{"id":5587,"date":"2024-12-10T12:26:14","date_gmt":"2024-12-10T11:26:14","guid":{"rendered":"https:\/\/www.rivistaeco.com\/?p=5587"},"modified":"2024-12-10T12:26:14","modified_gmt":"2024-12-10T11:26:14","slug":"what-happens-when-ai-systems-interact-with-each-other","status":"publish","type":"post","link":"https:\/\/www.rivistaeco.com\/en\/2024\/12\/10\/what-happens-when-ai-systems-interact-with-each-other\/","title":{"rendered":"What Happens When AI Systems Interact with Each Other?"},"content":{"rendered":"

In the race to develop artificial intelligence (AI) that supports society, particular attention must be paid to scenarios where multiple algorithms interact with each other. Addressing this requires an interdisciplinary approach spanning computer science and economics, fostering a serious dialogue about the desired outcomes that should guide the evolution of these systems.<\/span><\/i><\/p>\n

News reports describe residents of a San Francisco neighborhood complaining of sleepless nights caused by incessant honking from the street. At first glance, nothing unusual\u2014except for the nature of the impatient drivers. The noise, it turned out, came from a fleet of self-driving taxis belonging to a well-known American company.<\/span><\/p>\n

The situation was paradoxical: various autonomous cars were engaged in parking maneuvers, “arguing” through their horns, yet no humans were driving. This somewhat amusing incident highlights an important aspect of AI development. As AI systems become pervasive, scenarios increasingly arise where multiple “artificial agents” must interact with one another and a complex external environment.<\/span><\/p>\n

An agent can be thought of as an entity capable of perceiving its surroundings and making decisions based on this information. In the San Francisco example, each autonomous car was an independent artificial agent. The ideal outcome of their interaction was simple: orderly parking without collisions or disturbing the neighborhood. However, as the incident demonstrates, achieving this in practice is more challenging than it seems.<\/span><\/p>\n

The Challenges of Governing Multi-Agent Systems<\/b><\/h3>\n

With the continuous advancements in AI, reports of new and surprising achievements dominate the news. Yet, difficulties emerge when multiple artificial agents need to interact. Most AI systems are trained to develop cognitive skills in isolated, stable environments\u2014settings that do not change significantly over time. When such agents are deployed in different contexts or in dynamically changing environments, results may diverge from expectations.<\/span><\/p>\n

The human brain enables us to “read” novel situations based on prior knowledge and experiences. Equipping AI systems, which rely on statistical learning, with this generalization capability remains a key research challenge.<\/span><\/p>\n

The complexity increases when the environment features multiple artificial agents, each potentially pursuing different objectives and acting strategically. These agents must evaluate the impact of their actions not only on their goals but also on the objectives of other agents and the broader environment. It is crucial to understand how these “algorithmic” decisions influence the overall outcome of a multi-agent system\u2019s evolution. In extreme cases, an AI agent may face entities with diametrically opposed goals, functioning as adversaries. Interactions of this “zero-sum” type\u2014where one wins and the other loses\u2014require AI agents to be designed for satisfactory performance even under worst-case scenarios, not just typical or optimistic ones.<\/span><\/p>\n

A Multi-Agent System Example: Financial Markets<\/b><\/h3>\n

Scenarios involving multiple artificial agents abound and often significantly impact society. Consider financial markets, where over 70% of trades on U.S. stock exchanges are estimated to be executed by automated trading algorithms. These algorithms make decisions in milliseconds, reacting to events and other agents\u2019 decisions faster than any human trader could. High-frequency trading capabilities and parallelized actions on a large scale enable these systems to adopt strategies qualitatively different from those of human traders. Consequently, markets may evolve along unexpected and unpredictable trajectories.<\/span><\/p>\n

Under certain conditions, high-frequency trading can create inefficiencies in financial markets. The algorithms’ behavior is determined by myriad parameters chosen based on historical data and limited future scenarios. This raises a critical question: how would a market with extensive participation from trading algorithms react to unforeseen situations not accounted for during the algorithms’ optimization? Could the interaction among AI algorithms increase systemic risk, potentially leading to market collapse? Designing “economic platforms” to effectively mediate these interactions is essential, along with rethinking economic institutions to manage interactions among artificial agents.<\/span><\/p>\n

