Othello, also known as Reversi, is a popular 2-players board game. Olivaw is an intelligent agent playing Othello. Compared to the most famous ones (such as Saio), it exploits limited resources by autonomously learning how to improve its gameplay by playing against itself. In previous occasions, Othello players reported the impression of a sort of human-likeness in how Olivaw plays. We designed and ran an experimental study to better investigate these impressions in a controlled setting. Participants were asked to watch the moves of pre-recorded Othello games played by a human expert player against either another agent (i.e., Olivaw, Saio) or another human. The identity of the opponent, the outcome of the game (i.e., whether the human expert or the opponent player won), and the color of the players (i.e., black or white, black always playing first) were manipulated. We then asked participants to evaluate the human-likeness of the opponent player. Results confirm that the outcome of the match affects the perception of human-likeness of the players.

While different neural models often exhibit latent spaces that are alike when exposed to semantically related data, this intrinsic similarity is not always immediately discernible. Towards a better understanding of this phenomenon, our work shows how representations learned from these neural modules can be translated between different pre-trained networks via simpler transformations than previously thought. An advantage of this approach is the ability to estimate these transformations using standard, well-understood algebraic procedures that have closed-form solutions. Our method directly estimates a transformation between two given latent spaces, thereby enabling effective stitching of encoders and decoders without additional training. We extensively validate the adaptability of this translation procedure in different experimental settings: across various trainings, domains, architectures (e.g., ResNet, CNN, ViT), and in multiple downstream tasks (classification, reconstruction). Notably, we show how it is possible to zero-shot stitch text encoders and vision decoders, or vice-versa, yielding surprisingly good classification performance in this multimodal setting.

CLIP proved that aligning visual and language spaces is key to solving many vision tasks without explicit training, but required to train image and text encoders from scratch on a huge dataset. LiT improved this by only training the text encoder and using a pre-trained vision network. In this paper, we show that a common space can be created without any training at all, using single-domain encoders (trained with or without supervision) and a much smaller amount of image-text pairs. Furthermore, our model has unique properties. Most notably, deploying a new version with updated training samples can be done in a matter of seconds. Additionally, the representations in the common space are easily interpretable as every dimension corresponds to the similarity of the input to a unique entry in the multimodal dataset. Experiments on standard zero-shot visual benchmarks demonstrate the typical transfer ability of image-text models. Overall, our method represents a simple yet surprisingly strong baseline for foundation multi-modal models, raising important questions on their data efficiency and on the role of retrieval in machine learning.

In this article, we introduceOlivaw, an artificial intelligenceOthelloplayer adopting the design principles of the famous AlphaGo programs. The main motivation behindOlivawis to attain exceptional competence in a nontrivial board game at a tiny fraction of the cost of its illustrious predecessors. In this article, we show how the AlphaGo Zero’s paradigm can be successfully applied to the popular game ofOthellousing only commodity hardware and free cloud services. While being simpler thanChessorGo,Othellomaintains a considerable search space and difficulty in evaluating board positions. To achieve this result,Olivawimplements some improvements inspired by recent works to accelerate the standard AlphaGo Zero learning process. The main modification implies doubling the positions collected per game during the training phase, by including also positions not played but largely explored by the agent. We tested the strength ofOlivawin three different ways: by pitting it against Edax, considered by the strongest open-sourceOthelloengine, by playing anonymous games on the web platform OthelloQuest, and, finally, in two in-person matches against top-notch human players: a national champion and a former world champion.

Olivaw is an AI Othello playing agent which autonomously learns how to improve its gameplay by playing against itself. Some top-notch players (including former World Champions) reported that they had the impression that Olivaw's gameplay was human-like. To better investigate the processes related to these impressions, we conducted a pilot study using the Othello Game Evaluation App, a computer application we developed to evaluate pre-recorded Othello games in a controlled setting while assuring an adequate user experience. An exploratory analysis of the results shows that the participants mostly evaluated Olivaw as a human. When asked for a motivation for their choice, some of them reported that they evaluate poor game moves (and, consequently, losing the game) as an indication of the human-likeness of the player.