Artificial Intelligence (AI) and video games are two of the greatest technological achievements of our time. You can’t have one without the other, and for better or worse, they’ve been linked to each other since the very beginning. Wide-reaching technology that can benefit many industries, AI gaming has already made its mark on the world in many ways.
The twins of technology
As a catalyst for rapid change in a wide range of industries, AI has improved profits and streamlined workloads. Video games, on the other hand, have swiftly migrated from arcade machines to every digital device you can think of. If it has a screen, you can play a game on it. For a large majority of us, AI and video games have demonstrably become an integral part of daily life.
As if playing a game of leapfrog, AI and video games have helped each other advance at an exponential rate. From competent 1950’s chess AI to the machine learning marvels of today, we’ve come a long way in a relatively short amount of time.
We’re still discovering ways that AI can help us develop smarter games, and game technologies are conversely driving innovation in the field of AI. Below you’ll find some examples of the AI’s responsible for a new era of video games and how these intelligent machines have been taking the world of video games by storm.
Using artificial neural network technology, Google’s DeepMind AI programs can learn how to play games the same way humans do. Short-term memory systems allow these AI’s to store (as well as recall) any information they’ve been given. Describing these AI programs as general-purpose learning algorithms, DeepMind has showcased a number of successful examples in recent years.
In 2016, DeepMind’s AlphaGo AI was the subject of a highly rated documentary when it managed to beat a world champion Go player in a five-round match (Go is a complex variant of chess popular across Asia). To give you an idea of the rapid pace at which AI can develop, AlphaGo took months of learning to learn how to play Go. AlphaGo’s successor, AlphaZero, took a mere three days to learn how to beat AlphaGo and went on to beat its predecessor 100 times in a row.
AlphaStar is another DeepMind AI designed to play the popular real-time strategy game Starcraft 2. In a fraction of the time it would take a human player to do the same, AlphaStar managed to reach the highest levels of Starcraft 2’s multiplayer rankings. From July of 2019, AlphaStar played 1v1 matches against human players, and by October of the same year, AlphaStar was one of the 200 highest-ranked Starcraft 2 ‘players’ in the whole of Europe.
DeepMind’s latest AI, named Alpha57, has already demonstrated its ability to beat human players at over 50 different Atari classic games. Considering the aforementioned examples and AI’s astonishing rate of improvement, DeepMind’s next breakthrough might be just around the corner.
Generative Pre-trained Transformer, or GPT for short, is an AI program developed by the well-known research lab OpenAI. By using what’s known as a generative language model, GPT makes informed decisions based on how much experience you give it. The more you teach it, the more effective GPT’s decision-making becomes.
AI Dungeon is a great example of what GPT can do. Invented in 2019 by AI developer Nick Walton, AI dungeon is a text adventure game akin to tabletop role-playing games like Dungeons and Dragons. Using a similar structure to those early tabletop games, player characters are all human controlled in AI Dungeon, but the twist is that the dungeon master is a GPT program.
Every world in AI Dungeon, from spells and lore to enemies and locations, all of it is created entirely by GPT’s AI. Since AI Dungeon’s 2019 release, GPT has undergone several revisions. Licensed exclusively to Microsoft at the end of 2020, GPT-3 is the last-known iteration of OpenAI’s flagship technology.
Open AI Five, another OpenAI project, has been using the 5v5 battle arena video game Dawn of the Ancients (Dota) 2 as a proverbial stomping ground. In Dota 2, teams have to fight each other as well as basic enemies while also capturing objectives around the map.
With annual televised tournament sporting prizes worth millions of dollars, Dota 2 is one of the most well-known esport games around. In 2018 Open AI Five began trial matches in Dota 2 against human opponents. By mid-2019, it had defeated that year’s world champions. Barely a month later, it would go on to play over 4,000 matches, winning an unprecedented 9 games out of every 10 played.
AI is formidable in competitive multiplayer games; that much is clear. It’s also clear that video game mechanics can be used as steppingstones to develop more advanced AI. What’s not clear, however, is just how far this technology can go.
We’re in uncharted waters here, a thought that may terrify some people, but for others, the digital frontier of AI is a call to adventure, a chance to make discoveries that could benefit the whole of mankind.