The Solution Lies in an Interdisciplinary Approach<\/b><\/h3>\n

This challenge is twofold. From the perspective of individual agents, we must design AI algorithms capable of maintaining high performance even in unexpected scenarios and in the presence of strategic interactions with other entities. From a system-wide perspective, orchestrating the behavior of individual agents to achieve socially desirable global outcomes is critical.<\/span><\/p>\n

A fundamental starting point is to formally describe each entity’s objectives within a multi-agent system and calculate the best strategies to achieve them. To this end, AI researchers often borrow ideas and models from economic theory. Many economic models assume that actors make rational choices to maximize their goals based on their context. While this assumption may fall short in describing human behavior, it effectively models artificial agents. The branch of economics studying mathematical models of strategic interaction among rational agents is game theory.<\/span><\/p>\n

Enter Game Theory<\/b><\/h3>\n

Computational game theory allows us to predict the evolution of a multi-agent system and study its effects. It also facilitates the analysis and design of what David Parkes and Michael Wellman described in a 2015 article as \u201cAI economies,\u201d multi-agent economic systems where artificial agents execute transactions with other entities. These AI economies can be governed by rules and incentive systems much more complex than those designed for systems intended for humans. Artificial agents possess substantial computational power and are not subject to the cognitive limitations or bounded rationality that human decision-makers often face.<\/span><\/p>\n

The designer of an AI economic system cannot directly control the behavior and functioning of the AI agents participating in the system but can define the rules and incentives governing the system and the agents’ interactions. The \u201crules of the game\u201d can be designed to indirectly influence the choices made by artificial agents, ensuring that the system evolves toward a desired state. However, determining what constitutes desirable states to be pursued at the system level is a complex process that must involve all stakeholders in the system.<\/span><\/p>\n

To enable these kinds of AI economic systems, it will be necessary to foster collaboration among researchers, industry professionals, and policymakers. An initial step in this direction can be seen in the European Union’s Artificial Intelligence Act, which provides a framework for regulating AI systems.<\/span><\/p>\n

Returning to the example of financial markets, a system-level objective could be to mitigate the competitive advantage linked to the speed at which transactions are executed. The system designer cannot directly intervene in the algorithms of autonomous trading systems, as these are controlled by private companies. However, they can slow down the market in which the algorithms interact. How? By introducing new \u201crules of the game\u201d to disincentivize massive investments aimed at executing transactions with ultra-low latency. For instance, as suggested by economists Eric Budish, Peter Cramton, and John Shim in a 2015 article, a simple intervention could involve requiring all buy and sell orders to be grouped and executed together once per second. This straightforward rule would force all transactions to occur according to a “market clock,” approximately once per second, rendering any further reductions in algorithmic latency unnecessary.<\/span><\/p>\n

Challenges for the Future<\/b><\/h3>\n

Many systems of critical importance to society will be characterized by interactions among different AI algorithms, each developed with potentially misaligned objectives. In this context, it is necessary to act on two fronts. From the perspective of the individual agent, it will be essential to develop algorithms capable of ensuring good performance and achieving their objectives even in complex environments shared with other entities operating within the same context. On the other hand, from a system-wide perspective, tools will be needed to predict and control the evolution of multi-agent systems.<\/span><\/p>\n

In this case, one often has a useful “superpower”: the ability to determine the rules governing the environment in which the agents operate. The ultimate goal is to design AI economic systems whose effects are beneficial to society. To achieve this, it will be necessary to study problems that lie at the intersection of computer science and economics. Additionally, a profound dialogue must be established regarding the right outcomes toward which the evolution of these systems should be directed.<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"

