On Natural Language Processing, Game Theory, and Diplomacy

Beyond GPT in its different evolutions, there are other LLMs -as stated in Large Language Models  (LLMs): an Ontological Leap in AI– developed with a perfectly defined industry focus in mind. This is the case of CICERO.  

In November 2022, the Meta Fundamental AI Research Diplomacy Team (FAIR) and researchers from other academic institutions published the seminal paper Human-level Play in the Game of Diplomacy by Combining Language Models with Strategic Reasoning, laying the foundations for CICERO. 

CICERO is an AI agent that can use language to negotiate, persuade, and work with people to achieve strategic goals similar to the way humans do. It was the first AI to achieve human-level performance in the strategy game No-press Diplomacy. 

No-press Diplomacy is a complex strategy game, involving both cooperation and competition, that has served as a benchmark for multi-agent AI research. It is a 7-player zero-sum cooperative/competitive board game, featuring simultaneous moves and a heavy emphasis on negotiation and coordination. In the game a map of Europe is divided into 75 provinces. 34 of these provinces contain supply centers, and the goal of the game is for a player to control a majority (18) of the SCs. Each players begins the game controlling three or four supply centers and an equal number of units. Importantly, all actions occur simultaneously: players write down their orders and then reveal them at the same time. This makes Diplomacy an imperfect-information game in which an optimal policy may need to be stochastic in order to prevent predictability. 

Diplomacy is a game about people rather than pieces. It is designed in such a way that cooperation with other players is almost essential to achieve victory, even though only one player can ultimately win. It requires players to master the art of understanding other people’s motivations and perspectives; to make complex plans and adjust strategies; and then to use natural language to reach agreements with other people and to persuade them to form partnerships and alliances.

How Was Cicero Developed by FAIR?

In two-player zero-sum (2p0s) settings, principled self-play algorithms ensures that a player will not lose in expectation regardless of the opponent’s strategy, as exposed by John von Neumann in 1928 in his work Zur Theorie der Gesellschaftsspiele.

Theoretically, any finite 2p0s game -such as chess, go, or poker- can be solved via self-play given sufficient computing power and memory. However, in games involving cooperation, self-play alone no longer guarantees good performance when playing with humans, even with infinite computing power and memory. The clearest example of this is language. A self-play agent trained from scratch without human data in a cooperative game involving free-form communication channels would almost certainly not converge to using English, for instance, as the medium of communication. Owing to this, the afore-mentioned researchers developed a self-play reinforcement learning algorithm -named RL-DiL-piKL-, that provided a model of human play while simultaneously training an agent that responds well to this human model. The RL-DiL-piKL was used to train an agent, named Diplodocus. In a 200-game No-press Diplomacy tournament involving 62 human participants, two Diplodocus agents both achieved a higher average score than all other participants who played more than two games, and ranked first and third according to an Elo rating system -a method for calculating the relative skill levels of players in zero-sum games.

Which Are the Implications of this Breakthrough?

Despite almost silenced by the advent of GPT in its different versions, firstly this is an astonishing advance in the field of negotiation, and more particularly in the realm of diplomacy. Never an AI model has had such a brilliant performance in a fuzzy environment, seasoned by information asymmetries, common sense reasoning, ambiguous natural language, and statistical modeling. Secondly and more importantly, this is another evidence we are in a completely new AI era in which machines can and are scaling knowledge. 

These LLMs have caused a deep shift: we went from attempting to encode human-distilled insights into machines to delegating the learning process itself to machines. AI is ushering in a world in which decisions are made in three primary ways: by humans (which is familiar), by machines (which is becoming familiar), and by collaboration between humans and machines (which is not only unfamiliar but also unprecedented). We will begin to give AI fewer specific instructions about how exactly to achieve the goals we assign it. Much more frequently we will present AI with ambiguos goals and ask: “How, based on your conclusions, should we proceed?”

AI promises to transform all realms of human experience. And the core of its transformations will ultimately occur at the philosophical level, transforming how humans understand reality and our roles within it. In an age in which machines increasingly perform tasks only humans used to be capable of: what, then, will constitute our identity as human beings? 

With the rise of AI, the definition of the human role, human aspirations, and human fulfillment will change. For humans accustomed to monopoly on complex intelligence, AI will challenge self-perception. To make sense of our place in this world, our emphasis may need to shift from the centrality of human reason to the centrality of human dignity and autonomy. Human-AI collaboration does not occur between peers. Our task will be to understand the transformations that AI brings to human experience, the challenges it presents to human identity, and which aspects of these developments require regulation or counterbalancing by other human commitments.

The AI revolution has come to stay. Unless we develop new concepts to explain, interpret, and organize its consequent transformations, we will be unprepared to navigate them. We must rely on our most solid resources -reason, moral and ethical values, tradition…- to adapt our relationship with reality so it keeps on being human. 

Large Language Models (LLMs): an Ontological Leap in AI

More than the quasi-human interaction and the practically infinite use cases that could be covered with it, OpenAI’s ChatGPT has provided an ontological jolt of a depth that transcends the realm of AI itself.

Large language models (LLMs), such as GPT-3, YUAN 1.0, BERT, LaMDA, Wordcraft, HyperCLOVA, Megatron-Turing Natural Language Generation, or PanGu-Alpha represent a major advance in artificial intelligence and, in particular, toward the goal of human-like artificial general intelligence. LLMs have been called foundational models; i.e., the infrastructure that made LLMs possible –the combination of enormously large data sets, pre-trained transformer models, and the requirement of significant computing power– is likely to be the basis for the first general purpose AI technologies.

