The Turing test (originally called the imitation game) is a test of the ability of humans to differentiate other humans from human imposters. However, as AI advances it becomes clear that the Turing test is not enough. We should use the flea market test to determine if AI can mimic humans. What is a flea market test? Over the last few years, the idea of a flea market test for AI has developed during my visits to flea markets in Belgrade, Geneva, and other places. Last week, I made a video summarising my main observations and findings.
Following enthusiastic reactions to this video, I elaborated more on the need to seriously consider the dynamics of the flea market as the new AI Turing test. First of all, second-hand markets exist everywhere and are often called flea markets because of actual fleas that may still occupy them. Its etymology can be traced back to the French marché aux puces, literally translated as ‘market with fleas’. In English-speaking countries, in addition to the flea market, there are other names: trash & treasure markets in Australia, or car boot sales in the UK. In Spanish-speaking countries, a flea market is often described as el rastro or mercado de pulgas. In Chile, you can find a mercado persia (Persian market) or a feria artesanal (artisan fair) in Argentina. In Serbian, the name is buvljak (originating from the word buva, which means flea). Flea forms the etymological foundation of many terms used locally: Chinese (tiàozǎo shìchǎng), Korean (byeorugsijang), Arabic (souq al barghout), Hindi (phlee market), German (flohmarkt), Dutch (vlooienmarkt), Swedish (loppmarknad), Turkish (bit pazarı), and Finnish (kirpputori). In well-functioning economies, flea markets are marginal and exotic spaces. Flea markets play a more important economic role when economies slide into trouble and the informal economy gains higher relevance. Flea market diversity worldwide could correct the current monochromatic AI debates dominated by narratives developed in a few advanced AI circles. Lack of sensitivity to diversity might be a major problem, as AI continues to impact cultural and societal contexts worldwide. This diversity and local touch can be found in flea markets, from physical artefacts to, more importantly, ways of socialising, bargaining, and perhaps even laughing. Second, flea markets are intensely transactional. They are places for bargaining. Selling and buying are two of the oldest human activities. AI is still behind in grasping the transactional aspects of human interactions. Thirdly, at flea markets, you can find all strata of a society. It is often a place for survival and often the primary source of income for people experiencing poverty. It can be a place for the more affluent to find rare vintage or antique items. For collectors, flea markets are treasure troves. Often, flea markets are places in legal border zones, sometimes selling goods without clear origins, sometimes stolen ones. Fourth, flea markets have very few formal rules. In essence, they are places where you can find the rules of market economy and capitalism in their purest form, as goods and services are exchanged in more direct ways. Most of the modern economy is heavily regulated and, thus, easier to automate using AI. A flea market is much more challenging to write algorithms for. Nuance, tone of voice, and non-verbal communication are not AI’s strong points. Fifth, traders use many tacit rules and skills. As the video shows, AI can understand tacit rules, but interpreting them and putting them into practice isn’t as easy for AI as it is for humans. For example, as a buyer, sometimes you should show little interest. It would be best to strike the right balance between revealing and hiding your interest in a particular product. On timing, it is vital to start early or finish late. Early in the day, traders were keen to make their first sale as a good omen. Later in the day, they want to get a rid of the remaining goods even if at a lower price. Sixth, flea markets are places of storytelling. Traders create stories after analysing potential customers, their needs, and sometimes their dreams. AI won’t be able to compete with flea market sellers in inventing stories based on a complex and very quick analysis of context, information, and emotions. Seventh, flea market transactions involve high levels of emotions and human interactions. In the casual and relaxed atmosphere, communication is vibrant. In addition to verbal communication, traders read each other’s body language and engage in eye contact. Every detail and analysis are re-evaluated in real-time. Intensity and complexity of communication are not reflected in a relaxed and easy-going atmosphere. Eight, flea markets are arenas for persuasion. They involve a very personal activity that starts with building rapport, gathering sufficient information and knowledge, telling good stories, and using good timing. Persuasion requires a high level of emotional intelligence and empathy, which AI does not have and cannot easily mimic. Some specific activities can be automated, but holistic persuasion involving knowledge, intuition, and emotions will remain the ultimate frontier of human uniqueness, at least for some time. Ninth, flea market traders have a certain ethos. Bargaining is the core, while cheating is punished. Ultimately, there is satisfaction in successful negotiations, which hopefully make both parties equally (un)happy. AI regulators can learn from the flea market’s tacit ethics, which are not enforced through formal rules and legislation. The richness of flea market exchanges make it a realistic representation of human conditions with all their emotional, intellectual, and societal shades. Thus, the flea market AI test would be much more reliable than the Turing test in detecting whether the other party bargaining is a fellow human or a machine. In this test, we, as humans, can act as both buyers and sellers. I started this debate with a short video, and above, I listed nine reasons for the flea market AI test. It is my hope that the flea market metaphor can help us better understand both the potential and the risks that AI catalyses. It could inform current debates, emphasising the possibility that AI poses an existential threat to humanity. Paradoxically, flea markets could be our last refuge, where humanity can start developing answers for future AI challenges. I look forward to your comments and suggestions on how we can advance a better understanding of AI and its impact on humanity. 