In the race to develop artificial intelligence (AI) that supports society, particular attention must be paid to scenarios where multiple algorithms interact with each other. […]<\/p>\n","protected":false},"author":9026,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[1],"tags":[],"coauthors":[199],"class_list":["post-5587","post","type-post","status-publish","format-standard","hentry","category-non-categorizzato"],"acf":[],"yoast_head":"\nWhat Happens When AI Systems Interact with Each Other? - Rivista Eco<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.rivistaeco.com\/en\/2024\/12\/10\/what-happens-when-ai-systems-interact-with-each-other\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"What Happens When AI Systems Interact with Each Other? - Rivista Eco\" \/>\n<meta property=\"og:description\" content=\"In the race to develop artificial intelligence (AI) that supports society, particular attention must be paid to scenarios where multiple algorithms interact with each other. […]\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.rivistaeco.com\/en\/2024\/12\/10\/what-happens-when-ai-systems-interact-with-each-other\/\" \/>\n<meta property=\"og:site_name\" content=\"Rivista Eco\" \/>\n<meta property=\"article:published_time\" content=\"2024-12-10T11:26:14+00:00\" \/>\n<meta name=\"author\" content=\"Andrea Celli\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Andrea Celli\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"7 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/www.rivistaeco.com\\\/en\\\/2024\\\/12\\\/10\\\/what-happens-when-ai-systems-interact-with-each-other\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.rivistaeco.com\\\/en\\\/2024\\\/12\\\/10\\\/what-happens-when-ai-systems-interact-with-each-other\\\/\"},\"author\":{\"name\":\"Andrea Celli\",\"@id\":\"https:\\\/\\\/www.rivistaeco.com\\\/#\\\/schema\\\/person\\\/466b466f6dfe2a36303ffe73dcca12c1\"},\"headline\":\"What Happens When AI Systems Interact with Each Other?\",\"datePublished\":\"2024-12-10T11:26:14+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/www.rivistaeco.com\\\/en\\\/2024\\\/12\\\/10\\\/what-happens-when-ai-systems-interact-with-each-other\\\/\"},\"wordCount\":1306,\"articleSection\":[\"Non categorizzato\"],\"inLanguage\":\"en-US\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/www.rivistaeco.com\\\/en\\\/2024\\\/12\\\/10\\\/what-happens-when-ai-systems-interact-with-each-other\\\/\",\"url\":\"https:\\\/\\\/www.rivistaeco.com\\\/en\\\/2024\\\/12\\\/10\\\/what-happens-when-ai-systems-interact-with-each-other\\\/\",\"name\":\"What Happens When AI Systems Interact with Each Other? - Rivista Eco\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.rivistaeco.com\\\/#website\"},\"datePublished\":\"2024-12-10T11:26:14+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/www.rivistaeco.com\\\/#\\\/schema\\\/person\\\/466b466f6dfe2a36303ffe73dcca12c1\"},\"breadcrumb\":{\"@id\":\"https:\\\/\\\/www.rivistaeco.com\\\/en\\\/2024\\\/12\\\/10\\\/what-happens-when-ai-systems-interact-with-each-other\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/www.rivistaeco.com\\\/en\\\/2024\\\/12\\\/10\\\/what-happens-when-ai-systems-interact-with-each-other\\\/\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/www.rivistaeco.com\\\/en\\\/2024\\\/12\\\/10\\\/what-happens-when-ai-systems-interact-with-each-other\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"http:\\\/\\\/www.rivistaeco.com\\\/en\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"What Happens When AI Systems Interact with Each Other?\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/www.rivistaeco.com\\\/#website\",\"url\":\"https:\\\/\\\/www.rivistaeco.com\\\/\",\"name\":\"Rivista Eco\",\"description\":\"\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/www.rivistaeco.com\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/www.rivistaeco.com\\\/#\\\/schema\\\/person\\\/466b466f6dfe2a36303ffe73dcca12c1\",\"name\":\"Andrea Celli\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/f3b0f239267fc42b966f67449848f9dad6353ee89ebbbdfce07e7bdcb41bcf8b?s=96&d=mm&r=gcdfa42a2645678eb465e760504fa6425\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/f3b0f239267fc42b966f67449848f9dad6353ee89ebbbdfce07e7bdcb41bcf8b?