute. And this is also why you cannot ignore the phenomena: AI will kill all standardized low-quality outputs, but smart and productive people will become smarter and more productive.

For your information, I also asked AI to find prices of the main EVOO brands in the markets included in the graph and to complement the graph with such info. And it happened. Current data were extracted presumably from online stores and analyzed. But I won’t publish them for obvious reasons.

But now, let’s finish our experiment, asking the AI to correct the graph, placing the right % and creating a nice and evocative picture of fraudulent practices in the EVOO sector to close this blog.

That’s all folks!