In May 2020, OpenAI released GPT-3 (Generative Pre-trained Transformer 3), an artificial intelligence system based on deep learning techniques that can generate text. This analysis is done by a neural network, each layer of which analyzes a different aspect of the samples it is provided with; e.g., meanings of words, relations of words, sentence structures, and so on. It assigns arbitrary numerical values to words and then, after analyzing large amounts of texts, calculates the likelihood that one particular word will follow another. Amongst other tasks, GPT-3 can write short stories, novels, reportages, scientific papers, code, and mathematical formulas. It can write in different styles and imitate the style of the text prompt. It can also answer content-based questions; i.e., it learns the content of texts and can articulate this content. And it can grant as well concise summaries of lengthy passages.

OpenAI and the likes endow machines with a structuralist equipment: a formal logical analysis of language as a system in order to let machines participate in language. GPT-3 and other transformer-based language models stand in direct continuity with the linguist Saussure’s work: language comes into view as a logical system to which the speaker is merely incidental. These LLMs give rise to a new concept of language, implicit in which is a new understanding of human and machine. OpenAI, Google, Facebook, or Microsoft effectively are indeed catalyzers, which are triggering a disruption in the old concepts we have been living by so far: a machine with linguistic capabilities is simply a revolution.

Nonetheless, critiques have appeared as well against LLMs. The usual one is that no matter how good they may appear to be at using words, they do not have true language; based on the primeval seminal trailblazing work from the philologist Zipf, criticism have stated they are just technical systems made up of data, statistics, and predictions.

According to the linguist Emily Bender, “a language model is a system for haphazardly stitching together sequences of linguistic forms it has observed in its vast training data, according to probabilistic information about how they combine, but without any reference to meaning: a stochastic parrot. Quite the opposite we, human beings, are intentional subjects who can make things into objects of thought by inventing and endowing meaning.“

Machine learning engineers in companies like OpenAI, Google, Facebook, or Microsoft have experimentally established a concept of language at the center of which does not need to be the human. According to this new concept, language is a system organized by an internal combinatorial logic that is independent from whomever speaks (human or machine). They have undermined one of the most deeply rooted axioms in Western philosophy: humans have what animals and machines do not have, language and logos.

Some data: monthly, on average, humans publish about seventy million posts on the content management platform WordPress. Humans produce about fifty-six billion words a month, or 1.8 billion words a day on this content management platform. GPT-3 -before its scintillating launch- was producing around 4.5 billion words a day, more than twice what humans on WordPress were doing collectively. And that is just GPT-3; there are other LLMs. We are exposed to a flood of non-human words. What will it mean to be surrounded by a multitude of non-human forms of intelligence? How can we relate to these astonishingly powerful content-generator LLMs? Do machines require semantics or even a will to communicate with us?

These are philosophical questions that cannot be just solved with an engineering approach. The scope is much wider and the stakes are extremely high. LLMs can, as well as master and learn our human languages, make us reflect and question ourselves about the nature of language, knowledge, and intelligence. Large language models illustrate, for the first time in the history of AI, that language understanding can be decoupled from all the sensorial and emotional features we, human beings, share with each other. Gradually, it seems we are entering eventually a new epoch in AI.

Humanos y máquinas, un futuro con inteligencia artificial. AI Event at Fundación Telefónica

Last Thursday, April 7th, I was invited as CEO of hAItta to participate and give a speech about Natural Language Processing: the pending challenge in AI, at the event “Humanos y máquinas, un futuro con inteligencia artificial” in Fundación Telefónica.

It was a great pleasure for me to share the floor with Mr. Ramón López de Mántaras. director y profesor de investigación del Instituto de Investigación en Inteligencia Artificial del CSIC, Mr. Álvaro Otero, COO de BigML, Mr. Juan García Braschi, country manager España y CFO en Cabify, and Mr. Javier Placer Mendoza, CEO de Telefónica Open Future.

Bringing AI closer to society.

Full video of the event “Humanos y máquinas, un futuro con inteligencia artificial. Mar de datos 2016. Fundación Telefónica”

Summary video of the event and later interview

Winograd Schema Challenge: A Step beyond the Turing Test

The well-known Turing test was first proposed by Alan Turing (1950) as a practical way to defuse what seemed to him to be a pointless argument about whether or not machines could think. He put forward that, instead of formulating such a vague question, we should ask whether a machine would be capable of producing behavior that we would say required thought in people. The sort of behavior he had in mind was participating in a natural conversation in English over a teletype in what he called the Imitation Game. The idea, roughly, was the following: if an interrogator was unable to tell after a long, free flowing and unrestricted conversation with a machine whether s/he was dealing with a person or a machine, then we should be prepared to say that the machine was thinking. The Turing test does have some troubling aspects though.

Leer más »

Once upon a Time in 1954…

Die Grenzen meiner Sprache bedeuten die Grenzen meiner Welt

(The limits of my language are the limits of my world)

Ludwig Wittgenstein, Tractatus Logico-Philosophicus

… In a cold day of January it took place in Washington DC the Georgetown-IBM experiment, the first and most influential demonstration of automatic translation performed throughout the history. Developed jointly by the University of Georgetown and IBM, the experiment implied the automatic translation of more than 60 sentences from Russian into English. The sentences were chosen precisely; there was no syntactic analysis, which could manage to identify the sentence structure. The approach was mainly lexicographic, based on dictionaries in which a certain word had a link to some particular rules.

That episode was a success. Story has it that the level of euphoria amongst the researchers was such that it was stated that within three or five years the problem of the automatic translation would be solved… That was more than 60 years ago and the language problem –the comprehension and generation of messages by the machine- is still pending. Probably this is the last frontier which separates the human intelligence from the artificial intelligence.

Leer más »