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/f3b0f239267fc42b966f67449848f9dad6353ee89ebbbdfce07e7bdcb41bcf8b?s=96&d=mm&r=g\",\"caption\":\"Andrea Celli\"},\"url\":\"https:\\\/\\\/www.rivistaeco.com\\\/en\\\/author\\\/acelli\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"What Happens When AI Systems Interact with Each Other? - Rivista Eco","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.rivistaeco.com\/en\/2024\/12\/10\/what-happens-when-ai-systems-interact-with-each-other\/","og_locale":"en_US","og_type":"article","og_title":"What Happens When AI Systems Interact with Each Other? - Rivista Eco","og_description":"In the race to develop artificial intelligence (AI) that supports society, particular attention must be paid to scenarios where multiple algorithms interact with each other. […]","og_url":"https:\/\/www.rivistaeco.com\/en\/2024\/12\/10\/what-happens-when-ai-systems-interact-with-each-other\/","og_site_name":"Rivista Eco","article_published_time":"2024-12-10T11:26:14+00:00","author":"Andrea Celli","twitter_card":"summary_large_image","twitter_misc":{"Written by":"Andrea Celli","Est. reading time":"7 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.rivistaeco.com\/en\/2024\/12\/10\/what-happens-when-ai-systems-interact-with-each-other\/#article","isPartOf":{"@id":"https:\/\/www.rivistaeco.com\/en\/2024\/12\/10\/what-happens-when-ai-systems-interact-with-each-other\/"},"author":{"name":"Andrea Celli","@id":"https:\/\/www.rivistaeco.com\/#\/schema\/person\/466b466f6dfe2a36303ffe73dcca12c1"},"headline":"What Happens When AI Systems Interact with Each Other?","datePublished":"2024-12-10T11:26:14+00:00","mainEntityOfPage":{"@id":"https:\/\/www.rivistaeco.com\/en\/2024\/12\/10\/what-happens-when-ai-systems-interact-with-each-other\/"},"wordCount":1306,"articleSection":["Non categorizzato"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/www.rivistaeco.com\/en\/2024\/12\/10\/what-happens-when-ai-systems-interact-with-each-other\/","url":"https:\/\/www.rivistaeco.com\/en\/2024\/12\/10\/what-happens-when-ai-systems-interact-with-each-other\/","name":"What Happens When AI Systems Interact with Each Other? - Rivista Eco","isPartOf":{"@id":"https:\/\/www.rivistaeco.com\/#website"},"datePublished":"2024-12-10T11:26:14+00:00","author":{"@id":"https:\/\/www.rivistaeco.com\/#\/schema\/person\/466b466f6dfe2a36303ffe73dcca12c1"},"breadcrumb":{"@id":"https:\/\/www.rivistaeco.com\/en\/2024\/12\/10\/what-happens-when-ai-systems-interact-with-each-other\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.rivistaeco.com\/en\/2024\/12\/10\/what-happens-when-ai-systems-interact-with-each-other\/"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.rivistaeco.com\/en\/2024\/12\/10\/what-happens-when-ai-systems-interact-with-each-other\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"http:\/\/www.rivistaeco.com\/en\/"},{"@type":"ListItem","position":2,"name":"What Happens When AI Systems Interact with Each Other?"}]},{"@type":"WebSite","@id":"https:\/\/www.rivistaeco.com\/#website","url":"https:\/\/www.rivistaeco.com\/","name":"Rivista Eco","description":"","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.rivistaeco.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Person","@id":"https:\/\/www.rivistaeco.com\/#\/schema\/person\/466b466f6dfe2a36303ffe73dcca12c1","name":"Andrea Celli","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/secure.gravatar.com\/avatar\/f3b0f239267fc42b966f67449848f9dad6353ee89ebbbdfce07e7bdcb41bcf8b?s=96&d=mm&r=gcdfa42a2645678eb465e760504fa6425","url":"https:\/\/secure.gravatar.com\/avatar\/f3b0f239267fc42b966f67449848f9dad6353ee89ebbbdfce07e7bdcb41bcf8b?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/f3b0f239267fc42b966f67449848f9dad6353ee89ebbbdfce07e7bdcb41bcf8b?s=96&d=mm&r=g","caption":"Andrea Celli"},"url":"https:\/\/www.rivistaeco.com\/en\/author\/acelli\/"}]}},"jetpack_featured_media_url":"","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/www.rivistaeco.com\/en\/wp-json\/wp\/v2\/posts\/5587","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.rivistaeco.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.rivistaeco.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.rivistaeco.com\/en\/wp-json\/wp\/v2\/users\/9026"}],"replies":[{"embeddable":true,"href":"https:\/\/www.rivistaeco.com\/en\/wp-json\/wp\/v2\/comments?post=5587"}],"version-history":[{"count":1,"href":"https:\/\/www.rivistaeco.com\/en\/wp-json\/wp\/v2\/posts\/5587\/revisions"}],"predecessor-version":[{"id":5588,"href":"https:\/\/www.rivistaeco.com\/en\/wp-json\/wp\/v2\/posts\/5587\/revisions\/5588"}],"wp:attachment":[{"href":"https:\/\/www.rivistaeco.com\/en\/wp-json\/wp\/v2\/media?parent=5587"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.rivistaeco.com\/en\/wp-json\/wp\/v2\/categories?post=5587"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.rivistaeco.com\/en\/wp-json\/wp\/v2\/tags?post=5587"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/www.rivistaeco.com\/en\/wp-json\/wp\/v2\/coauthors?post=5587"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}